ff_example.py

import tensorflow as tf
import numpy as np
from sklearn.linear_model import LinearRegression
from sklearn.metrics import r2_score, mean_squared_error


NUM_LAYERS = np.random.randint(1, 11)
INPUT_DIM = np.random.randint(1, 129)
OUTPUT_DIM = np.random.randint(1, 129)


def random_model():
    inp = tf.keras.Input(shape=(INPUT_DIM,))
    for layer in range(NUM_LAYERS):
        if layer == 0:
            x = tf.keras.layers.Dense(
                np.random.randint(1, 1025),
                bias_initializer='glorot_uniform'
            )(inp)
        else:
            x = tf.keras.layers.Dense(
                np.random.randint(1, 1025),
                bias_initializer='glorot_uniform'
            )(x)

    out = tf.keras.layers.Dense(
        OUTPUT_DIM,
        bias_initializer='glorot_uniform'
    )(x)

    return tf.keras.Model(inputs=inp, outputs=out)


if __name__ == '__main__':
    model = random_model()
    model.summary()

    X = np.random.uniform(size=(1024, INPUT_DIM))
    y = model(X)

    linear_model = LinearRegression().fit(X, y)

    X_test = np.random.uniform(size=(1024, INPUT_DIM))
    y_true = model(X_test)
    y_pred = linear_model.predict(X_test)

    print(f'r2 score: {r2_score(y_true, y_pred)}')
    print(f'mse: {mean_squared_error(y_true, y_pred)}')