simple_neural_network_mlp_sklearn_regression_regulation

import os
import ipdb
import numpy as np
import matplotlib.pyplot as plt
from sklearn.neural_network import MLPRegressor
from sklearn.metrics import accuracy_score
from sklearn.preprocessing import OneHotEncoder, MinMaxScaler, StandardScaler
from sklearn.metrics import mean_squared_error
from sklearn import datasets
from sklearn.model_selection import train_test_split

data = datasets.load_diabetes()

X = data['data']
Y = data['target']

#ipdb.set_trace()

# use and the scaled input the encoded output

scalerX = MinMaxScaler(feature_range=(-1,1))
scalerY = MinMaxScaler(feature_range=(-1,1))

X_scaled = scalerX.fit_transform(X)
Y_scaled = scalerX.fit_transform(Y.reshape(-1,1))

#ipdb.set_trace()
X = X_scaled
Y = Y_scaled #.reshape(1,-1)[0]
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.3, random_state=0)


clf = MLPRegressor(solver='sgd', hidden_layer_sizes=(5,3),
                    activation='tanh',
                    learning_rate='constant',
                    alpha=0.001,   # regularization, try 0, 0.01, 0.001, and see total weights
                    learning_rate_init=0.1,
                    n_iter_no_change=100, # default is 10
                    tol=0.001,
                    max_iter=1000,
                    batch_size=1,
                    momentum=0,
                    random_state=0)

clf.fit(X_train, Y_train)
predictedY_train = clf.predict(X_train)
mean_square_error_train = mean_squared_error(Y_train, predictedY_train)
#similar a clf.score(X,Y)

predictedY_test = clf.predict(X_test)
mean_square_error_test = mean_squared_error(Y_test, predictedY_test)

print("Mean_Square_Error for train set is: %f" %mean_square_error_train)
print("Mean_Square_Error for test set is: %f" %mean_square_error_test)

# visualizar os pesos
print("Coefficients: %s" %clf.coefs_)


# Somar todos os coeficientes:
total_weights = 0
for layer_weights in clf.coefs_:
    total_weights += np.sum(np.abs(layer_weights))
print("Total Weights: %f" %total_weights)

# visualizar o número de iterações rodadas
print("Number of iterations: %d" %clf.n_iter_)

#for i,x in enumerate(X):
#    print("Input: %s \t Output: %s" %(x,clf.predict([x])[0]))

plt.plot(Y_train,label="Y for train set")
plt.plot(predictedY_train,label="predicted Y for train set")
plt.show()

plt.plot(Y_test,label="Y for test set")
plt.plot(predictedY_test,label="predicted Y for test set")
plt.show()