Repository navigation
Simple code to do clustering and SSE plot #3
Description
Activity
Progress:
- implemented a stripped-down version of Inverse-features that only creates an elbow graph
- implemented cut switch and cut_number variables. Cut can be found in the lines after the import of the pickle file. Setting cut to True, means the resultant dataframe will output a smaller one cut down to the number of rows set in cut_number.
- The Kmeans graph has clusters up to 15. The elbow/sudden drop occurs at ~6 clusters.
- The OUTPUT: cluster_dicts dictionary. You may set num_clusters to any number. The kmeans SSE clustering will cluster the dataframe by num_clusters and append to the dictionary. eg. num_clusters = 3, outputs three dataframes like this:
{"1": dataframe, "2": dataframe, "3": dataframe}
Notes:
- PCA Code is identical in inverse_features and PCA notebook
- It appears that the inverse features are already implemented in the original pickle files
Further progress:
- Created a function that returns a cluster DataFrame. Inputs are: Num_clusters, clustering_column_name (use to determine what the column name of the cluster column will be).
- One can use this function to return a pandas DataFrame which can then be saved using .to_pickle() to an appropriate directory.
- This function provides a very quick way of doing clustering.
Consider inverse_features code and strip it from everything except importing data frame, clustering, and plotting SSE. Test it for several small dataframes.
Outputs: one dataframe per number of clusters.
Don't add inverse features - it should be a separate function that expands the data frame.