"Graphic methods convey to the mind a more comprehensive grasp of essential features than do written reports, because one can naturally gather interesting details from a picture in far less time than from a written description. Further than this, the examination of a picture allows one to make deductions of his own, while in the case of a written description the reader must, to a great degree, accept the conclusions of the author." (Allan C Haskell, "How to Make and Use Graphic Charts", 1919)
"An important rule in the drafting of curve charts is that the amount scale should begin at zero. In comparisons of size the omission of the zero base, unless clearly indicated, is likely to give a misleading impression of the relative values and trend." (Rufus R Lutz, "Graphic Presentation Simplified", 1949)
"The quantile plot is a good general display since it is fairly easy to construct and does a good job of portraying many aspects of a distribution. Three convenient features of the plot are the following: First, in constructing it, we do not make any arbitrary choices of parameter values or cell boundaries [...] and no models for the data are fitted or assumed. Second, like a table, it is not a summary but a display of all the data. Third, on the quantile plot every point is plotted at a distinct location, even if there are duplicates in the data. The number of points that can be portrayed without overlap is limited only by the resolution of the plotting device. For a high resolution device several hundred points distinguished." (John M Chambers et al, "Graphical Methods for Data Analysis", 1983)
"Maps containing marks that indicate a variety of features at specific locations are easy to produce and often revealing for the reader. You can use dots, numbers, and shapes, with or without keys. The basic map must always be simple and devoid of unnecessary detail. There should be no ambiguity about what happens where." (Bruce Robertson, "How to Draw Charts & Diagrams", 1988)
"Maps used as charts do not need fine cartographic detail. Their purpose is to express ideas, explain relationships, or store data for consultation. Keep your maps simple. Edit out irrelevant detail. Without distortion, try to present the facts as the main feature of your map, which should serve only as a springboard for the idea you're trying to put across." (Bruce Robertson, "How to Draw Charts & Diagrams", 1988)
"Scatter charts show the relationships between information, plotted as points on a grid. These groupings can portray general features of the source data, and are useful for showing where correlationships occur frequently. Some scatter charts connect points of equal value to produce areas within the grid which consist of similar features." (Bruce Robertson, "How to Draw Charts & Diagrams", 1988)
"Boxplots provide information at a glance about center (median), spread (interquartile range), symmetry, and outliers. With practice they are easy to read and are especially useful for quick comparisons of two or more distributions. Sometimes unexpected features such as outliers, skew, or differences in spread are made obvious by boxplots but might otherwise go unnoticed." (Lawrence C Hamilton, "Regression with Graphics: A second course in applied statistics", 1991)
"A well-constructed graph can show several features of the data at once. Some graphs contain as much information as the original data, and so (unlike numerical summaries) do not actually simplify the data; rather, they express it in visual form. Unexpected or unusual features, which are not obvious within numerical tables, often jump to our attention once we draw a graph. Because the strengths and weaknesses of graphical methods are opposite those of numerical summary methods, the two work best in combination." (Lawrence C Hamilton, "Data Analysis for Social Scientists: A first course in applied statistics", 1995)
"When analyzing data it is many times advantageous to generate a variety of graphs using the same data. This is true whether there is little or lots of data. Reasons for this are: (1) Frequently, all aspects of a group of data can not be displayed on a single graph. (2) Multiple graphs generally result in a more in-depth understanding of the information. (3) Different aspects of the same data often become apparent. (4) Some types of graphs cause certain features of the data to stand out better (5) Some people relate better to one type of graph than another." (Robert L Harris, "Information Graphics: A Comprehensive Illustrated Reference", 1996)
"[…] a chart is a picture of relationships, and only the picture counts. Everything else - titles, labels, scale values - merely identifies and explains. The most important feature of the picture is the impression you receive. Scaling has an important controlling effect on that impression." (Gene Zelazny. "Say It with Charts: The executive’s guide to visual communication" 4th Ed., 2001)
"Anyone who has seen, and especially used, a highly responsive interactive visualization tool will be struck by two features. First, that a mere rearrangement of how the data is displayed can lead to a surprising degree of additional insight into that data. Second, that the very property of interactivity can considerably enhance that tool's effectiveness, especially if the computer's response follows a user's action virtually immediately, say within a fraction of a second." (Robert Spence, "Information Visualization", 2001)
"A feature shared by both the range and the interquartile range is that they are each calculated on the basis of just two values - the range uses the maximum and the minimum values, while the IQR uses the two quartiles. The standard deviation, on the other hand, has the distinction of using, directly, every value in the set as part of its calculation. In terms of representativeness, this is a great strength. But the chief drawback of the standard deviation is that, conceptually, it is harder to grasp than other more intuitive measures of spread." (Alan Graham, "Developing Thinking in Statistics", 2006)
"A useful feature of a stem plot is that the values maintain their natural order, while at the same time they are laid out in a way that emphasises the overall distribution of where the values are concentrated (that is, where the longer branches are). This enables you easily to pick out key values such as the median and quartiles." (Alan Graham, "Developing Thinking in Statistics", 2006)
"Old code rarely offers trendy graphics or flavor-of-the-month features, but it has one considerable ad - vantage: It tends to work. A program that has been well used is like an old garden that has been well tended or a vintage guitar that has been well played: Its rough edges have been filed away, its bugs have been found and fixed, and its performance is a known and valuable quantity." (Scott Rosenberg, "Dreaming in Code", 2007)
"Multivariate techniques often summarize or classify many variables to only a few groups or factors (e.g., cluster analysis or multi-dimensional scaling). Parallel coordinate plots can help to investigate the influence of a single variable or a group of variables on the result of a multivariate procedure. Plotting the input variables in a parallel coordinate plot and selecting the features of interest of the multivariate procedure will show the influence of different input variables." (Martin Theus & Simon Urbanek, "Interactive Graphics for Data Analysis: Principles and Examples", 2009)
"Parallel coordinate plots are often overrated concerning their ability to depict multivariate features. Scatterplots are clearly superior in investigating the relationship between two continuous variables and multivariate outliers do not necessarily stick out in a parallel coordinate plot. Nonetheless, parallel coordinate plots can help to find and understand features such as groups/clusters, outliers and multivariate structures in their multivariate context. The key feature is the ability to select and highlight individual cases or groups in the data, and compare them to other groups or the rest of the data." (Martin Theus & Simon Urbanek, "Interactive Graphics for Data Analysis: Principles and Examples", 2009)
"Histograms are often mistaken for bar charts but there are important differences. Histograms show distribution through the frequency of quantitative values (y axis) against defined intervals of quantitative values(x axis). By contrast, bar charts facilitate comparison of categorical values. One of the distinguishing features of a histogram is the lack of gaps between the bars [...]" (Andy Kirk, "Data Visualization: A successful design process", 2012)
"When various types of data are layered directly on top of one another, the viewer is able to spatially correlate multiple features. This is immediately intuitive in the case of spatial relationships […]" (Felice C Frankel & Angela H DePace, "Visual Strategies", 2012)
"When you decide how to depict your data, you decide on the abstraction. Will you present a graph? A cartoon? An accurate molecular model? And which features will you include in these representations? Your preferred abstraction should include all necessary information, exclude unnecessary information, and make use of your reader’s preexisting knowledge without being confined by it." (Felice C Frankel & Angela H DePace, "Visual Strategies", 2012)
"Three high-level targets are very broadly relevant, for all kinds of data: trends, outliers, and features. A trend is a high-level characterization of a pattern in the data. Simple examples of trends include increases, decreases, peaks, troughs, and plateaus. Almost inevitably, some data doesn’t fit well with that backdrop; those elements are the outliers. The exact definition of features is task dependent, meaning any particular structures of interest." (Tamara Munzner, "Visualization Analysis and Design", 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)
"The main advantage of decision tree models is that they are interpretable. It is relatively easy to understand the sequences of tests a decision tree carried out in order to make a prediction. This interpretability is very important in some domains. [...] Decision tree models can be used for datasets that contain both categorical and continuous descriptive features. A real advantage of the decision tree approach is that it has the ability to model the interactions between descriptive features. This arises from the fact that the tests carried out at each node in the tree are performed in the context of the results of the tests on the other descriptive features that were tested at the preceding nodes on the path from the root. Consequently, if there is an interaction effect between two or more descriptive features, a decision tree can model this." (John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)
"There are two kinds of mistakes that an inappropriate inductive bias can lead to: underfitting and overfitting. Underfitting occurs when the prediction model selected by the algorithm is too simplistic to represent the underlying relationship in the dataset between the descriptive features and the target feature. Overfitting, by contrast, occurs when the prediction model selected by the algorithm is so complex that the model fits to the dataset too closely and becomes sensitive to noise in the data."(John D Kelleher et al, "Fundamentals of Machine Learning for Predictive Data Analytics: Algorithms, Worked Examples, and Case Studies", 2015)
No comments:
Post a Comment