PCA is a form of dimension reduction

It groups or reduces the number of features (dimensions) in your data into a more manageable number of columns.

PCA is a simple type of Singular Value Decomposition (SVD)

SVD is a linear algebra method for refactoring a matrix (dataset) into a smaller matrix by finding multiple lines of best fit.

It’s closely related to Factor Analysis.

Think of a pizza…

PCA can be used for…

• Exploration and Visualization (usually uses 2 dimensions)
• Modeling and Data Preparation (any number of dimensions)

Choose a number of components appropriate to your use case!

You don’t need to split your data, but you should definitely standardize it!

As always pay attention to the hyperparameters.

• n_components: the number of components the model will produce
• whiten: an adjustment of results, good for when you will use components in another model
• svd_solver: the exact SVD method you’ll use, usually can be left as ‘auto’
• random_state

Use .fit_transform() to fit the model and get components in the same function.

The resulting data will have the same number of rows but a new number of columns.

Put this data into a new dataframe to do something with it!

The .components_ and .explained_variance_ attributes can help you understand your results.

Try PCA for the penguins dataset.

1. Select features and prepare data. Consider standardizations as well as null values.
2. Run PCA to reduce to 2 dimensions and plot the results.
3. Run PCA with 3 dimensions and use .components_ to assess results.
4. Use the 3-dimension PCA results to re-run last week’s K-means clustering.

Good luck! 🐧🐧🐧