modelTrainingFunctions_original.py

import pandas as pd
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import SGDClassifier


def modelTraining(X_train,y_train,ID_train):

    ID_list = ID_train.drop_duplicates().tolist()
    pred_probs = pd.DataFrame([])

    #leave-one-out based on recordings (we leave out one recording as test at a time)
    for i in range(0,len(ID_list)):

        idnow = ID_list[i]
        ID_train_list = ID_train.to_list()
        val_index = [i for i, x in enumerate(ID_train_list) if x == idnow]
        train_index = [i for i, x in enumerate(ID_train_list) if x != idnow]

        X_train1, X_val1 = X_train.iloc[train_index], X_train.iloc[val_index]
        y_train1, y_val1 = y_train.iloc[train_index], y_train.iloc[val_index]


        #normalize train set
        scaler = StandardScaler()
        scaler.fit(X_train1)
        X_trainNorm1 = scaler.transform(X_train1.values)
        X_valNorm1 = scaler.transform(X_val1.values)

        #TODO: optimize the penaly weight
        #https://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LogisticRegressionCV.html
        logReg = SGDClassifier(loss='log', penalty='elasticnet')
        logReg.fit(X_trainNorm1, y_train1)
        y_hat_prob = logReg.predict_proba(X_valNorm1)
        classes =logReg.classes_
        pred_probs = pred_probs.append(pd.DataFrame({'ID': ID_train[val_index], str(classes[0]): y_hat_prob[:,0],
                                                     str(classes[1]): y_hat_prob[:,1]}),ignore_index=True, sort=False)    

    return pred_probs

def predict_class(prob_dry,prob_wet):
    if prob_dry > prob_wet :
        return 'Dry'
    else:
        return 'Wet'

#get probability per recording
def get_predClass_per_audio(pred_probs, label_dict):

    pred_probs_perID = pred_probs.groupby('ID').aggregate('mean').reset_index()

    pred_probs_perID['pred_class'] = pred_probs_perID.apply(lambda x: predict_class(x['Dry'], x['Wet']), axis=1)

    #add actual classes
    pred_probs_perID['label'] = pred_probs_perID["ID"].map(label_dict)
    return pred_probs_perID