def predict_model(train_x,
train_y,
test_x,
test_y=None,
class_weights=None,
random_state=0):
# Learn the KNN model
params = {
'layers': [[1000, 0.90], [100, 0.60]],
'loss': 'categorical_crossentropy'
}
model = fp.PMC_NeuralNetwork_T1(train_x.shape[1],
train_y.shape[1],
params,
bags=1)
model.fit(train_x, train_y, test_x, test_y, batch_size=64, nb_epoch=20)
# Predict on the test instances
test_predicted = model.predict_proba(test_x)
counter = 1
if test_y is not None:
print("...score = {:.5f}".format(
fp.brier_score(test_y, test_predicted / counter, class_weights)))
# Add more epochs
for k in range(4):
model.fit(train_x, train_y, test_x, test_y, batch_size=128)
test_predicted = test_predicted + model.predict_proba(test_x)
counter += 1
if test_y is not None:
print("...score = {:.5f}".format(
fp.brier_score(test_y, test_predicted / counter,
class_weights)))
test_predicted = test_predicted / counter
return (test_predicted)
'L2_XGB_vD1_0', 'L2_XGB_vB2_1', 'L2_NN_vD1_0', 'L2_XGB_vC2_0'
]
# Set weights
weights = np.ones(len(files_txt))
if validate:
validated = np.zeros((16124, 20))
for i, file_txt in enumerate(files_txt):
validated = validated + fd.load_L1_train([file_txt]) * weights[i]
validated = validated / np.sum(weights)
all_train_y = fd.load_train_y([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
all_train_y = all_train_y[np.isfinite(all_train_y.sum(1))]
print("Final Submission local-CV score: {}".format(
fp.brier_score(all_train_y, validated, class_weights)))
# Predict L3 test data
predicted = np.zeros((16600, 20))
for i, file_txt in enumerate(files_txt):
predicted = predicted + fd.load_submissions([file_txt]) * weights[i]
predicted = predicted / np.sum(weights)
name_to_save = 'L3_WA_vD2'
# Save files
directory = '../final_submission/'
if not os.path.exists(directory):
os.makedirs(directory)
# Submission
submission = pd.concat((pd.DataFrame(rows), pd.DataFrame(predicted)), axis=1)
