def test_configuration(self, conf_dict):
params = Params.ConfigurableParams(conf_dict)
print("[GridSearch] Starting CrossValidation with id {}".format(
params.ID))
name = self.model_name + "_" + str(params.ID)
cv = KFoldCrossValidation(self.train_set,
self.train_labels,
params=params,
K=Params.NUM_FOLDS,
model_name=name)
try:
cv.perform()
cv.dump()
self.results[name] = cv.mean_accuracy
print("{}\t{}".format(name, cv.mean_accuracy))
except:
print("[RandomSearch] Problem with CrossValidation with id {}".
format(params.ID))
cv.report_error()
print("[GridSearch] Ended CrossValidation with id {}".format(
params.ID))
validation_s = MLCUP2017.cup_validation_set
validation_l = MLCUP2017.cup_validation_labels
mues = MLCUP2017.mean_per_attribute_dataset[len(train_s[0]) + 1:]
sigmas = MLCUP2017.std_per_attribute_dataset[len(train_s[0]) + 1:]
print("mues {}".format(mues))
print("sigmas {}".format(sigmas))
for model_name, params_dict in best_models_with_high_epochs:
model_name = os.path.basename(model_name)
print("MODEL {}".format(model_name))
params = Params.ConfigurableParams(params_dict)
params.MAX_EPOCH = int(params_dict["mean epochs"])
myNN = NeuralNetwork.NeuralNetwork(params)
for size in params.LAYERS_SIZE:
myNN.addLayer(size)
myNN.set_train(train_s, train_l)
myNN.set_validation(validation_s, validation_l)
myNN.learn()
Plotting.plot_loss_accuracy_per_epoch(myNN)
a = Statistics.MEELoss()
