eda-fe-model 0.3.2

A python package to handle EDA and feature extraction and also return the best hyperparameters for a tabular classification problem.

EDA FEATURE_EXTRACTOR MODEL

A python package to do EDA, feature selection and display the best hyperparameters for a pre-built classification model.

Useful for datasets with no NaNs or null values present. Can be used for normal classification tasks, next update will work for Regression type problems and incorporate a sorted arrangement of variables.

before utilizing the package, ensure no Null or NaN values remain.

pip install eda-fe-model

pip install eda-fe-model==0.3.2

Using the library

from eda_fe_model import package

package.EDA()
package.feature_extraction()

Use to_categorical from keras.utils, to One Hot Encode the labels

package.build_best_model()
package.model_create()

EDA

package.EDA accepts the following:

        dataset = pandas dataframe
        columns_drop = columns to drop as a list. Accepts None
        one_hot_encode = True/False
        label_encode = True/False
        normalize = True/False
        standardize = True/False
        target_varaible = single target, y, as array
        test_size = percentage of the dataset to be used for testing purposes
        random_state

If the dataset only consists of categorical variables, set normalize or standardize to True.

returns the splitted dataset: x_train, x_test, y_train, y_test (respectively)

FEATURE EXTRACTION

package.feature_extraction accepts the following:

        train_X = train dataset consisting of predictors
        train_Y = train labels
        test_X = test dataset consisting of predictors
        test_Y = test labels
        rfe = True/False; Do you want to use Random Feature Extractor
        dim_out = Used only if rfe=True; output dimension; number of features to be selected 
        distribution = Distibution of the dataset you want to use for GLM

If rfe is False, set dim_out and distribution to be None, to return the input x and y for train and test datasets.
Try changing the distribution if error due to convergence appear.

returns x_train and x_test datasets with the user entered dimension/predictors

BUILD BEST MODEL

package.build_best_model accepts the follwoing:

        x = train dataset consisting of predictors
        y = One HOt Encoded training labels

returns a RandomizedSearchCV object.

Best Score: results.best_score_
Best Parameters: results.best_params_

CREATING THE MODEL

package.model_create accpets the best parameters from the build_best_model() and runs the model for a user specified epochs.

        x = the new train dataset consisting of just the predictors.
        y = One Hot encoded training labels