def main():
# Get train,test dataset and hyperparameters from file.
train, test = create_data(path_to_file, DATA)
hyperparameters = parse_configuration_file()
# Clearing out previous file.
with open("error.txt", "w"):
pass
# Perform training.
start = time.time()
ensemble = gbrt(train, hyperparameters, test)
training_time = time.time() - start
# Get training statistics.
train_rmse = ensemble.compute_dataset_rmse(train.get_dataframe_copy())
test_rmse = ensemble.compute_dataset_rmse(test.get_dataframe_copy())
# Write hyperparameters, statistics and training time to file.
with open("outcome.txt", "w") as outcome_file:
outcome_file.write("HyperParameters:\n" + "=" * 20 + "\n")
outcome_file.write("{}".format(str(hyperparameters)))
outcome_file.write("\n\nErrors:\n" + "=" * 20)
outcome_file.write("\nTrain error = {}".format(train_rmse))
outcome_file.write("\nTest error = {}".format(test_rmse))
outcome_file.write("\n\nRunningTime:\n" + "=" * 20 + "\n")
outcome_file.write("{}".format(training_time))
def iterate_sample_parameter(train_set, test_set):
hyperparameters = GBRTHyperparameters(*DEFAULT_HYPERPARAMETERS,)
results = {}
results["time"] = []
for sampling_portion in SAMPLING_OPTIONS:
errors = []
hyperparameters.sampling_portion = sampling_portion
start = time.time()
gbrt(train_set, hyperparameters, test_set, errors)
training_time = time.time() - start
results[sampling_portion] = errors[:]
results["time"].append((sampling_portion, training_time))
with open('sampling_deliverable.pickle', 'wb') as handle:
pickle.dump(results, handle, protocol=pickle.HIGHEST_PROTOCOL)
def iterate_threshold_parameter(train_set, test_set):
hyperparameters = GBRTHyperparameters(*DEFAULT_HYPERPARAMETERS,)
results = {}
results["time"] = []
for threshold in THRESHOLD_OPTIONS:
errors = []
hyperparameters.num_threshold = threshold
start = time.time()
gbrt(train_set, hyperparameters, test_set, errors)
training_time = time.time() - start
results[threshold] = errors[:]
results["time"].append((threshold, training_time))
with open('threshold_deliverable.pickle', 'wb') as handle:
pickle.dump(results, handle, protocol=pickle.HIGHEST_PROTOCOL)
def test_compute_mse(self):
x1 = np.arange(12)
y = [1] * 3 + [3] * 3 + [16] * 3 + [20] * 3
df = pd.DataFrame()
df["x1"] = x1
df["y"] = y
dataset = TrainingDataset(df, "y")
hyperparameters = GBRTHyperparameters(1, 3, 4, 1, 1, 0)
ensemble = gbrt(dataset, hyperparameters)
self.assertEqual(ensemble.compute_dataset_mse(df, 1), 2.5)
def iterate_depth_parameter(train_set, test_set):
hyperparameters = GBRTHyperparameters(*DEFAULT_HYPERPARAMETERS,)
results = {}
for depth in DEPTH_OPTIONS:
errors = []
hyperparameters.max_depth = depth
tree = gbrt(train_set, hyperparameters, test_set, errors)
results[depth] = errors[:]
with open('depth_deliverable.pickle', 'wb') as handle:
pickle.dump(results, handle, protocol=pickle.HIGHEST_PROTOCOL)
def nottest_gbrt_residual(self):
x1 = np.arange(12)
y = [1] * 3 + [3] * 3 + [10] * 3 + [20] * 3
df = pd.DataFrame()
df["x1"] = x1
df["y"] = y
dataset = TrainingDataset(df, "SalePrice")
ensemble = gbrt(dataset, 1, 3, 4)
self.assertEqual(ensemble.evaluate(df.iloc[0], 1), 2)
self.assertEqual(ensemble.evaluate(df.iloc[9], 1), 15)
def test_gbrt_residual(self):
x1 = np.arange(6)
y = [0, 2, 100, 102, 106, 109.5]
df = pd.DataFrame()
df["x1"] = x1
df["y"] = y
hyperparameters = GBRTHyperparameters(*DEFAULT_HYPERPARAMETERS, )
dataset = TrainingDataset(df, "SalePrice")
ensemble = gbrt(dataset, hyperparameters)
self.assertEqual(ensemble.evaluate(df.iloc[0], 1), 2)
self.assertEqual(ensemble.evaluate(df.iloc[9], 1), 15)
def test_real_data(self):
train, test = create_data("../../data/")
hyperparameters = GBRTHyperparameters(0, 2, 3, 1, 1, 0)
ensemble = gbrt(train, hyperparameters, test)
get_features_importance(ensemble)