Applied Data Science and Visualization 2020
Summary by Lode Notermans
Lecture 1
Different kinds of data analysis
Explorative data analysis (EDA)
Unsupervised learning
Supervised learning
Trade-off between prediction accuracy and model interpretability
Measuring the quality of fit
Bias Variance trade-off
Lecture 2
Grammar of graphics
Colors are perceived relatively
Lecture 3
Cross validation
Leave-one-out cross-validation (LOOCV)
k-fold Cross validation
Somewhat higher bias than LOOCV due to smaller training set for each iteration
(but smaller bias than Validation-set method due to larger training set than VS.)
Bias-Variance Trade-Off for k-Fold Cross-Validation
Using CV for classification problems
Lecture 4
Model Selection and Regularization
Subset selection
Forward stepwise selection
Backward stepwise selection
Penalized (regularized) regression
Ridge regression
Lasso regression
Lasso versus ridge regression, which method is better?
Selecting the best λ
Lecture 5
k-Nearest neighbors
Logistic regression
Interpretation of logistic regression
Evaluation of classifiers
Confusion matrix
ROC curve
,Lecture 6
Polynomial regression
Step functions
Splines
Natural splines
Choosing K (number of knots) and placement
Smoothing splines
Local regression
Multiple predictors: Generalized additive models
Conclusions
Lecture 7
What is a Shiny app?
Building basic UI
Reactivity
Final remarks on shiny
Lecture 8
Lecture 9
,Lecture 1
ISLR Chapter 1, 2
Different kinds of data analysis
Exploratory Confirmatory
Description EDA, One-sample t-test
Unsupervised learning
Prediction Supervised learning Macroeconomics
Explanation John Snow and the cholera Causal modeling
outbreak
Prescription Personalized medicine A/B testing
Exploratory data analysis (EDA)
Describing interesting patterns: use graphs,
summaries, to understand subgroups, detect
anomalies (“outliers”), understand the data
Examples: boxplot, five-number summary,
histograms, scatterplots…
Unsupervised learning
Inputs, but no outputs. Try to learn
structure and relationships from
these data, like detecting
unobserved groups (clusters) from
the data.
Assumptions about structural properties of the data
Dimension reduction methods
, ● Principal components analysis
● Factor analysis
● Random projections
Clustering
● K-means clustering
Supervised learning
Building a statistical model for predicting /
estimating an output based on one or more
inputs.
Most widely used machine-learning methods
are supervised
● Spam classifiers of e-mail
● Face recognizers over images
● Medical diagnosis systems for patients
Methods include
● (logistic) Regression
● Decision trees/random forests
● Support vector machines
● Neural networks
Classification Regression
Classification: predict to which category an Regression: predict a quantitative outcome
observation belongs (qualitative outcomes)
Trade-off between prediction accuracy and model interpretability
If the researcher is mainly interested in inference, a
more restrictive model is more useful, but when the
researcher is interested in prediction a more loose
model is more useful. While one might argue that a
more flexible model is always better, flexible models
lead to such complicated estimates of f that it is
difficult to understand how any individual predictor is
associated with the response. So a trade-off between
flexibility and interpretability has to be found. Because
Summary by Lode Notermans
Lecture 1
Different kinds of data analysis
Explorative data analysis (EDA)
Unsupervised learning
Supervised learning
Trade-off between prediction accuracy and model interpretability
Measuring the quality of fit
Bias Variance trade-off
Lecture 2
Grammar of graphics
Colors are perceived relatively
Lecture 3
Cross validation
Leave-one-out cross-validation (LOOCV)
k-fold Cross validation
Somewhat higher bias than LOOCV due to smaller training set for each iteration
(but smaller bias than Validation-set method due to larger training set than VS.)
Bias-Variance Trade-Off for k-Fold Cross-Validation
Using CV for classification problems
Lecture 4
Model Selection and Regularization
Subset selection
Forward stepwise selection
Backward stepwise selection
Penalized (regularized) regression
Ridge regression
Lasso regression
Lasso versus ridge regression, which method is better?
Selecting the best λ
Lecture 5
k-Nearest neighbors
Logistic regression
Interpretation of logistic regression
Evaluation of classifiers
Confusion matrix
ROC curve
,Lecture 6
Polynomial regression
Step functions
Splines
Natural splines
Choosing K (number of knots) and placement
Smoothing splines
Local regression
Multiple predictors: Generalized additive models
Conclusions
Lecture 7
What is a Shiny app?
Building basic UI
Reactivity
Final remarks on shiny
Lecture 8
Lecture 9
,Lecture 1
ISLR Chapter 1, 2
Different kinds of data analysis
Exploratory Confirmatory
Description EDA, One-sample t-test
Unsupervised learning
Prediction Supervised learning Macroeconomics
Explanation John Snow and the cholera Causal modeling
outbreak
Prescription Personalized medicine A/B testing
Exploratory data analysis (EDA)
Describing interesting patterns: use graphs,
summaries, to understand subgroups, detect
anomalies (“outliers”), understand the data
Examples: boxplot, five-number summary,
histograms, scatterplots…
Unsupervised learning
Inputs, but no outputs. Try to learn
structure and relationships from
these data, like detecting
unobserved groups (clusters) from
the data.
Assumptions about structural properties of the data
Dimension reduction methods
, ● Principal components analysis
● Factor analysis
● Random projections
Clustering
● K-means clustering
Supervised learning
Building a statistical model for predicting /
estimating an output based on one or more
inputs.
Most widely used machine-learning methods
are supervised
● Spam classifiers of e-mail
● Face recognizers over images
● Medical diagnosis systems for patients
Methods include
● (logistic) Regression
● Decision trees/random forests
● Support vector machines
● Neural networks
Classification Regression
Classification: predict to which category an Regression: predict a quantitative outcome
observation belongs (qualitative outcomes)
Trade-off between prediction accuracy and model interpretability
If the researcher is mainly interested in inference, a
more restrictive model is more useful, but when the
researcher is interested in prediction a more loose
model is more useful. While one might argue that a
more flexible model is always better, flexible models
lead to such complicated estimates of f that it is
difficult to understand how any individual predictor is
associated with the response. So a trade-off between
flexibility and interpretability has to be found. Because
