import numpy as np

from sklearn.linear_model import LogisticRegression

from keras.datasets import fashion_mnist
from sklearn.metrics import accuracy_score
from sklearn.metrics import confusion_matrix

import seaborn as sns;

(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()
x_train = x_train[len(x_train)-1000:]
y_train = y_train[len(y_train)-1000:]

x_test = x_test[len(x_test)-1000:]
y_test = y_test[len(y_test)-1000:]

images_train =  []
for image_train in x_train:
    images_train.append(image_train.flatten())

images_test = []

for image_test in x_test:
    images_test.append(image_test.flatten())

images_train = np.array(images_train)
images_test = np.array(images_test)

print(len(x_test))
print("Starting grid search\n")
from sklearn.model_selection import GridSearchCV
from sklearn.neural_network import MLPClassifier

parameters = {'solver': ['sgd'], 'max_iter': [1000], \
              'hidden_layer_sizes': [(15,10,5), (30, 20, 10,)], \
              'random_state': [1], 'tol': [1e-7], 'batch_size': [20], 'shuffle': [True, False], \
              'momentum': [0.95], 'activation': ['logistic'], 'alpha': [1e-5]}

grid_search = GridSearchCV(MLPClassifier(), parameters, n_jobs=-1, verbose=1, cv=2)
grid_search.fit(images_train, y_train)

print(grid_search.score(images_test, y_test))
print(grid_search.best_params_)

print("Starting MLPClassifier\n")


neural_network = MLPClassifier(hidden_layer_sizes=(30,20,10),random_state=1)

neural_network.fit(images_train, y_train)
#neural_network.fit(preprocessing.StandardScaler().fit_transform(images_train), y_train)

conf_matrix_neural_network = confusion_matrix(y_test, \
                                              neural_network.predict(images_test))

#print("Confusion_matrix:")
#print(conf_matrix_neural_network)

sns.heatmap(conf_matrix_neural_network)

acc = accuracy_score(y_test, neural_network.predict(images_test))
print("Neural network model accuracy is {0:0.2f}".format(acc))