def train(search=True):
# read data
x_train, y_train, feature_names = util.read_train_data()
#todo correct
x_train[np.isinf(x_train)] = 1
y_train[np.isinf(y_train)] = 1
np.nan_to_num(x_train)
np.nan_to_num(y_train)
x_train = x_train.astype(np.float32)
y_train = y_train.astype(np.float32)
# set model
if search:
model = find_best_model()
else:
model = XGBRegressor()
# train
timer = Timer()
model.fit(x_train, y_train)
timer.end()
# save
save_model(model)
return model
def train(search=True):
# read data
x_train, y_train, feature_names = util.read_train_data()
# model
if search:
model = find_best_model()
else:
model = RandomForestRegressor()
# train
timer = Timer()
model.fit(x_train, y_train)
timer.end()
if search:
print("\nBest parameters:\n")
for param in model.best_params_:
print("{}: {}".format(param, model.best_params_[param]))
print()
# save
save_model(model)
return model
def train():
# read data
x_train, y_train, feature_names = util.read_train_data()
# model
model = BayesianRidge()
# train
timer = Timer()
model.fit(x_train, y_train)
timer.end()
# save
save_model(model)
return model
def train():
# read data
x_train, y_train, feature_names = util.read_train_data()
# model
model = LogisticRegression()
# train
timer = Timer()
model.fit(x_train, y_train)
timer.end()
# save
save_model(model)
return model
def train(neurons=[10],
epochs=50,
retrain=None,
optimizer="adadelta",
train_path=settings.DATASET_TRAIN_PATH,
test_path=settings.DATASET_TEST_PATH):
# read data
print(train_path)
x_train, y_train, feature_names = util.read_train_data(path=train_path)
x_test, y_test, feature_names = util.read_test_data(path=test_path)
# get model
if retrain is None:
# get model
input_size = x_train.shape[1]
if len(y_train.shape) == 1:
output_size = 1
else:
output_size = y_train.shape[1]
model = get_model(input_size,
output_size,
neurons=neurons,
optimizer=optimizer)
else:
model = retrain
# train
timer = Timer()
history = model.fit(x_train,
y_train,
validation_data=(x_test, y_test),
epochs=epochs,
batch_size=12)
timer.end()
# graphs
#visualization.accuracy__graph(history.history)
#visualization.model_loss_graph(history.history)
# save model
save_model(model)
return model
def test_models(init, final, sep=1, epochs=10):
# read data
x_train, y_train, _ = util.read_train_data()
x_test, y_test, _ = util.read_test_data()
historys = []
for i in range(init, final + 1, sep):
print("\nNeurons {} - start".format(i))
model, neurons, history = _test_model([i], epochs, x_train, y_train,
x_test, y_test)
historys.append((neurons, history))
save_model(model, sufix="_" + str(i))
print("Neurons {} - done".format(i))
pickle.dump(historys, open(HISTORYS_FILE, 'wb'))
def train(max_depth=None, max_leaf_nodes=None, search=True):
"""
:param max_depth:
:param max_leaf_nodes:
:return:
"""
# read data
x_train, y_train, feature_names = util.read_train_data()
# set model
if search:
model = find_best_model()
else:
model = DecisionTreeRegressor(random_state=0,
max_depth=max_depth,
max_leaf_nodes=max_leaf_nodes)
# train
timer = Timer()
model.fit(x_train, y_train)
timer.end()
if search:
#print("\nAll Results:\n")
#results = pd.DataFrame(model.cv_results_)
#dskc_terminal.markdown_table(results)
print("\nBest parameters:\n")
for param in model.best_params_:
print("{}: {}".format(param, model.best_params_[param]))
print()
# save graph
# _save_tree_graph(model, feature_names)
save_model(model)
return model