class Learner[source]

Learner(X_train:ndarray, y_train:ndarray, X_test:ndarray, y_test:ndarray, ohe:bool=False, scaler:bool=False, pca:bool=True, pca_n_components:int=50, param_grids:list=None)

Class for training and prediction.

	Type	Default	Details
X_train	ndarray		X_train numpy ndarray
y_train	ndarray		y_train numpy ndarray
X_test	ndarray		X_test numpy ndarray
y_test	ndarray		y_test numpy ndarray
ohe	bool	False	to use one hot encoding or not
scaler	bool	False	to use standard scaling or not
pca	bool	True	to use principal component analysis or not
pca_n_components	int	50	PCA number of components
param_grids	list	None	param_grid for grid search, if None - gets default grid from utils

Learner.create_pipeline[source]

Learner.create_pipeline()

Create and return pipeline

Learner.train[source]

Learner.train(scoring:str='accuracy', cv:int=5, n_jobs:int=-1)

Run GridSearchCV for all models on X_train and y_train of dataset.
Returns:
    train_results: list of grid search results
    grid_list: list of trained grid objects

	Type	Default	Details
scoring	str	accuracy	must be one of https://scikit-learn.org/stable/modules/model_evaluation.html#scoring-parameter
cv	int	5	defaults to 5-fold CV
n_jobs	int	-1	defaults to -1 to use all cores

Learner.get_top_5_train_results[source]

Learner.get_top_5_train_results()

Return top 5 results for each grid

Learner.predict[source]

Learner.predict()

Get predictions on the dataset's X_test from best estimators of GridSearchCV.

Learner.pick_k[source]

Learner.pick_k(max_clusters:int=10, pca_n_components:int=50)

Plot elbow and silohutte curves & print silohutte scores to help determine the ideal 'k' for Kmeans.

	Type	Default	Details
max_clusters	int	10	max number of clusters to try out
pca_n_components	int	50	number of components to reduce to in PCA

The pick_k method does the following to help determine the ideal k for KMeans:

• It first concats X_train and X_test of this dataset into a single ndarray 'X'
• then encodes X using OneHotEncoder
• then sclaes X using StandardScaler
• then dimensionality reduces X using PCA
• then plots elbow & silhouette plots for X and prints silhouette scores, and returns the PCA-reduced X.

Learner.analyze_clusters[source]

Learner.analyze_clusters(X_pca:ndarray, k:int, random_state:int=10)

Perform KMeans clustering, print cluster counts and plot clusters from the result.

	Type	Default	Details
X_pca	ndarray		dim reduced X numpy ndarray
k	int		the chosen value of k for KMeans
random_state	int	10	random state for KMeans

Learner.run_label_spreading[source]

Learner.run_label_spreading(pca_n_components:int=50)

Run Label Spreading, print report, append results to predict_results.

	Type	Default	Details
pca_n_components	int	50	number of components to reduce to in PCA