biolab / orange-lecture-notes Goto Github PK

Add a chapter at the end of the book with a how-to on organizing the data:

• Select Columns
• Select Rows
• Discretize
• Continuize
• Edit Domain
• Impute
• Feature Constructor?

Add page numbers, skipping the title page (and possibly index).

Add a simple task with painted data (diagonal split) and try Tree on it. Then try Logistic Regression. Observe the tree and observe the nomogram. Reason about the difference.

If external people contribute a chapter to this repository, would it be possible to name them as chapter authors and where?

Just an idea, if at any time we get more people contributing. :)

Wouldn't it be a good idea to bring overfitting in classification (How to Cheat) under one umbrella with overfitting in regression? Currently, overfitting in classification is introduced under a different name before the notion of overfitting is introduced. This is somewhat confusing.

Add a chapter on MDS visualization.

Add chapter on model calibration, showing Calibration plot, Calibrated model. Use heart-disease.

The "map" at the end of this chapter shows the US cities mirrored horizontally (flipped), or from the perspective of inside the earth - not from Australian perspective. That would have been upside down.

Add an assignment explaining the difference between clustering and classification. Focus on the difference between important features (classification) and all features (clustering) and how clustering corresponds (or doesn't) to class labels.

Add an assignment that makes people think about PCA on random data. Talk about 2 PCA components (visualizations) and multiple PCA variables in the model (pros and cons).

The chapter on Classification starts with the sentence "We have seen the iris data before". However, the iris dataset in the first chapter of older lecture notes has now been replaced by a gene expression dataset, so the reader may not have seen it before.
By the way, the gene expression dataset is rather hard to understand for non-biologists. I would prefer a return to the iris dataset in the first chapter.

Add a chapter on t-SNE visualization.