https://training-course-material.com/index.php?action=history&feed=atom&title=Machine_Learning_with_RMachine Learning with R - Revision history2026-05-13T23:38:49ZRevision history for this page on the wikiMediaWiki 1.45.1https://training-course-material.com/index.php?title=Machine_Learning_with_R&diff=24675&oldid=prevBernard Szlachta at 00:37, 17 February 20152015-02-17T00:37:16Z<p></p>
<p><b>New page</b></p><div>[[Category:R]]<br />
[[Category:Machine Learning]]<br />
<br />
<slideshow style="nobleprog" headingmark="⌘" incmark="…" scaled="false" font="Trebuchet MS" ><br />
;title: Introduction to R with exercises<br />
;author: MIHALY BARASZ for NobleProg Ltd<br />
</slideshow><br />
<br />
== TABLE OF CONTENTS ⌘==<br />
* Sources and further reading<br />
* Machine Learning vs. Statistical Learning<br />
* Linear regression<br />
* Exercise for linear regression<br />
* Exercise for linear regression (contd.)<br />
* R best practices<br />
* Logistic regression<br />
* Testing, cross-validation<br />
* Classification exercise<br />
* Presenting the results<br />
* Deploying your results<br />
* Generalized Linear Models<br />
* Generalized Linear Model (cont.)<br />
* Regularization<br />
* Regularization more generally<br />
== 1 SOURCES AND FURTHER READING ⌘==<br />
Source materials<br />
* “An Introduction to Statistical Learning”<br />
** Available for free in PDF form online<br />
** Online course by Trevor Hastie and Rob Tibshirani<br />
* Andrew Ng's “Machine Learning” online course<br />
Further reading<br />
* “Think Stats” and “Think Bayes”<br />
**both by Allen B. Downey<br />
**both available for free online<br />
**programming in Python<br />
== 2 MACHINE LEARNING VS. STATISTICAL LEARNING ⌘==<br />
* Different origins<br />
* Different focus<br />
* Highly convergent in the recent years<br />
== 3 LINEAR REGRESSION ⌘==<br />
The simplest model for estimating a numerical response<br />
Y=β0+β1X1+β2X2+⋯+βpXp+ε<br />
Details<br />
* Understanding the results<br />
* Assessing the accuracy<br />
* Iterpreting the coefficients<br />
* Understanding factors<br />
* Adding higher oder terms and interactions<br />
== 4 EXERCISE FOR LINEAR REGRESSION ⌘==<br />
* Data file: Advertising.csv (from ISLR)<br />
* Multivariate linear regression<br />
**Which variables are important<br />
**Do they not have any predicting power?<br />
**How much precision do we lose by dropping the "unimportant" variables?<br />
== 5 EXERCISE FOR LINEAR REGRESSION (CONTD.) ⌘==<br />
* Interactions between variables<br />
Regression with all interactions<br />
**Comparing results<br />
**What are interations<br />
**Visualizing interactions<br />
<br />
== 6 R BEST PRACTICES ⌘==<br />
* Organizing your work (and data)<br />
* Reusable work<br />
* Plotting<br />
* Learning<br />
== 7 LOGISTIC REGRESSION ⌘==<br />
Response is categorical: Yes or No.<br />
f(X)=β0+β1X1+β2X2+⋯+βpXp<br />
Find a suitable f(X) and classify to Yes if f(X)>0 and to No otherwise. What is a good f?<br />
* Minimizes the training error? Not fine-grained enough; hard to optimize for.<br />
* Map f(X) to probabilities and maximize for the likelihood of training data.<br />
== 8 TESTING, CROSS-VALIDATION ⌘==<br />
* Training vs. Test-set performance<br />
* Bias-Variance trade-off (under/overfitting)<br />
* Strategies for estimating test error; Cross-Validation<br />
* Bootstra<br />
== 9 CLASSIFICATION EXERCISE ⌘==<br />
* Data: "defaulters" from the ISLR package<br />
== 10 PRESENTING THE RESULTS ⌘==<br />
* Session in R Markdown<br />
== 11 DEPLOYING YOUR RESULTS ⌘==<br />
* Exporting a model to a spreadsheet<br />
* Porting to a different programming environment<br />
* Using R as a library<br />
* Deploying R applications to web<br />
**Shiny: http://shiny.rstudio.com/<br />
== 12 GENERALIZED LINEAR MODELS ⌘==<br />
* What's common in linear regression and logistic regression?<br />
* How do they fit under one common assumption?<br />
* What is the family parameter in glm?<br />
== 13 GENERALIZED LINEAR MODEL (CONT.) ⌘==<br />
* Common underlying assumption: a linear function of the predictors determines the distribution of the response.<br />
* The parameters of the linear function are determined in a way to maximize the likelihood of the observations.<br />
f(X)=β0+β1X1+β2X2+⋯+βpXp<br />
For example, given the value of predictors X we assume that the distribution of the response depends only on f(X):<br />
* Linear regression: N(f(X),σ2) with a constant σ2 (its value doesn't matter)<br />
* Two class classification: binomial, with the probability of the Yes class being p, where logp1−p=f(X)<br />
Deviance: negative log likelihood (times two :)). This is what we actually minimize in practice. In case of linear regression…<br />
== 14 REGULARIZATION ⌘==<br />
* Prediction accuracy; especially if p>n.<br />
* Model interpretability: removes irrelevant features. Feature selection.<br />
== 15 REGULARIZATION MORE GENERALLY ⌘==<br />
Methods<br />
* Subset selection<br />
* Shrinkage (aka. regularization). Ridge regression, lasso<br />
* Dimension reduction. Pricipal components regression; partial least squares.<br />
== 16 REGULARIZATION EXERCISE ⌘==<br />
* Data: regul.csv<br />
== 17 TREE BASED METHODS ⌘==<br />
* Decision trees<br />
* Random forests (baggin, bootstrap)<br />
* Boosting<br />
== 18 UNSUPERVISED LEARNING ⌘==<br />
* Reasons, goals<br />
* Methods<br />
== 19 PRINCIPAL COMPONENTS ANALYSIS ⌘==<br />
== 20 CLUSTERING ⌘==<br />
* Goals<br />
* Examples<br />
* Challenges<br />
== 21 K-MEANS CLUSTERING ⌘==<br />
Demonstration of R magic</div>Bernard Szlachta