'Auditorium', 'bar', 'classroom', 'closet', 'movietheater',
'restaurant', 'other'
]
no_code = False
data_train, category_train = generate_matrix(categories, 'Train', no_code)
data_test, category_test = generate_matrix(categories,
'Test',
no_code=True)
perceptron = MultilayerPerceptronClassifier( \
hidden_layer_sizes = (12,), \
max_iter = 1000, \
# random_state = 123,
algorithm = 'sgd',
batch_size = 200,
shuffle = True,
alpha = 0.00002,\
learning_rate_init = 0.0001, )
# verbose = True,)
category_predicted = perceptron.fit(data_train,
category_train).predict(data_test)
if not no_code:
category_predicted = matrix_category_decode(category_predicted)
accuracy = accuracy_score(category_test, category_predicted)
cm = confusion_matrix(category_test, category_predicted, categories)
# cm_title = 'Confusion matrix, without normalization ('+selected_set_name+', '+perceptr_name+')'
def getResults(iters, hidden_layer_size, not_coded):
categories = [
'Auditorium', 'bar', 'classroom', 'closet', 'movietheater',
'restaurant', 'other'
]
no_code = not_coded
data_train, category_train = generate_matrix(categories, 'Train', no_code)
data_test, category_test = generate_matrix(categories,
'Test',
no_code=True)
perceptron = MultilayerPerceptronClassifier( \
hidden_layer_sizes = (hidden_layer_size,), \
max_iter = iters, \
algorithm = 'sgd',
batch_size = 200,
shuffle = True,
alpha = 0.00002,\
learning_rate_init = 0.0001,
verbose=False)
perceptron_model = perceptron.fit(data_train, category_train)
category_predicted = perceptron_model.predict(data_test)
# Probamos en el set de entrenamiento para medir sobreajuste
category_predicted_train = perceptron_model.predict(data_train)
if not no_code:
category_predicted = matrix_category_decode(category_predicted)
category_predicted_train = matrix_category_decode(
category_predicted_train)
category_train = matrix_category_decode(category_train)
accuracy = accuracy_score(category_test, category_predicted)
accuracy_train = accuracy_score(category_train, category_predicted_train)
cm = confusion_matrix(category_test, category_predicted, categories)
cm_train = confusion_matrix(category_train, category_predicted_train,
categories)
cm_normalized = normalize_matrix(cm)
cm_train_normalized = normalize_matrix(cm_train)
coded_title = 'String coded' if no_code else 'List coded'
coded_name = 'no_coded' if no_code else 'coded'
outputs = perceptron.n_outputs_
cm_normalized_title_base = coded_title + ' outputs: ' + str(outputs)
cm_normalized_title = cm_normalized_title_base + ' Acc. sobre test: ' + str(
accuracy)
cm_train_normalized_title = cm_normalized_title_base + ' Acc. sobre train ' + str(
accuracy_train)
plt.figure()
plot_confusion_matrix(cm_normalized,
cm_normalized_title,
ticks=(categories))
print 'Accuracy test: ' + str(accuracy)
variable_names = str(iters) + "_" + str(hidden_layer_size) + "_" + str(
coded_name)
plt.savefig('cm_' + variable_names + '_' +
coded_title.lower().replace(" ", "_") + '_' +
str(time.time()).replace('.', '') + '.png',
bbox_inches='tight')
plt.close()
plt.figure()
plot_confusion_matrix(cm_train_normalized,
cm_train_normalized_title,
ticks=(categories))
print 'Accuracy train: ' + str(accuracy_train)
plt.savefig('cm_train_' + variable_names + '_' +
coded_title.lower().replace(" ", "_") + '_' +
str(time.time()).replace('.', '') + '.png',
bbox_inches='tight')
plt.close()
with open("results.csv", "a") as myfile:
myfile.write(
str(iters) + "," + str(hidden_layer_size) + "," + coded_name +
"," + str(accuracy) + "," + str(accuracy_train) + "\n")