elasticnet_train.seq

import sys
import python

pydef get_all_training_files_in_folder(directory) -> list[tuple[str, str]]:
    import os
    files = []
    for item in os.listdir(directory):
        item_path = os.path.join(directory, item)
        if os.path.isfile(item_path):
            if item.endswith(".csv"):
                    files.append((item_path, item_path[:-4] + ".labels"))
    return files

pydef training(directory, train_files, learning_rate, l1, l2):
    import os
    import numpy as np
    y_list = []
    X_list = []
    for csv_f, lbl_f in train_files:
        with open(lbl_f) as lbl_file:
            for line in lbl_file:
                age, Fage = line.rstrip().split(',')
                y_list.append(Fage)
        with open(csv_f) as csv_file:
            X = np.genfromtxt(csv_file, dtype=float, delimiter=',')
            X_list.append(X)
    X_train = np.concatenate(X_list)
    y_train = np.array(y_list, dtype=np.float)
    #Dimensions of the normalized data
    N = X_train.shape[0]
    D = X_train.shape[1]
    #Initialize the gradient descent and learning rate
    beta = np.random.randn(D) / np.sqrt(D)
    #Standard scaling
    X_train = (X_train - np.mean(X_train, axis=0)) / np.std(X_train, axis=0)
    for i in range(10000):
        Yhat = X_train.dot(beta)
        delta = Yhat - y_train
        beta = beta - (learning_rate *(X_train.T.dot(delta) + l1*np.sign(beta) + l2*2*beta))
        mse = delta.dot(delta)/N
        print(mse)
    intercept = np.sum(y_train - np.dot(X_train, beta))/N
    np.save(os.path.join(directory, "enet_betas.npy"), beta)
    np.save(os.path.join(directory, "enet_intercept.npy"), intercept)
    #print(str(sum(beta != 0)) + " Variables and Neglected " +  str(sum(beta == 0)) + " Variables")

input_folder = sys.argv[1]
save_folder = sys.argv[2]
learning_rate = 0.0000001
#L1 regularization term alpha
l1 = 0.02255706
#L2 regularization term lambda
l2 = 0.5

train_files = get_all_training_files_in_folder(input_folder)
training(train_files, learning_rate, l1, l2)