def fit(self, X, y, test_data=None):
# create tmp dir to hold data and model (especially the latter)
tmp_dir = tempfile.mkdtemp()
issparse = sps.issparse(X)
f_format = "svm" if issparse else "csv"
train_filepath = os.path.abspath("{}/X.{}".format(tmp_dir, f_format))
io_utils.dump_data(X, y, train_filepath, issparse)
if test_data:
valid = []
for i, (x_test, y_test) in enumerate(test_data):
test_filepath = os.path.abspath("{}/X{}_test.{}".format(
tmp_dir, i, f_format))
valid.append(test_filepath)
io_utils.dump_data(x_test, y_test, test_filepath, issparse)
self.param['valid'] = ",".join(valid)
self.param['task'] = 'train'
self.param['data'] = train_filepath
self.param['output_model'] = os.path.join(tmp_dir,
"LightGBM_model.txt")
calls = ["{}={}\n".format(k, self.param[k]) for k in self.param]
if self.config == "":
conf_filepath = os.path.join(tmp_dir, "train.conf")
with open(conf_filepath, 'w') as f:
f.writelines(calls)
process = subprocess.Popen(
[self.exec_path, "config={}".format(conf_filepath)],
stdout=subprocess.PIPE,
bufsize=1)
else:
process = subprocess.Popen(
[self.exec_path, "config={}".format(self.config)],
stdout=subprocess.PIPE,
bufsize=1)
with process.stdout:
for line in iter(process.stdout.readline, b''):
print(line.strip().decode('utf-8')) if self.verbose else None
# wait for the subprocess to exit
process.wait()
with open(self.param['output_model'], mode='r') as file:
self.model = file.read()
shutil.rmtree(tmp_dir)
if test_data and self.param['early_stopping_round'] > 0:
self.best_round = max(
map(int, re.findall("Tree=(\d+)", self.model))) + 1
def predict_proba(self, X):
tmp_dir = tempfile.mkdtemp()
issparse = sps.issparse(X)
f_format = "svm" if issparse else "csv"
predict_filepath = os.path.abspath(
os.path.join(tmp_dir, "X_to_pred.{}".format(f_format)))
output_model = os.path.abspath(os.path.join(tmp_dir, "model"))
conf_filepath = os.path.join(tmp_dir, "predict.conf")
output_results = os.path.abspath(
os.path.join(tmp_dir, "LightGBM_predict_result.txt"))
with open(output_model, mode="w") as file:
file.write(self.model)
io_utils.dump_data(X, np.zeros(X.shape[0]), predict_filepath, issparse)
calls = [
"task = predict\n", "data = {}\n".format(predict_filepath),
"input_model = {}\n".format(output_model),
"output_result={}\n".format(output_results)
]
with open(conf_filepath, 'w') as f:
f.writelines(calls)
process = subprocess.Popen(
[self.exec_path, "config={}".format(conf_filepath)],
stdout=subprocess.PIPE)
if self.verbose:
while process.poll() is None:
#line = process.stdout.readline()
line = process.communicate()[0] # Martin Kersner
print(line.strip().decode('utf-8'))
else:
process.communicate()
raw_probabilities = np.loadtxt(output_results, dtype=float)
if self.param['application'] == 'multiclass':
y_prob = raw_probabilities
elif self.param['application'] == 'binary':
probability_of_one = raw_probabilities
probability_of_zero = 1 - probability_of_one
y_prob = np.transpose(
np.vstack((probability_of_zero, probability_of_one)))
else:
raise
shutil.rmtree(tmp_dir)
return y_prob
def predict(self, X):
tmp_dir = tempfile.mkdtemp()
issparse = sps.issparse(X)
f_format = "svm" if issparse else "csv"
predict_filepath = os.path.abspath(
os.path.join(tmp_dir, "X_to_pred.{}".format(f_format)))
output_model = os.path.abspath(os.path.join(tmp_dir, "model"))
output_results = os.path.abspath(
os.path.join(tmp_dir, "LightGBM_predict_result.txt"))
conf_filepath = os.path.join(tmp_dir, "predict.conf")
with open(output_model, mode="w") as file:
file.write(self.model)
io_utils.dump_data(X, np.zeros(X.shape[0]), predict_filepath, issparse)
calls = [
"task = predict\n", "data = {}\n".format(predict_filepath),
"input_model = {}\n".format(output_model),
"output_result={}\n".format(output_results)
]
with open(conf_filepath, 'w') as f:
f.writelines(calls)
process = subprocess.Popen(
[self.exec_path, "config={}".format(conf_filepath)],
stdout=subprocess.PIPE)
if self.verbose:
while process.poll() is None:
line = process.stdout.readline()
print(line.strip().decode('utf-8'))
else:
process.communicate()
y_pred = np.loadtxt(output_results, dtype=float)
shutil.rmtree(tmp_dir)
return y_pred