🔭Data Science: Transformations (Just the Quotes)

"Logging size transforms the original skewed distribution into a more symmetrical one by pulling in the long right tail of the distribution toward the mean. The short left tail is, in addition, stretched. The shift toward symmetrical distribution produced by the log transform is not, of course, merely for convenience. Symmetrical distributions, especially those that resemble the normal distribution, fulfill statistical assumptions that form the basis of statistical significance testing in the regression model." (Edward R Tufte, "Data Analysis for Politics and Policy", 1974)

"Logging skewed variables also helps to reveal the patterns in the data. […] the rescaling of the variables by taking logarithms reduces the nonlinearity in the relationship and removes much of the clutter resulting from the skewed distributions on both variables; in short, the transformation helps clarify the relationship between the two variables. It also […] leads to a theoretically meaningful regression coefficient." (Edward R Tufte, "Data Analysis for Politics and Policy", 1974)

"The logarithmic transformation serves several purposes: (1) The resulting regression coefficients sometimes have a more useful theoretical interpretation compared to a regression based on unlogged variables. (2) Badly skewed distributions - in which many of the observations are clustered together combined with a few outlying values on the scale of measurement - are transformed by taking the logarithm of the measurements so that the clustered values are spread out and the large values pulled in more toward the middle of the distribution. (3) Some of the assumptions underlying the regression model and the associated significance tests are better met when the logarithm of the measured variables is taken." (Edward R Tufte, "Data Analysis for Politics and Policy", 1974)

"The logarithm is one of many transformations that we can apply to univariate measurements. The square root is another. Transformation is a critical tool for visualization or for any other mode of data analysis because it can substantially simplify the structure of a set of data. For example, transformation can remove skewness toward large values, and it can remove monotone increasing spread. And often, it is the logarithm that achieves this removal." (William S Cleveland, "Visualizing Data", 1993)

"Compound errors can begin with any of the standard sorts of bad statistics - a guess, a poor sample, an inadvertent transformation, perhaps confusion over the meaning of a complex statistic. People inevitably want to put statistics to use, to explore a number's implications. [...] The strengths and weaknesses of those original numbers should affect our confidence in the second-generation statistics." (Joel Best, "Damned Lies and Statistics: Untangling Numbers from the Media, Politicians, and Activists", 2001)

"All forms of complex causation, and especially nonlinear transformations, admittedly stack the deck against prediction. Linear describes an outcome produced by one or more variables where the effect is additive. Any other interaction is nonlinear. This would include outcomes that involve step functions or phase transitions. The hard sciences routinely describe nonlinear phenomena. Making predictions about them becomes increasingly problematic when multiple variables are involved that have complex interactions. Some simple nonlinear systems can quickly become unpredictable when small variations in their inputs are introduced." (Richard N Lebow, "Forbidden Fruit: Counterfactuals and International Relations", 2010)

"Either a logarithmic or a square-root transformation of the data would produce a new series more amenable to fit a simple trigonometric model. It is often the case that periodic time series have rounded minima and sharp-peaked maxima. In these cases, the square root or logarithmic transformation seems to work well most of the time." (DeWayne R Derryberry, "Basic data analysis for time series with R", 2014)

"Transformations of data alter statistics. For example, the mean of a data set can be found, but it is not easy to relate the mean of a data set to the mean of the logarithm of that data set. The median is far friendlier to transformations. If the median of a data set is found, then the logarithm of the data set is analyzed; the median of the log transformed data will be the log of the original median." (DeWayne R Derryberry, "Basic data analysis for time series with R", 2014) 

"Transforming data to measurements of a different kind can clarify and simplify hypotheses that have already been generated and can reveal patterns that would otherwise be hidden." (Forrest W Young et al, "Visual Statistics: Seeing data with dynamic interactive graphics", 2016)

"Feature generation (or engineering, as it is often called) is where the bulk of the time is spent in the machine learning process. As social science researchers or practitioners, you have spent a lot of time constructing features, using transformations, dummy variables, and interaction terms. All of that is still required and critical in the machine learning framework. One difference you will need to get comfortable with is that instead of carefully selecting a few predictors, machine learning systems tend to encourage the creation of lots of features and then empirically use holdout data to perform regularization and model selection. It is common to have models that are trained on thousands of features." (Rayid Ghani & Malte Schierholz, "Machine Learning", 2017)

"Data analysis and data mining are concerned with unsupervised pattern finding and structure determination in data sets. The data sets themselves are explicitly linked as a form of representation to an observational or otherwise empirical domain of interest. 'Structure' has long been understood as symmetry which can take many forms with respect to any transformation, including point, translational, rotational, and many others. Symmetries directly point to invariants, which pinpoint intrinsic properties of the data and of the background empirical domain of interest. As our data models change, so too do our perspectives on analysing data." (Fionn Murtagh, "Data Science Foundations: Geometry and Topology of Complex Hierarchic Systems and Big Data Analytics", 2018)

"Many statistical procedures perform more effectively on data that are normally distributed, or at least are symmetric and not excessively kurtotic (fat-tailed), and where the mean and variance are approximately constant. Observed time series frequently require some form of transformation before they exhibit these distributional properties, for in their 'raw' form they are often asymmetric." (Terence C Mills, "Applied Time Series Analysis: A practical guide to modeling and forecasting", 2019)

🔭Data Science: Optimization (Just the Quotes)

"[...] any hope that we are smart enough to find even transiently optimum solutions to our data analysis problems is doomed to failure, and, indeed, if taken seriously, will mislead us in the allocation of effort, thus wasting both intellectual and computational effort." (John W Tukey, "Choosing Techniques for the Analysis of Data", 1981)

"In constructing a model, we always attempt to maximize its usefulness. This aim is closely connected with the relationship among three key characteristics of every systems model: complexity, credibility, and uncertainty. This relationship is not as yet fully understood. We only know that uncertainty (predictive, prescriptive, etc.) has a pivotal role in any efforts to maximize the usefulness of systems models. Although usually (but not always) undesirable when considered alone, uncertainty becomes very valuable when considered in connection to the other characteristics of systems models: in general, allowing more uncertainty tends to reduce complexity and increase credibility of the resulting model. Our challenge in systems modelling is to develop methods by which an optimal level of allowable uncertainty can be estimated for each modelling problem." (George J Klir & Bo Yuan, "Fuzzy Sets and Fuzzy Logic: Theory and Applications", 1995)

"[...] an algorithm’s average performance is determined by how 'aligned' it is with the underlying probability distribution over optimization problems on which it is run." (David H Wolpert & William G Macready, "No free lunch theorems for optimization", IEEE Transactions on Evolutionary Computation 1 (1), 1997)

"[...] despite the NFL theorems, algorithms can have a priori distinctions that hold even if nothing is specified concerning the optimization problems. In particular, we show that there can be 'head-to-head' minimax distinctions between a pair of algorithms, i.e., that when considering one function at a time ,a pair of algorithms may be distinguishable, even if they are not when one looks over all functions." (David H Wolpert & William G Macready, "No free lunch theorems for optimization", IEEE Transactions on Evolutionary Computation 1 (1), 1997)

"[...] if you have a general optimization involving uncertainty and very little prior knowledge, the situation is rather hopeless. Due to the NFL theorem, you cannot do any better than a blind search. Each blind search evaluation will be very expensive, with no hope of future improvement, theoretical or otherwise. And the number of performance searches required to get anywhere is simply too large. Neither time nor theoretical or technological progress are on your side. No grand optimization algorithm to end all algorithms is possible." (Yu-Chi Ho, "The no free lunch theorem and the human-machine interface", IEEE Control Systems Magazine, 1999)

"The No Free Lunch (NFL) theorem […] tells us that without any structural assumptions on an optimization problem, no algorithm can perform better on average than blind search." (Yu-Chi Ho, "The no free lunch theorem and the human-machine interface", IEEE Control Systems Magazine, 1999)

"A model is an imitation of reality and a mathematical model is a particular form of representation. We should never forget this and get so distracted by the model that we forget the real application which is driving the modelling. In the process of model building we are translating our real world problem into an equivalent mathematical problem which we solve and then attempt to interpret. We do this to gain insight into the original real world situation or to use the model for control, optimization or possibly safety studies." (Ian T Cameron & Katalin Hangos, "Process Modelling and Model Analysis", 2001)

"Because No Free Lunch theorems dictate that no optimization algorithm can be considered more efficient than any other when considering all possible functions, the desired function class plays a prominent role in the model. In particular, this provides a tractable way to answer the traditionally difficult question of what algorithm is best matched to a particular class of functions. Among the benefits of the model are the ability to specify the function class in a straightforward manner, a natural way to specify noisy or dynamic functions, and a new source of insight into No Free Lunch theorems for optimization." (Christopher K Monson, "No Free Lunch, Bayesian Inference, and Utility: A Decision-Theoretic Approach to Optimization", [thesis] 2006)

"There may be no significant difference between the point of view of inferring the true structure and that of making a prediction if an infinitely large quantity of data is available or if the data are noiseless. However, in modeling based on a finite quantity of real data, there is a significant gap between these two points of view, because an optimal model for prediction purposes may be different from one obtained by estimating the 'true model'." (Genshiro Kitagawa & Sadanori Konis, "Information Criteria and Statistical Modeling", 2007)

"A priori, it is clear that no method will always be the best [...]. However, it is reasonable to argue that each method will have a set of functions, a type of data, and a range of sample sizes for which it is optimal – a sort of catchment region for each procedure. Ideally, one could partition a space of regression problems into catchment regions, depending on which methods were under consideration, and determine which catchment region seemed most appropriate for each method. This ideal solution would amount to a selection principle for nonparametric methods. Unfortunately, it is unclear how to do this, not least because the catchment regions are unknown." (Bertrand Clarke et al, "Principles and Theory for Data Mining and Machine Learning", 2009)

"When generating trees, it is usually optimal to grow a larger tree than is justifiable and then prune it back. The main reason this works well is that stop splitting rules do not look far enough forward. That is, stop splitting rules tend to underfit, meaning that even if a rule stops at a split for which the next candidate splits give little improvement, it may be that splitting them one layer further will give a large improvement in accuracy." (Bertrand Clarke et al, "Principles and Theory for Data Mining and Machine Learning", 2009)

"The problem of comparing classifiers is not at all an easy task. There is no single classifier that works best on all given problems, phenomenon related to the 'No-free-lunch' metaphor, i.e., each classifier (’restaurant’) provides a specific technique associated with the corresponding costs (’menu’ and ’price’ for it). It is hence up to us, using the information and knowledge at hand, to find the optimal trade-off." (Florin Gorunescu, "Data Mining Concepts, Models and Techniques", 2011)

"In an emergency, a data product that just produces more data is of little use. Data scientists now have the predictive tools to build products that increase the common good, but they need to be aware that building the models is not enough if they do not also produce optimized, implementable outcomes." (Jeremy Howard et al, "Designing Great Data Products", 2012)

"Briefly speaking, to solve a Machine Learning problem means you optimize a model to fit all the data from your training set, and then you use the model to predict the results you want. Therefore, evaluating a model need to see how well it can be used to predict the data out of the training set. Usually there are three types of the models: underfitting, fair and overfitting model [...]. If we want to predict a value, both (a) and (c) in this figure cannot work well. The underfitting model does not capture the structure of the problem at all, and we say it has high bias. The overfitting model tries to fit every sample in the training set and it did it, but we say it is of high variance. In other words, it fails to generalize new data." (Shudong Hao, "A Beginner’s Tutorial for Machine Learning Beginners", 2014)

"Deep learning is an area of machine learning that emerged from the intersection of neural networks, artificial intelligence, graphical modeling, optimization, pattern recognition and signal processing." (N D Lewis, "Deep Learning Made Easy with R: A Gentle Introduction for Data Science", 2016)

"Optimization is more than finding the best simulation results. It is itself a complex and evolving field that, subject to certain information constraints, allows data scientists, statisticians, engineers, and traders alike to perform reality checks on modeling results." (Chris Conlan, "Automated Trading with R: Quantitative Research and Platform Development", 2016)

"Data scientists should have some domain expertise. Most data science projects begin with a real-world, domain-specific problem and the need to design a data-driven solution to this problem. As a result, it is important for a data scientist to have enough domain expertise that they understand the problem, why it is important, and how a data science solution to the problem might fit into an organization’s processes. This domain expertise guides the data scientist as she works toward identifying an optimized solution." (John D Kelleher & Brendan Tierney, "Data Science", 2018)

"Optimization is the process of finding the maximum or minimum of a given function (also known as a fitness function), by calculating the best values for its variables (also known as a 'solution'). Despite the simplicity of this definition, it is not an easy process; often involves restrictions, as well as complex relationships among the various variables. Even though some functions can be optimized using some mathematical process, most functions we encounter in data science are not as simple, requiring a more advanced technique." (Yunus E Bulut & Zacharias Voulgaris, "AI for Data Science: Artificial Intelligence Frameworks and Functionality for Deep Learning, Optimization, and Beyond", 2018)

"Optimization systems (or optimizers, as they are often referred to) aim to optimize in a systematic way, oftentimes using a heuristics-based approach. Such an approach enables the AI system to use a macro level concept as part of its low-level calculations, accelerating the whole process and making it more light-weight. After all, most of these systems are designed with scalability in mind, so the heuristic approach is most practical." (Yunus E Bulut & Zacharias Voulgaris, "AI for Data Science: Artificial Intelligence Frameworks and Functionality for Deep Learning, Optimization, and Beyond", 2018)

"The no free lunch theorems set limits on the range of optimality of any method. That is, each methodology has a ‘catchment area’ where it is optimal or nearly so. Often, intuitively, if the optimality is particularly strong then the effectiveness of the methodology falls off more quickly outside its catchment area than if its optimality were not so strong. Boosting is a case in point: it seems so well suited to binary classification that efforts to date to extend it to give effective classification (or regression) more generally have not been very successful. Overall, it remains to characterize the catchment areas where each class of predictors performs optimally, performs generally well, or breaks down." (Bertrand S Clarke & Jennifer L. Clarke, "Predictive Statistics: Analysis and Inference beyond Models", 2018)

"Cross-validation is a useful tool for finding optimal predictive models, and it also works well in visualization. The concept is simple: split the data at random into a 'training' and a 'test' set, fit the model to the training data, then see how well it predicts the test data. As the model gets more complex, it will always fit the training data better and better. It will also start off getting better results on the test data, but there comes a point where the test data predictions start going wrong." (Robert Grant, "Data Visualization: Charts, Maps and Interactive Graphics", 2019)

"Random forests are essentially an ensemble of trees. They use many short trees, fitted to multiple samples of the data, and the predictions are averaged for each observation. This helps to get around a problem that trees, and many other machine learning techniques, are not guaranteed to find optimal models, in the way that linear regression is. They do a very challenging job of fitting non-linear predictions over many variables, even sometimes when there are more variables than there are observations. To do that, they have to employ 'greedy algorithms', which find a reasonably good model but not necessarily the very best model possible." (Robert Grant, "Data Visualization: Charts, Maps and Interactive Graphics", 2019)

🔭Data Science: There's No Free Lunch (Just the Quotes)

"There is nothing in any object, consider'd in itself, which can afford us a reason for drawing a conclusion beyond it; […] even after the observation of the frequent or constant conjunction of objects, we have no reason to draw any inference concerning any object beyond those of which we have had experience." (David Hume, "A Treatise of Human Nature", 1739) [considered as the first attempt to underline the limits of inductive inference] 

"Consider any of the heuristics that people have come up with for supervised learning: avoid overfitting, prefer simpler to more complex models, boost your algorithm, bag it, etc. The no free lunch theorems say that all such heuristics fail as often (appropriately weighted) as they succeed. This is true despite formal arguments some have offered trying to prove the validity of some of these heuristics." (David H Wolpert, "The lack of a priori distinctions between learning algorithms", Neural Computation Vol. 8(7), 1996)

"[...] an algorithm’s average performance is determined by how 'aligned' it is with the underlying probability distribution over optimization problems on which it is run." (David H Wolpert & William G Macready, "No free lunch theorems for optimization", IEEE Transactions on Evolutionary Computation 1 (1), 1997)

"[...] despite the NFL theorems, algorithms can have a priori distinctions that hold even if nothing is specified concerning the optimization problems. In particular, we show that there can be 'head-to-head' minimax distinctions between a pair of algorithms, i.e., that when considering one function at a time ,a pair of algorithms may be distinguishable, even if they are not when one looks over all functions." (David H Wolpert & William G Macready, "No free lunch theorems for optimization", IEEE Transactions on Evolutionary Computation 1 (1), 1997)

"The NFL theorems do not directly address minimax properties of search. For example, say we are considering two deterministic algorithms a1 and a2. It may very well be that there exist cost functions such that a1’s histogram is much better (according to some appropriate performance measure) than a2’s, but no cost functions for which the reverse is true. For the NFL theorem to be obeyed in such a scenario, it would have to be true that there are many more f for which a2’s histogram is better than a1’s than vice-versa, but it is only slightly better for all those f. For such a scenario, in a certain sense a1 has better 'head-to-head' minimax behavior than a2; there are f for which a1 beats a2 badly, but none for which a1 does substantially worse than a2." (David H Wolpert & William G Macready, "No free lunch theorems for optimization", IEEE Transactions on Evolutionary Computation 1 (1), 1997)

"[...] the NFL theorems mean that if an algorithm does particularly well on average for one class of problems then it must do worse on average over the remaining problems. In particular, if an algorithm performs better than random search on some class of problems then in must perform worse than random search on the remaining problems. Thus comparisons reporting the performance of a particular algorithm with a particular parameter setting on a few sample problems are of limited utility. While such results do indicate behavior on the narrow range of problems considered, one should be very wary of trying to generalize those results to other problems." (David H Wolpert & William G Macready, "No free lunch theorems for optimization", IEEE Transactions on Evolutionary Computation 1 (1), 1997)

"[...] if you have a general optimization involving uncertainty and very little prior knowledge, the situation is rather hopeless. Due to the NFL theorem, you cannot do any better than a blind search. Each blind search evaluation will be very expensive, with no hope of future improvement, theoretical or otherwise. And the number of performance searches required to get anywhere is simply too large. Neither time nor theoretical or technological progress are on your side. No grand optimization algorithm to end all algorithms is possible." (Yu-Chi Ho, "The no free lunch theorem and the human-machine interface", IEEE Control Systems Magazine, 1999)

"The No Free Lunch (NFL) theorem […] tells us that without any structural assumptions on an optimization problem, no algorithm can perform better on average than blind search." (Yu-Chi Ho, "The no free lunch theorem and the human-machine interface", IEEE Control Systems Magazine, 1999)

"[...] a general-purpose universal optimization strategy is theoretically impossible, and the only way one strategy can outperform another is if it is specialized to the specific problem under consideration." (Yu-Chi Ho & David L Pepyne, "Simple explanation of the no-free-lunch theorem and its implications", Journal of Optimization Theory and Applications 115, 2002)

"Because No Free Lunch is a well-known concept, researchers are increasingly interested in finding function classes that are well matched to their algorithms. The determination of an algorithm–class pairing is generally approximated by testing the algorithm on one of a number of benchmark functions, then generalizing the results by distilling various defining characteristics of those benchmarks." (Christopher K Monson, "No Free Lunch, Bayesian Inference, and Utility: A Decision-Theoretic Approach to Optimization", [thesis] 2006)

"Because No Free Lunch theorems dictate that no optimization algorithm can be considered more efficient than any other when considering all possible functions, the desired function class plays a prominent role in the model. In particular, this provides a tractable way to answer the traditionally difficult question of what algorithm is best matched to a particular class of functions. Among the benefits of the model are the ability to specify the function class in a straightforward manner, a natural way to specify noisy or dynamic functions, and a new source of insight into No Free Lunch theorems for optimization." (Christopher K Monson, "No Free Lunch, Bayesian Inference, and Utility: A Decision-Theoretic Approach to Optimization", [thesis] 2006)

"No Free Lunch dictates that any algorithm may be deceived, a difficulty to which the inference algorithm is not immune." (Christopher K Monson, "No Free Lunch, Bayesian Inference, and Utility: A Decision-Theoretic Approach to Optimization", [thesis] 2006)

"That the complexity of the problem is inherently tied to the flexibility of the representation of the prior serves to clarify part of No Free Lunch, especially the proof that problem difficulties cannot be ranked in the absence of a specific algorithm: it is not, in fact, the choice of algorithm that makes ranking possible among problems, but the choice of representation." (Christopher K Monson, "No Free Lunch, Bayesian Inference,and Utility: A Decision Theoretic Approach to Optimization", [thesis] 2006)

"The No Free Lunch work is a framework that addresses the core aspects of search, focusing on the connection between fitness functions and effective search algorithms. The central importance of this connection is demonstrated by the No Free Lunch theorem, which states that, averaged over all problems, all search algorithms perform equally. This result implies that if we are comparing a genetic algorithm to some other algorithm (e.g., simulated annealing, or even random search) and the genetic algorithm to some other algorithm (e.g., simulated annealing, or even random search) performs better for some class of problems, then the other algorithm necessarily performs better on problems outside the class. Thus it is essential to incorporate knowledge of the problem into the search algorithm." (S N Sivanandam & S N Deepa, "Introduction to Genetic Algorithms", 2008)

"A priori, it is clear that no method will always be the best [...]. However, it is reasonable to argue that each method will have a set of functions, a type of data, and a range of sample sizes for which it is optimal – a sort of catchment region for each procedure. Ideally, one could partition a space of regression problems into catchment regions, depending on which methods were under consideration, and determine which catchment region seemed most appropriate for each method. This ideal solution would amount to a selection principle for nonparametric methods. Unfortunately, it is unclear how to do this, not least because the catchment regions are unknown." (Bertrand Clarke et al, "Principles and Theory for Data Mining and Machine Learning", 2009)

"Methods perform well if the conditions of their derivation are met, with no method capable of covering all possible conditions. More strongly put, each good method has a domain on which it may be best, and different methods have different domains so the task is to characterize those domains and then figure out which domain a given problem’s solution is likely to be in. This of course, is extraordinarily difficult in its own right." (Bertrand Clarke et al, "Principles and Theory for Data Mining and Machine Learning", 2009)

"The well-known 'No Free Lunch' theorem indicates that there does not exist a pattern classification method that is inherently superior to any other, or even to random guessing without using additional information. It is the type of problem, prior information, and the amount of training samples that determine the form of classifier to apply. In fact, corresponding to different real-world problems, different classes may have different underlying data structures. A classifier should adjust the discriminant boundaries to fit the structures which are vital for classification, especially for the generalization capacity of the classifier." (Hui Xue et al, "SVM: Support Vector Machines", 2009)

"Another important fact, having impact on EA [Evolutionary Algorithm] use is so called No Free Lunch Theorem (NFLT) [...]. Main idea of this theorem is that there is no ideal algorithm which would be able to solve any problem. Simply, if there are for example two algorithms A and B, then for certain subset of possible problems is more suitable algorithms A and for another subset algorithm B. All those subsets can be of course totally disconnected, or/and overlapped." (Ivan Zelinka & Hendrik Richter, "Evolutionary Algorithms for Chaos Researchers", Studies in Computational Intelligence Vol. 267, 2010)

"Each learning algorithm dictates a certain model that comes with a set of assumptions. This inductive bias leads to error if the assumptions do not hold for the data. Learning is an ill-posed problem and with finite data, each algorithm converges to a different solution and fails under different circumstances. The performance of a learner may be fine-tuned to get the highest possible accuracy on a validation set, but this finetuning is a complex task and still there are instances on which even the best learner is not accurate enough. The idea is that there may be another base-learner learner that is accurate on these. By suitably combining multiple base learners then, accuracy can be improved." (Ethem Alpaydin, "Introduction to Machine Learning" 2nd Ed, 2010)

"The problem of comparing classifiers is not at all an easy task. There is no single classifier that works best on all given problems, phenomenon related to the 'No-free-lunch' metaphor, i.e., each classifier (’restaurant’) provides a specific technique associated with the corresponding costs (’menu’ and ’price’ for it). It is hence up to us, using the information and knowledge at hand, to find the optimal trade-off." (Florin Gorunescu, "Data Mining Concepts, Models and Techniques", 2011)

"As a consequence of the no free lunch theorem, we need to develop many different types of models, to cover the wide variety of data that occurs in the real world. And for each model, there may be many different algorithms we can use to train the model, which make different speed-accuracy-complexity tradeoffs." (Kevin P Murphy, "Machine Learning: A Probabilistic Perspective", 2012)

"Much of machine learning is concerned with devising different models, and different algorithms to fit them. We can use methods such as cross validation to empirically choose the best method for our particular problem. However, there is no universally best model - this is sometimes called the no free lunch theorem. The reason for this is that a set of assumptions that works well in one domain may work poorly in another." (Kevin P Murphy, "Machine Learning: A Probabilistic Perspective", 2012)

"The validity of NFL theorems largely depends on the validity of their fundamental assumptions. However, whether these assumptions are valid in practice is another question. Often, these assumptions are too stringent, and thus free lunches are possible." (Xin-She Yang, "Free Lunch or No Free Lunch: That is not Just a Question?", 2012)

"The idea of feature learning is to automate the process of finding a good representation of the input space. As mentioned before, the No-Free-Lunch theorem tells us that we must incorporate some prior knowledge on the data distribution in order to build a good feature representation." (Shai Shalev-Shwartz & Shai Ben-David, "Understanding Machine Learning: From Theory to Algorithms", 2014)

"We emphasize that while there are some common techniques for feature learning one may want to try, the No-Free-Lunch theorem implies that there is no ultimate feature learner. Any feature learning algorithm might fail on some problem. In other words, the success of each feature learner relies (sometimes implicitly) on some form of prior assumption on the data distribution. Furthermore, the relative quality of features highly depends on the learning algorithm we are later going to apply using these features." (Shai Shalev-Shwartz & Shai Ben-David, "Understanding Machine Learning: From Theory to Algorithms", 2014)

"Choosing an appropriate classification algorithm for a particular problem task requires practice: each algorithm has its own quirks and is based on certain assumptions. To restate the 'No Free Lunch' theorem: no single classifier works best across all possible scenarios. In practice, it is always recommended that you compare the performance of at least a handful of different learning algorithms to select the best model for the particular problem; these may differ in the number of features or samples, the amount of noise in a dataset, and whether the classes are linearly separable or not." (Sebastian Raschka, "Python Machine Learning", 2015)

"Learning theory claims that a machine learning algorithm can generalize well from a finite training set of examples. This seems to contradict some basic principles of logic. Inductive reasoning, or inferring general rules from a limited set of examples, is not logically valid. To logically infer a rule describing every member of a set, one must have information about every member of that set." (Ian Goodfellow et al, "Deep Learning", 2015)

"The no free lunch theorem for machine learning states that, averaged over all possible data generating distributions, every classification algorithm has the same error rate when classifying previously unobserved points. In other words, in some sense, no machine learning algorithm is universally any better than any other. The most sophisticated algorithm we can conceive of has the same average performance (over all possible tasks) as merely predicting that every point belongs to the same class. [...] the goal of machine learning research is not to seek a universal learning algorithm or the absolute best learning algorithm. Instead, our goal is to understand what kinds of distributions are relevant to the 'real world' that an AI agent experiences, and what kinds of machine learning algorithms perform well on data drawn from the kinds of data generating distributions we care about." (Ian Goodfellow et al, "Deep Learning", 2015)

"The no free lunch theorem implies that we must design our machine learning algorithms to perform well on a specific task. We do so by building a set of preferences into the learning algorithm. When these preferences are aligned with the learning problems we ask the algorithm to solve, it performs better." (Ian Goodfellow et al, "Deep Learning", 2015)

"The 'No free lunch' theorem demonstrates that it is not possible to find one algorithm behaving better for any problem. On the other hand, we know that we can work with different degrees of knowledge of the problem which we expect to solve, and that it is not the same to work without knowledge of the problem (hypothesis of the 'no free lunch' theorem) than to work with partial knowledge about the problem, knowledge that allows us to design algorithms with specific characteristics which can make them more suitable to solve of the problem." (Salvador García et al, "Data Preprocessing in Data Mining", 2015)

"Roughly stated, the No Free Lunch theorem states that in the lack of prior knowledge (i.e. inductive bias) on average all predictive algorithms that search for the minimum classification error (or extremum over any risk metric) have identical performance according to any measure." (N D Lewis, "Deep Learning Made Easy with R: A Gentle Introduction for Data Science", 2016)

"However, because ML algorithms are biased to look for different types of patterns, and because there is no one learning bias across all situations, there is no one best ML algorithm. In fact, a theorem known as the 'no free lunch theorem' states that there is no one best ML algorithm that on average outperforms all other algorithms across all possible data sets." (John D Kelleher & Brendan Tierney, "Data Science", 2018)

"The no free lunch theorems set limits on the range of optimality of any method. That is, each methodology has a ‘catchment area’ where it is optimal or nearly so. Often, intuitively, if the optimality is particularly strong then the effectiveness of the methodology falls off more quickly outside its catchment area than if its optimality were not so strong. Boosting is a case in point: it seems so well suited to binary classification that efforts to date to extend it to give effective classification (or regression) more generally have not been very successful. Overall, it remains to characterize the catchment areas where each class of predictors performs optimally, performs generally well, or breaks down." (Bertrand S Clarke & Jennifer L. Clarke, "Predictive Statistics: Analysis and Inference beyond Models", 2018)

"The No Free Lunch theorems prove that under a uniform distribution over induction problems (search problems or learning problems), all induction algorithms perform equally.  […] the importance of the theorems arises by using them to analyze scenarios involving non-uniform distributions, and to compare different algorithms, without any assumption about the distribution over problems at all. In particular, the theorems prove that anti-cross-validation (choosing among a set of candidate algorithms based on which has worst out-of-sample behavior) performs as well as cross-validation, unless one makes an assumption - which has never been formalized - about how the distribution over induction problems, on the one hand, is related to the set of algorithms one is choosing among using (anti-)cross validation, on the other. In addition, they establish strong caveats concerning the significance of the many results in the literature which establish the strength of a particular algorithm without assuming a particular distribution." (David H Wolpert, "What is important about the No Free Lunch theorems?", 2020)

"A well-known theorem called the 'no free lunch' theorem proves exactly what we anecdotally witness when designing and building learning systems. The theorem states that any bias-free learning system will perform no better than chance when applied to arbitrary problems. This is a fancy way of stating that designers of systems must give the system a bias deliberately, so it learns what’s intended. As the theorem states, a truly bias- free system is useless." (Erik J Larson, "The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do", 2021)

🔭Data Science: Data Exploration (Just the Quotes)

"Exploratory data analysis can never be the whole story, but nothing else can serve as the foundation stone – as the first step." (John W Tukey, "Exploratory Data Analysis", 1977)

"Modern data graphics can do much more than simply substitute for small statistical tables. At their best, graphics are instruments for reasoning about quantitative information. Often the most effective way to describe, explore, and summarize a set of numbers even a very large set - is to look at pictures of those numbers. Furthermore, of all methods for analyzing and communicating statistical information, well-designed data graphics are usually the simplest and at the same time the most powerful." (Edward R Tufte, "The Visual Display of Quantitative Information", 1983)

"Unless exploratory data analysis uncovers indications, usually quantitative ones, there is likely to nothing for confirmatory data analysis to consider." (John W Tukey, "Exploratory Data Analysis", 1977)

"Many of the applications of visualization in this book give the impression that data analysis consists of an orderly progression of exploratory graphs, fitting, and visualization of fits and residuals. Coherence of discussion and limited space necessitate a presentation that appears to imply this. Real life is usually quite different. There are blind alleys. There are mistaken actions. There are effects missed until the very end when some visualization saves the day. And worse, there is the possibility of the nearly unmentionable: missed effects." (William S Cleveland, "Visualizing Data", 1993)

"The scatterplot is a useful exploratory method for providing a first look at bivariate data to see how they are distributed throughout the plane, for example, to see clusters of points, outliers, and so forth." (William S Cleveland, "Visualizing Data", 1993)

"The science of statistics may be described as exploring, analyzing and summarizing data; designing or choosing appropriate ways of collecting data and extracting information from them; and communicating that information. Statistics also involves constructing and testing models for describing chance phenomena. These models can be used as a basis for making inferences and drawing conclusions and, finally, perhaps for making decisions." (Fergus Daly et al, "Elements of Statistics", 1995)

"Compound errors can begin with any of the standard sorts of bad statistics - a guess, a poor sample, an inadvertent transformation, perhaps confusion over the meaning of a complex statistic. People inevitably want to put statistics to use, to explore a number's implications. [...] The strengths and weaknesses of those original numbers should affect our confidence in the second-generation statistics." (Joel Best, "Damned Lies and Statistics: Untangling Numbers from the Media, Politicians, and Activists", 2001)

"Data mining is more of an art than a science. No one can tell you exactly how to choose columns to include in your data mining models. There are no hard and fast rules you can follow in deciding which columns either help or hinder the final model. For this reason, it is important that you understand how the data behaves before beginning to mine it. The best way to achieve this level of understanding is to see how the data is distributed across columns and how the different columns relate to one another. This is the process of exploring the data." (Seth Paul et al. "Preparing and Mining Data with Microsoft SQL Server 2000 and Analysis", 2002)

"Every statistical analysis is an interpretation of the data, and missingness affects the interpretation. The challenge is that when the reasons for the missingness cannot be determined there is basically no way to make appropriate statistical adjustments. Sensitivity analyses are designed to model and explore a reasonable range of explanations in order to assess the robustness of the results." (Gerald van Belle, "Statistical Rules of Thumb", 2002)

"Since the aim of exploratory data analysis is to learn what seems to be, it should be no surprise that pictures play a vital role in doing it well." (John W. Tukey, "John W Tukey’s Works on Interactive Graphics", The Annals of Statistics Vol. 30 (6), 2002)

"If we attempt to map the world of a story before we explore it, we are likely either to (a) prematurely limit our exploration, so as to reduce the amount of material we need to consider, or (b) explore at length but, recognizing the impossibility of taking note of everything, and having no sound basis for choosing what to include, arbitrarily omit entire realms of information. The opportunities are overwhelming." (Peter Turchi, "Maps of the Imagination: The writer as cartographer", 2004)

"Exploratory Data Analysis is more than just a collection of data-analysis techniques; it provides a philosophy of how to dissect a data set. It stresses the power of visualisation and aspects such as what to look for, how to look for it and how to interpret the information it contains. Most EDA techniques are graphical in nature, because the main aim of EDA is to explore data in an open-minded way. Using graphics, rather than calculations, keeps open possibilities of spotting interesting patterns or anomalies that would not be apparent with a calculation (where assumptions and decisions about the nature of the data tend to be made in advance)." (Alan Graham, "Developing Thinking in Statistics", 2006)

"There are two main reasons for using graphic displays of datasets: either to present or to explore data. Presenting data involves deciding what information you want to convey and drawing a display appropriate for the content and for the intended audience. [...] Exploring data is a much more individual matter, using graphics to find information and to generate ideas. Many displays may be drawn. They can be changed at will or discarded and new versions prepared, so generally no one plot is especially important, and they all have a short life span." (Antony Unwin, "Good Graphics?" [in "Handbook of Data Visualization"], 2008)

"All graphics present data and allow a certain degree of exploration of those same data. Some graphics are almost all presentation, so they allow just a limited amount of exploration; hence we can say they are more infographics than visualization, whereas others are mostly about letting readers play with what is being shown, tilting more to the visualization side of our linear scale. But every infographic and every visualization has a presentation and an exploration component: they present, but they also facilitate the analysis of what they show, to different degrees." (Alberto Cairo, "The Functional Art", 2011)

"But if you don’t present your data to readers so they can see it, read it, explore it, and analyze it, why would they trust you?" (Alberto Cairo, "The Functional Art", 2011)

"Data scientists combine entrepreneurship with patience, the willingness to build data products incrementally, the ability to explore, and the ability to iterate over a solution. They are inherently interdisciplinary. They can tackle all aspects of a problem, from initial data collection and data conditioning to drawing conclusions. They can think outside the box to come up with new ways to view the problem, or to work with very broadly defined problems: 'there’s a lot of data, what can you make from it?'" (Mike Loukides, "What Is Data Science?", 2011)

"Don’t rush to write a headline or an entire story or to design a visualization immediately after you find an interesting pattern, data point, or fact. Stop and think. Look for other sources and for people who can help you escape from tunnel vision and confirmation bias. Explore your information at multiple levels of depth and breadth, looking for extraneous factors that may help explain your findings. Only then can you make a decision about what to say, and how to say it, and about what amount of detail you need to show to be true to the data." (Alberto Cairo, "The Functional Art", 2011)

"The process of visually exploring data can be summarized in a single sentence: find patterns and trends lurking in the data and then observe the deviations from those patterns. Interesting stories may arise from both the norm - also called the smooth - and the exceptions." (Alberto Cairo, "The Functional Art", 2011)

"A viewer’s eye must be guided to 'read' the elements in a logical order. The design of an exploratory graphic needs to allow for the additional component of discovery - guiding the viewer to first understand the overall concept and then engage her to further explore the supporting information." (Felice C Frankel & Angela H DePace, "Visual Strategies", 2012)

"The process of visual analysis can potentially go on endlessly, with seemingly infinite combinations of variables to explore, especially with the rich opportunities bigger data sets give us. However, by deploying a disciplined and sensible balance between deductive and inductive enquiry you should be able to efficiently and effectively navigate towards the source of the most compelling stories." (Andy Kirk, "Data Visualization: A successful design process", 2012)

"Data mining is a craft. As with many crafts, there is a well-defined process that can help to increase the likelihood of a successful result. This process is a crucial conceptual tool for thinking about data science projects. [...] data mining is an exploratory undertaking closer to research and development than it is to engineering." (Foster Provost, "Data Science for Business", 2013)

"Early exploration of a dataset can be overwhelming, because you don’t know where to start. Ask questions about the data and let your curiosities guide you. […] Make multiple charts, compare all your variables, and see if there are interesting bits that are worth a closer look. Look at your data as a whole and then zoom in on categories and individual data points. […] Subcategories, the categories within categories (within categories), are often more revealing than the main categories. As you drill down, there can be higher variability and more interesting things to see." (Nathan Yau, "Data Points: Visualization That Means Something", 2013)

"Good visualization is a winding process that requires statistics and design knowledge. Without the former, the visualization becomes an exercise only in illustration and aesthetics, and without the latter, one of only analyses. On their own, these are fine skills, but they make for incomplete data graphics. Having skills in both provides you with the luxury - which is growing into a necessity - to jump back and forth between data exploration and storytelling." (Nathan Yau, "Data Points: Visualization That Means Something", 2013)

"Visualization can be appreciated purely from an aesthetic point of view, but it’s most interesting when it’s about data that’s worth looking at. That’s why you start with data, explore it, and then show results rather than start with a visual and try to squeeze a dataset into it. It’s like trying to use a hammer to bang in a bunch of screws. […] Aesthetics isn’t just a shiny veneer that you slap on at the last minute. It represents the thought you put into a visualization, which is tightly coupled with clarity and affects interpretation." (Nathan Yau, "Data Points: Visualization That Means Something", 2013)

"Exploratory analysis is what you do to understand the data and figure out what might be noteworthy or interesting to highlight to others." (Cole N Knaflic, "Storytelling with Data: A Data Visualization Guide for Business Professionals", 2015)

"Highlighting one aspect can make other things harder to see one word of warning in using preattentive attributes: when you highlight one point in your story, it can actually make other points harder to see. When you’re doing exploratory analysis, you should mostly avoid the use of preattentive attributes for this reason. When it comes to explanatory analysis, however, you should have a specific story you are communicating to your audience. Leverage preattentive attributes to help make that story visually clear." (Cole N Knaflic, "Storytelling with Data: A Data Visualization Guide for Business Professionals", 2015)

"Exploratory data analysis is the search for patterns and trends in a given data set. Visualization techniques play an important part in this quest. Looking carefully at your data is important for several reasons, including identifying mistakes in collection/processing, finding violations of statistical assumptions, and suggesting interesting hypotheses." (Steven S Skiena, "The Data Science Design Manual", 2017)

"Exploring data generates hypotheses about patterns in our data. The visualizations and tools of dynamic interactive graphics ease and improve the exploration, helping us to 'see what our data seem to say'." (Forrest W Young et al, "Visual Statistics: Seeing data with dynamic interactive graphics", 2016)

"With time series though, there is absolutely no substitute for plotting. The pertinent pattern might end up being a sharp spike followed by a gentle taper down. Or, maybe there are weird plateaus. There could be noisy spikes that have to be filtered out. A good way to look at it is this: means and standard deviations are based on the naïve assumption that data follows pretty bell curves, but there is no corresponding 'default' assumption for time series data (at least, not one that works well with any frequency), so you always have to look at the data to get a sense of what’s normal. [...] Along the lines of figuring out what patterns to expect, when you are exploring time series data, it is immensely useful to be able to zoom in and out." (Field Cady, "The Data Science Handbook", 2017)

"Dashboards are a type of multiform visualization used to summarize and monitor data. These are most useful when proxies have been well validated and the task is well understood. This design pattern brings a number of carefully selected attributes together for fast, and often continuous, monitoring - dashboards are often linked to updating data streams. While many allow interactivity for further investigation, they typically do not depend on it. Dashboards are often used for presenting and monitoring data and are typically designed for at-a-glance analysis rather than deep exploration and analysis." (Danyel Fisher & Miriah Meyer, "Making Data Visual", 2018)

"Models are formal structures represented in mathematics and diagrams that help us to understand the world. Mastery of models improves your ability to reason, explain, design, communicate, act, predict, and explore." (Scott E Page, "The Model Thinker", 2018)

"[…] the data itself can lead to new questions too. In exploratory data analysis (EDA), for example, the data analyst discovers new questions based on the data. The process of looking at the data to address some of these questions generates incidental visualizations - odd patterns, outliers, or surprising correlations that are worth looking into further." (Danyel Fisher & Miriah Meyer, "Making Data Visual", 2018)

"Analysis is a two-step process that has an exploratory and an explanatory phase. In order to create a powerful data story, you must effectively transition from data discovery (when you’re finding insights) to data communication (when you’re explaining them to an audience). If you don’t properly traverse these two phases, you may end up with something that resembles a data story but doesn’t have the same effect. Yes, it may have numbers, charts, and annotations, but because it’s poorly formed, it won’t achieve the same results." (Brent Dykes, "Effective Data Storytelling: How to Drive Change with Data, Narrative and Visuals", 2019)

🔭Data Science: Symmetry (Just the Quotes)

"The framing of hypotheses is, for the enquirer after truth, not the end, but the beginning of his work. Each of his systems is invented, not that he may admire it and follow it into all its consistent consequences, but that he may make it the occasion of a course of active experiment and observation. And if the results of this process contradict his fundamental assumptions, however ingenious, however symmetrical, however elegant his system may be, he rejects it without hesitation. He allows no natural yearning for the offspring of his own mind to draw him aside from the higher duty of loyalty to his sovereign, Truth, to her he not only gives his affections and his wishes, but strenuous labour and scrupulous minuteness of attention." (William Whewell, "Philosophy of the Inductive Sciences" Vol. 2, 1847)

"Rule 2. Any summary of a distribution of numbers in terms of symmetric functions should not give an objective degree of belief in any one of the inferences or predictions to be made therefrom that would cause human action significantly different from what this action would be if the original distributions had been taken as evidence." (Walter A Shewhart, "Economic Control of Quality of Manufactured Product", 1931)

"Logging size transforms the original skewed distribution into a more symmetrical one by pulling in the long right tail of the distribution toward the mean. The short left tail is, in addition, stretched. The shift toward symmetrical distribution produced by the log transform is not, of course, merely for convenience. Symmetrical distributions, especially those that resemble the normal distribution, fulfill statistical assumptions that form the basis of statistical significance testing in the regression model." (Edward R Tufte, "Data Analysis for Politics and Policy", 1974)

"Symmetry is also important because it can simplify our thinking about the distribution of a set of data. If we can establish that the data are (approximately) symmetric, then we no longer need to describe the  shapes of both the right and left halves. (We might even combine the information from the two sides and have effectively twice as much data for viewing the distributional shape.) Finally, symmetry is important because many statistical procedures are designed for, and work best on, symmetric data." (John M Chambers et al, "Graphical Methods for Data Analysis", 1983)

"There are several reasons why symmetry is an important concept in data analysis. First, the most important single summary of a set of data is the location of the center, and when data meaning of 'center' is unambiguous. We can take center to mean any of the following things, since they all coincide exactly for symmetric data, and they are together for nearly symmetric data: (l) the Center Of symmetry. (2) the arithmetic average or center Of gravity, (3) the median or 50%. Furthermore, if data a single point of highest concentration instead of several (that is, they are unimodal), then we can add to the list (4) point of highest concentration. When data are far from symmetric, we may have trouble even agreeing on what we mean by center; in fact, the center may become an inappropriate summary for the data." (John M Chambers et al, "Graphical Methods for Data Analysis", 1983)

"If a distribution were perfectly symmetrical, all symmetry-plot points would be on the diagonal line. Off-line points indicate asymmetry. Points fall above the line when distance above the median is greater than corresponding distance below the median. A consistent run of above-the-line points indicates positive skew; a run of below-the-line points indicates negative skew." (Lawrence C Hamilton, "Regression with Graphics: A second course in applied statistics", 1991)

"Remember that normality and symmetry are not the same thing. All normal distributions are symmetrical, but not all symmetrical distributions are normal. With water use we were able to transform the distribution to be approximately symmetrical and normal, but often symmetry is the most we can hope for. For practical purposes, symmetry (with no severe outliers) may be sufficient. Transformations are not a magic wand, however. Many distributions cannot even be made symmetrical." (Lawrence C Hamilton, "Regression with Graphics: A second course in applied statistics", 1991)

"Chaos demonstrates that deterministic causes can have random effects […] There's a similar surprise regarding symmetry: symmetric causes can have asymmetric effects. […] This paradox, that symmetry can get lost between cause and effect, is called symmetry-breaking. […] From the smallest scales to the largest, many of nature's patterns are a result of broken symmetry; […]" (Ian Stewart & Martin Golubitsky, "Fearful Symmetry: Is God a Geometer?", 1992)

"In everyday language, the words 'pattern' and 'symmetry' are used almost interchangeably, to indicate a property possessed by a regular arrangement of more-or-less identical units […]” (Ian Stewart & Martin Golubitsky, "Fearful Symmetry: Is God a Geometer?", 1992)

"Nature behaves in ways that look mathematical, but nature is not the same as mathematics. Every mathematical model makes simplifying assumptions; its conclusions are only as valid as those assumptions. The assumption of perfect symmetry is excellent as a technique for deducing the conditions under which symmetry-breaking is going to occur, the general form of the result, and the range of possible behaviour. To deduce exactly which effect is selected from this range in a practical situation, we have to know which imperfections are present." (Ian Stewart & Martin Golubitsky, "Fearful Symmetry", 1992)

"Data that are skewed toward large values occur commonly. Any set of positive measurements is a candidate. Nature just works like that. In fact, if data consisting of positive numbers range over several powers of ten, it is almost a guarantee that they will be skewed. Skewness creates many problems. There are visualization problems. A large fraction of the data are squashed into small regions of graphs, and visual assessment of the data degrades. There are characterization problems. Skewed distributions tend to be more complicated than symmetric ones; for example, there is no unique notion of location and the median and mean measure different aspects of the distribution. There are problems in carrying out probabilistic methods. The distribution of skewed data is not well approximated by the normal, so the many probabilistic methods based on an assumption of a normal distribution cannot be applied." (William S Cleveland, "Visualizing Data", 1993)

"A normal distribution is most unlikely, although not impossible, when the observations are dependent upon one another - that is, when the probability of one event is determined by a preceding event. The observations will fail to distribute themselves symmetrically around the mean." (Peter L Bernstein, "Against the Gods: The Remarkable Story of Risk", 1996)

"Symmetry is basically a geometrical concept. Mathematically it can be defined as the invariance of geometrical patterns under certain operations. But when abstracted, the concept applies to all sorts of situations. It is one of the ways by which the human mind recognizes order in nature. In this sense symmetry need not be perfect to be meaningful. Even an approximate symmetry attracts one's attention, and makes one wonder if there is some deep reason behind it." (Eguchi Tohru & ‎K Nishijima ," Broken Symmetry: Selected Papers Of Y Nambu", 1995)

"How deep truths can be defined as invariants – things that do not change no matter what; how invariants are defined by symmetries, which in turn define which properties of nature are conserved, no matter what. These are the selfsame symmetries that appeal to the senses in art and music and natural forms like snowflakes and galaxies. The fundamental truths are based on symmetry, and there’s a deep kind of beauty in that." (K C Cole, "The Universe and the Teacup: The Mathematics of Truth and Beauty", 1997)

"Symmetry and skewness can be judged, but boxplots are not entirely useful for judging shape. It is not possible to use a boxplot to judge whether or not a dataset is bell-shaped, nor is it possible to judge whether or not a dataset may be bimodal." (Jessica M Utts & Robert F Heckard, "Mind on Statistics", 2007)

"The concept of symmetry (invariance) with its rigorous mathematical formulation and generalization has guided us to know the most fundamental of physical laws. Symmetry as a concept has helped mankind not only to define ‘beauty’ but also to express the ‘truth’. Physical laws tries to quantify the truth that appears to be ‘transient’ at the level of phenomena but symmetry promotes that truth to the level of ‘eternity’." (Vladimir G Ivancevic & Tijana T Ivancevic,"Quantum Leap", 2008)

"The concept of symmetry is used widely in physics. If the laws that determine relations between physical magnitudes and a change of these magnitudes in the course of time do not vary at the definite operations (transformations), they say, that these laws have symmetry (or they are invariant) with respect to the given transformations. For example, the law of gravitation is valid for any points of space, that is, this law is in variant with respect to the system of coordinates." (Alexey Stakhov et al, "The Mathematics of Harmony", 2009)

"A pattern is a design or model that helps grasp something. Patterns help connect things that may not appear to be connected. Patterns help cut through complexity and reveal simpler understandable trends. […] Patterns can be temporal, which is something that regularly occurs over time. Patterns can also be spatial, such as things being organized in a certain way. Patterns can be functional, in that doing certain things leads to certain effects. Good patterns are often symmetric. They echo basic structures and patterns that we are already aware of." (Anil K. Maheshwari, "Business Intelligence and Data Mining", 2015)

"One kind of probability - classic probability - is based on the idea of symmetry and equal likelihood […] In the classic case, we know the parameters of the system and thus can calculate the probabilities for the events each system will generate. […] A second kind of probability arises because in daily life we often want to know something about the likelihood of other events occurring […]. In this second case, we need to estimate the parameters of the system because we don’t know what those parameters are. […] A third kind of probability differs from these first two because it’s not obtained from an experiment or a replicable event - rather, it expresses an opinion or degree of belief about how likely a particular event is to occur. This is called subjective probability […]." (Daniel J Levitin, "Weaponized Lies", 2017)

"Variables which follow symmetric, bell-shaped distributions tend to be nice as features in models. They show substantial variation, so they can be used to discriminate between things, but not over such a wide range that outliers are overwhelming." (Steven S Skiena, "The Data Science Design Manual", 2017)

"Data analysis and data mining are concerned with unsupervised pattern finding and structure determination in data sets. The data sets themselves are explicitly linked as a form of representation to an observational or otherwise empirical domain of interest. 'Structure' has long been understood as symmetry which can take many forms with respect to any transformation, including point, translational, rotational, and many others. Symmetries directly point to invariants, which pinpoint intrinsic properties of the data and of the background empirical domain of interest. As our data models change, so too do our perspectives on analysing data." (Fionn Murtagh, "Data Science Foundations: Geometry and Topology of Complex Hierarchic Systems and Big Data Analytics", 2018)

"It is not enough to give a single summary for a distribution - we need to have an idea of the spread, sometimes known as the variability. [...] The range is a natural choice, but is clearly very sensitive to extreme values [...] In contrast the inter-quartile range (IQR) is unaffected by extremes. This is the distance between the 25th and 75th percentiles of the data and so contains the ‘central half’ of the numbers [...] Finally the standard deviation is a widely used measure of spread. It is the most technically complex measure, but is only really appropriate for well-behaved symmetric data since it is also unduly influenced by outlying values." (David Spiegelhalter, "The Art of Statistics: Learning from Data", 2019)

"Many statistical procedures perform more effectively on data that are normally distributed, or at least are symmetric and not excessively kurtotic (fat-tailed), and where the mean and variance are approximately constant. Observed time series frequently require some form of transformation before they exhibit these distributional properties, for in their 'raw' form they are often asymmetric." (Terence C Mills, "Applied Time Series Analysis: A practical guide to modeling and forecasting", 2019)

"Mean-averages can be highly misleading when the raw data do not form a symmetric pattern around a central value but instead are skewed towards one side [...], typically with a large group of standard cases but with a tail of a few either very high (for example, income) or low (for example, legs) values." (David Spiegelhalter, "The Art of Statistics: Learning from Data", 2019)

🔭Data Science: Definitions (Just the Quotes)

"The errors of definitions multiply themselves according as the reckoning proceeds; and lead men into absurdities, which at last they see but cannot avoid, without reckoning anew from the beginning." (Thomas Hobbes, "The Moral and Political Works of Thomas Hobbes of Malmesbury", 1750)

"It is the essence of a scientific definition to be causative, not by introduction of imaginary somewhats, natural or supernatural, under the name of causes, but by announcing the law of action in the particular case, in subordination to the common law of which all the phenomena are modifications or results." (Samuel T Coleridge, "Hints Towards the Formation of a More Comprehensive Theory of Life, The Nature of Life", 1847)

"The dimmed outlines of phenomenal things all merge into one another unless we put on the focusing-glass of theory, and screw it up sometimes to one pitch of definition and sometimes to another, so as to see down into different depths through the great millstone of the world." (James C Maxwell, "Are There Real Analogies in Nature?", 1856)

"Being built on concepts, hypotheses, and experiments, laws are no more accurate or trustworthy than the wording of the definitions and the accuracy and extent of the supporting experiments." (Gerald Holton, "Introduction to Concepts and Theories in Physical Science", 1952)

"We cannot define truth in science until we move from fact to law. And within the body of laws in turn, what impresses us as truth is the orderly coherence of the pieces. They fit together like the characters of a great novel, or like the words of a poem. Indeed, we should keep that last analogy by us always, for science is a language, and like a language it defines its parts by the way they make up a meaning. Every word in a sentence has some uncertainty of definition, and yet the sentence defines its own meaning and that of its words conclusively. It is the internal unity and coherence of science which gives it truth, and which makes it a better system of prediction than any less orderly language." (Jacob Bronowski, "The Common Sense of Science", 1953)

"Scientific method is the way to truth, but it affords, even in principle, no unique definition of truth. Any so-called pragmatic definition of truth is doomed to failure equally." (Willard v O Quine, "Word and Object", 1960)

"This other world is the so-called physical world image; it is merely an intellectual structure. To a certain extent it is arbitrary. It is a kind of model or idealization created in order to avoid the inaccuracy inherent in every measurement and to facilitate exact definition." (Max Planck, "The Philosophy of Physics", 1963)

"The assumptions and definitions of mathematics and science come from our intuition, which is based ultimately on experience. They then get shaped by further experience in using them and are occasionally revised. They are not fixed for all eternity." (Richard Hamming, "Methods of Mathematics Applied to Calculus, Probability, and Statistics", 1985)

"First, good statistics are based on more than guessing. [...] Second, good statistics are based on clear, reasonable definitions. Remember, every statistic has to define its subject. Those definitions ought to be clear and made public. [...] Third, good statistics are based on clear, reasonable measures. Again, every statistic involves some sort of measurement; while all measures are imperfect, not all flaws are equally serious. [...] Finally, good statistics are based on good samples." (Joel Best, "Damned Lies and Statistics: Untangling Numbers from the Media, Politicians, and Activists", 2001)

"While some social problems statistics are deliberate deceptions, many - probably the great majority - of bad statistics are the result of confusion, incompetence, innumeracy, or selective, self-righteous efforts to produce numbers that reaffirm principles and interests that their advocates consider just and right. The best response to stat wars is not to try and guess who's lying or, worse, simply to assume that the people we disagree with are the ones telling lies. Rather, we need to watch for the standard causes of bad statistics - guessing, questionable definitions or methods, mutant numbers, and inappropriate comparisons." (Joel Best, "Damned Lies and Statistics: Untangling Numbers from the Media, Politicians, and Activists", 2001)

"The goal of data science is to improve decision making by basing decisions on insights extracted from large data sets. As a field of activity, data science encompasses a set of principles, problem definitions, algorithms, and processes for extracting nonobvious and useful patterns from large data sets. It is closely related to the fields of data mining and machine learning, but it is broader in scope." (John D Kelleher & Brendan Tierney, "Data Science", 2018)

"Numbers can easily confuse us when they are unmoored from a clear definition." (Tim Harford, "The Data Detective: Ten easy rules to make sense of statistics", 2020)

"The whole discipline of statistics is built on measuring or counting things. […] it is important to understand what is being measured or counted, and how. It is surprising how rarely we do this. Over the years, as I found myself trying to lead people out of statistical mazes week after week, I came to realize that many of the problems I encountered were because people had taken a wrong turn right at the start. They had dived into the mathematics of a statistical claim - asking about sampling errors and margins of error, debating if the number is rising or falling, believing, doubting, analyzing, dissecting - without taking the ti- me to understand the first and most obvious fact: What is being measured, or counted? What definition is being used?" (Tim Harford, "The Data Detective: Ten easy rules to make sense of statistics", 2020)

More quotes on "Definitions" at the-web-of-knowledge.blogspot.com. 

🔭Data Science: Training (Just the Quotes)

"A neural network is characterized by (1) its pattern of connections between the neurons (called its architecture), (2) its method of determining the weights on the connections (called its training, or learning, algorithm), and (3) its activation function." (Laurene Fausett, "Fundamentals of Neural Networks", 1994)

"An artificial neural network (or simply a neural network) is a biologically inspired computational model that consists of processing elements (neurons) and connections between them, as well as of training and recall algorithms." (Nikola K Kasabov, "Foundations of Neural Networks, Fuzzy Systems, and Knowledge Engineering", 1996)

"[…] an obvious difference between our best classifiers and human learning is the number of examples required in tasks such as object detection. […] the difficulty of a learning task depends on the size of the required hypothesis space. This complexity determines in turn how many training examples are needed to achieve a given level of generalization error. Thus the complexity of the hypothesis space sets the speed limit and the sample complexity for learning." (Tomaso Poggio & Steve Smale, "The Mathematics of Learning: Dealing with Data", Notices of the AMS, 2003)

"Learning a complicated function that matches the training data closely but fails to recognize the underlying process that generates the data. As a result of overfitting, the model performs poor on new input. Overfitting occurs when the training patterns are sparse in input space and/or the trained networks are too complex." (Frank Padberg, "Counting the Hidden Defects in Software Documents", 2010)

"Decision trees are also considered nonparametric models. The reason for this is that when we train a decision tree from data, we do not assume a fixed set of parameters prior to training that define the tree. Instead, the tree branching and the depth of the tree are related to the complexity of the dataset it is trained on. If new instances were added to the dataset and we rebuilt the tree, it is likely that we would end up with a (potentially very) different tree." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)

"Boosting defines an objective function to measure the performance of a model given a certain set of parameters. The objective function contains two parts: regularization and training loss, both of which add to one another. The training loss measures how predictive our model is on the training data. The most commonly used training loss function includes mean squared error and logistic regression. The regularization term controls the complexity of the model, which helps avoid overfitting." (Danish Haroon, "Python Machine Learning Case Studies", 2017)

"Decision trees are important for a few reasons. First, they can both classify and regress. It requires literally one line of code to switch between the two models just described, from a classification to a regression. Second, they are able to determine and share the feature importance of a given training set." (Russell Jurney, "Agile Data Science 2.0: Building Full-Stack Data Analytics Applications with Spark", 2017)

"Early stopping and regularization can ensure network generalization when you apply them properly. [...] With early stopping, the choice of the validation set is also important. The validation set should be representative of all points in the training set. When you use Bayesian regularization, it is important to train the network until it reaches convergence. The sum-squared error, the sum-squared weights, and the effective number of parameters should reach constant values when the network has converged. With both early stopping and regularization, it is a good idea to train the network starting from several different initial conditions. It is possible for either method to fail in certain circumstances. By testing several different initial conditions, you can verify robust network performance." (Mark H Beale et al, "Neural Network Toolbox™ User's Guide", 2017)

"Variance is a prediction error due to different sets of training samples. Ideally, the error should not vary from one training sample to another sample, and the model should be stable enough to handle hidden variations between input and output variables. Normally this occurs with the overfitted model." (Umesh R Hodeghatta & Umesha Nayak, "Business Analytics Using R: A Practical Approach", 2017)

"One of the most common problems that you will encounter when training deep neural networks will be overfitting. What can happen is that your network may, owing to its flexibility, learn patterns that are due to noise, errors, or simply wrong data. [...] The essence of overfitting is to have unknowingly extracted some of the residual variation (i.e., the noise) as if that variation represented the underlying model structure. The opposite is called underfitting - when the model cannot capture the structure of the data." (Umberto Michelucci, "Applied Deep Learning: A Case-Based Approach to Understanding Deep Neural Networks", 2018)

"The premise of classification is simple: given a categorical target variable, learn patterns that exist between instances composed of independent variables and their relationship to the target. Because the target is given ahead of time, classification is said to be supervised machine learning because a model can be trained to minimize error between predicted and actual categories in the training data. Once a classification model is fit, it assigns categorical labels to new instances based on the patterns detected during training." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"The trick is to walk the line between underfitting and overfitting. An underfit model has low variance, generally making the same predictions every time, but with extremely high bias, because the model deviates from the correct answer by a significant amount. Underfitting is symptomatic of not having enough data points, or not training a complex enough model. An overfit model, on the other hand, has memorized the training data and is completely accurate on data it has seen before, but varies widely on unseen data. Neither an overfit nor underfit model is generalizable - that is, able to make meaningful predictions on unseen data." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"There is a trade-off between bias and variance [...]. Complexity increases with the number of features, parameters, depth, training epochs, etc. As complexity increases and the model overfits, the error on the training data decreases, but the error on test data increases, meaning that the model is less generalizable." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"Cross-validation is a useful tool for finding optimal predictive models, and it also works well in visualization. The concept is simple: split the data at random into a 'training' and a 'test' set, fit the model to the training data, then see how well it predicts the test data. As the model gets more complex, it will always fit the training data better and better. It will also start off getting better results on the test data, but there comes a point where the test data predictions start going wrong." (Robert Grant, "Data Visualization: Charts, Maps and Interactive Graphics", 2019)

"Any machine learning model is trained based on certain assumptions. In general, these assumptions are the simplistic approximations of some real-world phenomena. These assumptions simplify the actual relationships between features and their characteristics and make a model easier to train. More assumptions means more bias. So, while training a model, more simplistic assumptions = high bias, and realistic assumptions that are more representative of actual phenomena = low bias." (Imran Ahmad, "40 Algorithms Every Programmer Should Know", 2020)

🔭Data Science: Training Data (Just the Quotes)

"Neural networks are a computing technology whose fundamental purpose is to recognize patterns in data. Based on a computing model similar to the underlying structure of the human brain, neural networks share the brains ability to learn or adapt in response to external inputs. When exposed to a stream of training data, neural networks can discover previously unknown relationships and learn complex nonlinear mappings in the data. Neural networks provide some fundamental, new capabilities for processing business data. However, tapping these new neural network data mining functions requires a completely different application development process from traditional programming." (Joseph P Bigus, "Data Mining with Neural Networks: Solving business problems from application development to decision support", 1996)

"[…] an obvious difference between our best classifiers and human learning is the number of examples required in tasks such as object detection. […] the difficulty of a learning task depends on the size of the required hypothesis space. This complexity determines in turn how many training examples are needed to achieve a given level of generalization error. Thus the complexity of the hypothesis space sets the speed limit and the sample complexity for learning." (Tomaso Poggio & Steve Smale, "The Mathematics of Learning: Dealing with Data", Notices of the AMS, 2003)

"Learning a complicated function that matches the training data closely but fails to recognize the underlying process that generates the data. As a result of overfitting, the model performs poor on new input. Overfitting occurs when the training patterns are sparse in input space and/or the trained networks are too complex." (Frank Padberg, "Counting the Hidden Defects in Software Documents", 2010)

"Overfitting occurs when a formula describes a set of data very closely, but does not lead to any sensible explanation for the behavior of the data and does not predict the behavior of comparable data sets. In the case of overfitting, the formula is said to describe the noise of the system rather than the characteristic behavior of the system. Overfitting occurs frequently with models that perform iterative approximations on training data, coming closer and closer to the training data set with each iteration. Neural networks are an example of a data modeling strategy that is prone to overfitting." (Jules H Berman, "Principles of Big Data: Preparing, Sharing, and Analyzing Complex Information", 2013)

"Briefly speaking, to solve a Machine Learning problem means you optimize a model to fit all the data from your training set, and then you use the model to predict the results you want. Therefore, evaluating a model need to see how well it can be used to predict the data out of the training set. Usually there are three types of the models: underfitting, fair and overfitting model [...]. If we want to predict a value, both (a) and (c) in this figure cannot work well. The underfitting model does not capture the structure of the problem at all, and we say it has high bias. The overfitting model tries to fit every sample in the training set and it did it, but we say it is of high variance. In other words, it fails to generalize new data." (Shudong Hao, "A Beginner’s Tutorial for Machine Learning Beginners", 2014)

"Neural networks can model very complex patterns and decision boundaries in the data and, as such, are very powerful. In fact, they are so powerful that they can even model the noise in the training data, which is something that definitely should be avoided. One way to avoid this overfitting is by using a validation set in a similar way as with decision trees.[...] Another scheme to prevent a neural network from overfitting is weight regularization, whereby the idea is to keep the weights small in absolute sense because otherwise they may be fitting the noise in the data. This is then implemented by adding a weight size term (e.g., Euclidean norm) to the objective function of the neural network." (Bart Baesens, "Analytics in a Big Data World: The Essential Guide to Data Science and Its Applications", 2014)

"A predictive model overfits the training set when at least some of the predictions it returns are based on spurious patterns present in the training data used to induce the model. Overfitting happens for a number of reasons, including sampling variance and noise in the training set. The problem of overfitting can affect any machine learning algorithm; however, the fact that decision tree induction algorithms work by recursively splitting the training data means that they have a natural tendency to segregate noisy instances and to create leaf nodes around these instances. Consequently, decision trees overfit by splitting the data on irrelevant features that only appear relevant due to noise or sampling variance in the training data. The likelihood of overfitting occurring increases as a tree gets deeper because the resulting predictions are based on smaller and smaller subsets as the dataset is partitioned after each feature test in the path." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)

"Cross-validation is a method of splitting all of your data into two parts: training and validation. The training data is used to build the machine learning model, whereas the validation data is used to validate that the model is doing what is expected. This increases our ability to find and determine the underlying errors in a model." (Matthew Kirk, "Thoughtful Machine Learning", 2015)

"Tree pruning identifies and removes subtrees within a decision tree that are likely to be due to noise and sample variance in the training set used to induce it. In cases where a subtree is deemed to be overfitting, pruning the subtree means replacing the subtree with a leaf node that makes a prediction based on the majority target feature level (or average target feature value) of the dataset created by merging the instances from all the leaf nodes in the subtree. Obviously, pruning will result in decision trees being created that are not consistent with the training set used to build them. In general, however, we are more interested in creating prediction models that generalize well to new data rather than that are strictly consistent with training data, so it is common to sacrifice consistency for generalization capacity." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)

"When memorization happens, you may have the illusion that everything is working well because your machine learning algorithm seems to have fitted the in sample data so well. Instead, problems can quickly become evident when you start having it work with out-of-sample data and you notice that it produces errors in its predictions as well as errors that actually change a lot when you relearn from the same data with a slightly different approach. Overfitting occurs when your algorithm has learned too much from your data, up to the point of mapping curve shapes and rules that do not exist [...]. Any slight change in the procedure or in the training data produces erratic predictions." (John P Mueller & Luca Massaron, "Machine Learning for Dummies", 2016)

"Bias is error from incorrect assumptions built into the model, such as restricting an interpolating function to be linear instead of a higher-order curve. [...] Errors of bias produce underfit models. They do not fit the training data as tightly as possible, were they allowed the freedom to do so. In popular discourse, I associate the word 'bias' with prejudice, and the correspondence is fairly apt: an apriori assumption that one group is inferior to another will result in less accurate predictions than an unbiased one. Models that perform lousy on both training and testing data are underfit." (Steven S Skiena, "The Data Science Design Manual", 2017)

"Bias occurs normally when the model is underfitted and has failed to learn enough from the training data. It is the difference between the mean of the probability distribution and the actual correct value. Hence, the accuracy of the model is different for different data sets (test and training sets). To reduce the bias error, data scientists repeat the model-building process by resampling the data to obtain better prediction values." (Umesh R Hodeghatta & Umesha Nayak, "Business Analytics Using R: A Practical Approach", 2017)

"Boosting defines an objective function to measure the performance of a model given a certain set of parameters. The objective function contains two parts: regularization and training loss, both of which add to one another. The training loss measures how predictive our model is on the training data. The most commonly used training loss function includes mean squared error and logistic regression. The regularization term controls the complexity of the model, which helps avoid overfitting." (Danish Haroon, "Python Machine Learning Case Studies", 2017)

"From a typical training set, many alternative decision trees can be created. As a rule, smaller trees are to be preferred, their main advantages being interpretability, removal of irrelevant and redundant attributes, and lower danger of overfitting noisy training data." (Miroslav Kubat, "An Introduction to Machine Learning" 2nd Ed., 2017)

"High-bias models typically produce simpler models that do not overfit and in those cases the danger is that of underfitting. Models with low-bias are typically more complex and that complexity enables us to represent the training data in a more accurate way. The danger here is that the flexibility provided by higher complexity may end up representing not only a relationship in the data but also the noise. Another way of portraying the bias-variance trade-off is in terms of complexity v simplicity." (Jesús Rogel-Salazar, "Data Science and Analytics with Python", 2017) 

"In machine learning, a model is defined as a function, and we describe the learning function from the training data as inductive learning. Generalization refers to how well the concepts are learned by the model by applying them to data not seen before. The goal of a good machine-learning model is to reduce generalization errors and thus make good predictions on data that the model has never seen." (Umesh R Hodeghatta & Umesha Nayak, "Business Analytics Using R: A Practical Approach", 2017)

"Multilayer perceptrons share with polynomial classifiers one unpleasant property. Theoretically speaking, they are capable of modeling any decision surface, and this makes them prone to overfitting the training data."  (Miroslav Kubat," An Introduction to Machine Learning" 2nd Ed., 2017)

"The danger of overfitting is particularly severe when the training data is not a perfect gold standard. Human class annotations are often subjective and inconsistent, leading boosting to amplify the noise at the expense of the signal. The best boosting algorithms will deal with overfitting though regularization. The goal will be to minimize the number of non-zero coefficients, and avoid large coefficients that place too much faith in any one classifier in the ensemble." (Steven S Skiena, "The Data Science Design Manual", 2017)

"Variance is a prediction error due to different sets of training samples. Ideally, the error should not vary from one training sample to another sample, and the model should be stable enough to handle hidden variations between input and output variables. Normally this occurs with the overfitted model." (Umesh R Hodeghatta & Umesha Nayak, "Business Analytics Using R: A Practical Approach", 2017)

"Variance is error from sensitivity to fluctuations in the training set. If our training set contains sampling or measurement error, this noise introduces variance into the resulting model. [...] Errors of variance result in overfit models: their quest for accuracy causes them to mistake noise for signal, and they adjust so well to the training data that noise leads them astray. Models that do much better on testing data than training data are overfit." (Steven S Skiena, "The Data Science Design Manual", 2017)

"The premise of classification is simple: given a categorical target variable, learn patterns that exist between instances composed of independent variables and their relationship to the target. Because the target is given ahead of time, classification is said to be supervised machine learning because a model can be trained to minimize error between predicted and actual categories in the training data. Once a classification model is fit, it assigns categorical labels to new instances based on the patterns detected during training." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"The trick is to walk the line between underfitting and overfitting. An underfit model has low variance, generally making the same predictions every time, but with extremely high bias, because the model deviates from the correct answer by a significant amount. Underfitting is symptomatic of not having enough data points, or not training a complex enough model. An overfit model, on the other hand, has memorized the training data and is completely accurate on data it has seen before, but varies widely on unseen data. Neither an overfit nor underfit model is generalizable - that is, able to make meaningful predictions on unseen data." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"There is a trade-off between bias and variance [...]. Complexity increases with the number of features, parameters, depth, training epochs, etc. As complexity increases and the model overfits, the error on the training data decreases, but the error on test data increases, meaning that the model is less generalizable." (Benjamin Bengfort et al, "Applied Text Analysis with Python: Enabling Language-Aware Data Products with Machine Learning", 2018)

"Cross-validation is a useful tool for finding optimal predictive models, and it also works well in visualization. The concept is simple: split the data at random into a 'training' and a 'test' set, fit the model to the training data, then see how well it predicts the test data. As the model gets more complex, it will always fit the training data better and better. It will also start off getting better results on the test data, but there comes a point where the test data predictions start going wrong." (Robert Grant, "Data Visualization: Charts, Maps and Interactive Graphics", 2019)

"The classifier accuracy would be extra ordinary when the test data and the training data are overlapping. But when the model is applied to a new data it will fail to show acceptable accuracy. This condition is called as overfitting." (Jesu V  Nayahi J & Gokulakrishnan K, "Medical Image Classification", 2019)