def _hp_search(self,
input_df,
validation_df,
metric,
recipe,
mc,
resources_per_trial,
remote_dir):
ft = TimeSequenceFeatureTransformer(self.future_seq_len,
self.dt_col,
self.target_col,
self.extra_features_col,
self.drop_missing)
if isinstance(input_df, list):
feature_list = ft.get_feature_list(input_df[0])
else:
feature_list = ft.get_feature_list(input_df)
# model = VanillaLSTM(check_optional_config=False)
model = TimeSequenceModel(check_optional_config=False, future_seq_len=self.future_seq_len)
# prepare parameters for search engine
search_space = recipe.search_space(feature_list)
runtime_params = recipe.runtime_params()
num_samples = runtime_params['num_samples']
stop = dict(runtime_params)
search_algorithm_params = recipe.search_algorithm_params()
search_algorithm = recipe.search_algorithm()
fixed_params = recipe.fixed_params()
del stop['num_samples']
searcher = RayTuneSearchEngine(logs_dir=self.logs_dir,
resources_per_trial=resources_per_trial,
name=self.name,
remote_dir=remote_dir,
)
searcher.compile(input_df,
search_space=search_space,
stop=stop,
search_algorithm_params=search_algorithm_params,
search_algorithm=search_algorithm,
fixed_params=fixed_params,
# feature_transformers=TimeSequenceFeatures,
feature_transformers=ft,
# model=model,
future_seq_len=self.future_seq_len,
validation_df=validation_df,
metric=metric,
mc=mc,
num_samples=num_samples)
# searcher.test_run()
searcher.run()
best = searcher.get_best_trials(k=1)[0] # get the best one trial, later could be n
pipeline = self._make_pipeline(best,
feature_transformers=ft,
model=model,
remote_dir=remote_dir)
return pipeline
def setup_method(self, method):
# super().setup_method(method)
train_data = pd.DataFrame(data=np.random.randn(64, 4))
val_data = pd.DataFrame(data=np.random.randn(16, 4))
test_data = pd.DataFrame(data=np.random.randn(16, 4))
future_seq_len = 1
past_seq_len = 6
# use roll method in time_sequence
tsft = TimeSequenceFeatureTransformer()
self.x_train, self.y_train = tsft._roll_train(
train_data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
self.x_val, self.y_val = tsft._roll_train(
val_data, past_seq_len=past_seq_len, future_seq_len=future_seq_len)
self.x_test = tsft._roll_test(test_data, past_seq_len=past_seq_len)
self.config = {
'epochs': 1,
"lr": 0.001,
"lstm_1_units": 16,
"dropout_1": 0.2,
"lstm_2_units": 10,
"dropout_2": 0.2,
"batch_size": 32,
}
self.model = VanillaLSTM(check_optional_config=False,
future_seq_len=future_seq_len)
def test_predict(self):
train_data = pd.DataFrame(data=np.random.randn(64, 4))
test_data = pd.DataFrame(data=np.random.randn(16, 4))
future_seq_len = 1
past_seq_len = 6
# use roll method in time_sequence
tsft = TimeSequenceFeatureTransformer()
x_train, y_train = tsft._roll_train(train_data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
x_test = tsft._roll_test(test_data, past_seq_len=past_seq_len)
config = {
'epochs': 2,
"lr": 0.001,
"lstm_1_units": 16,
"dropout_1": 0.2,
"lstm_2_units": 10,
"dropout_2": 0.2,
"batch_size": 32,
}
model = VanillaLSTM(check_optional_config=False,
future_seq_len=future_seq_len)
model.fit_eval(x_train, y_train, **config)
y_pred = model.predict(x_test)
assert y_pred.shape == (x_test.shape[0], 1)
def test_evaluate(self):
train_data = pd.DataFrame(data=np.random.randn(64, 4))
val_data = pd.DataFrame(data=np.random.randn(16, 4))
future_seq_len = 1
past_seq_len = 6
# use roll method in time_sequence
tsft = TimeSequenceFeatureTransformer()
x_train, y_train = tsft._roll_train(train_data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
x_val, y_val = tsft._roll_train(val_data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
config = {
'epochs': 1,
"lr": 0.001,
"lstm_1_units": 16,
"dropout_1": 0.2,
"lstm_2_units": 10,
"dropout_2": 0.2,
"batch_size": 32,
}
model = VanillaLSTM(check_optional_config=False,
future_seq_len=future_seq_len)
model.fit_eval(x_train, y_train, **config)
print("evaluate:", model.evaluate(x_val, y_val))
def setup_method(self, method):
tf.keras.backend.clear_session()
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
self.model = MTNetKeras()
self.config = {"long_num": self.long_num,
"time_step": self.time_step,
"ar_window": np.random.randint(1, 3),
"cnn_height": np.random.randint(1, 3),
"epochs": 1}
class TestZouwuModelMTNetForecaster(TestCase):
def setUp(self):
tf.keras.backend.clear_session()
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
def tearDown(self):
pass
def create_data(self):
def gen_train_sample(data, past_seq_len, future_seq_len):
data = pd.DataFrame(data)
x, y = self.ft._roll_train(data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
return x, y
def gen_test_sample(data, past_seq_len):
test_data = pd.DataFrame(data)
x = self.ft._roll_test(test_data, past_seq_len=past_seq_len)
return x
self.long_num = 6
self.time_step = 2
look_back = (self.long_num + 1) * self.time_step
look_forward = 1
self.x_train, self.y_train = gen_train_sample(
data=np.random.randn(64, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_val, self.y_val = gen_train_sample(data=np.random.randn(16, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_test = gen_test_sample(data=np.random.randn(16, 4),
past_seq_len=look_back)
def test_forecast_mtnet(self):
# TODO hacking to fix a bug
target_dim = 1
model = MTNetForecaster(target_dim=target_dim,
feature_dim=self.x_train.shape[-1],
long_series_num=self.long_num,
series_length=self.time_step)
x_train_long, x_train_short = model.preprocess_input(self.x_train)
x_val_long, x_val_short = model.preprocess_input(self.x_val)
x_test_long, x_test_short = model.preprocess_input(self.x_test)
model.fit([x_train_long, x_train_short],
self.y_train,
validation_data=([x_val_long, x_val_short], self.y_val),
batch_size=32,
distributed=False)
assert model.evaluate([x_val_long, x_val_short], self.y_val)
predict_result = model.predict([x_test_long, x_test_short])
assert predict_result.shape == (self.x_test.shape[0], target_dim)
def _hp_search(self, input_df, validation_df, metric, recipe, mc,
resources_per_trial, remote_dir):
ft = TimeSequenceFeatureTransformer(self.future_seq_len, self.dt_col,
self.target_col,
self.extra_features_col,
self.drop_missing)
if isinstance(input_df, list):
feature_list = ft.get_feature_list(input_df[0])
else:
feature_list = ft.get_feature_list(input_df)
def model_create_func():
# model = VanillaLSTM(check_optional_config=False)
_model = TimeSequenceModel(check_optional_config=False,
future_seq_len=self.future_seq_len)
return _model
model = model_create_func()
# prepare parameters for search engine
search_space = recipe.search_space(feature_list)
metric_mode = TimeSequencePredictor._get_metric_mode(metric)
searcher = RayTuneSearchEngine(
logs_dir=self.logs_dir,
resources_per_trial=resources_per_trial,
name=self.name,
remote_dir=remote_dir,
)
searcher.compile(
input_df,
model_create_func=model_create_func(),
search_space=search_space,
recipe=recipe,
feature_transformers=ft,
future_seq_len=self.future_seq_len,
validation_df=validation_df,
metric=metric,
metric_mode=metric_mode,
mc=mc,
)
# searcher.test_run()
analysis = searcher.run()
pipeline = self._make_pipeline(analysis,
metric_mode,
feature_transformers=ft,
model=model,
remote_dir=remote_dir)
return pipeline
def fit_with_fixed_configs(self,
input_df,
validation_df=None,
mc=False,
**user_configs):
"""
Fit pipeline with fixed configs. The model will be trained from initialization
with the hyper-parameter specified in configs. The configs contain both identity configs
(Eg. "future_seq_len", "dt_col", "target_col", "metric") and automl tunable configs
(Eg. "past_seq_len", "batch_size").
We recommend calling get_default_configs to see the name and default values of configs you
you can specify.
:param input_df: one data frame or a list of data frames
:param validation_df: one data frame or a list of data frames
:param user_configs: you can overwrite or add more configs with user_configs. Eg. "epochs"
:return:
"""
# self._check_configs()
if self.config is None:
self.config = self.get_default_configs()
if user_configs is not None:
self.config.update(user_configs)
ft_id_config_set = {
'future_seq_len', 'dt_col', 'target_col', 'extra_features_col',
'drop_missing'
}
ft_id_configs = {a: self.config[a] for a in ft_id_config_set}
self.feature_transformers = TimeSequenceFeatureTransformer(
**ft_id_configs)
model_id_config_set = {'future_seq_len'}
ft_id_configs = {a: self.config[a] for a in model_id_config_set}
self.model = TimeSequenceModel(check_optional_config=False,
**ft_id_configs)
all_available_features = self.feature_transformers.get_feature_list(
input_df)
self.config.update({"selected_features": all_available_features})
(x_train, y_train) = self.feature_transformers.fit_transform(
input_df, **self.config)
if self._is_val_df_valid(validation_df):
validation_data = self.feature_transformers.transform(
validation_df)
else:
validation_data = None
self.model.fit_eval(x_train,
y_train,
validation_data=validation_data,
mc=mc,
verbose=1,
**self.config)
def __init__(self, feature_transformers=None, model=None, config=None):
"""
initialize a pipeline
:param model: the internal model
:param feature_transformers: the feature transformers
"""
if feature_transformers is None:
assert model is None and config is None
self.feature_transformers = TimeSequenceFeatureTransformer()
self.model = VanillaLSTM(check_optional_config=False)
print("Initialize new time sequence pipeline.")
else:
self.feature_transformers = feature_transformers
self.model = model
self.config = config
def test_evaluate_predict_future_more_1(self):
target_col = "values"
metrics = ["mse", "r2"]
future_seq_len = np.random.randint(2, 6)
train_df, test_df, tsp, test_sample_num = self.get_input_tsp(
future_seq_len, target_col)
pipeline = tsp.fit(train_df, test_df)
mse, rs = pipeline.evaluate(test_df, metrics=metrics)
assert len(mse) == future_seq_len
assert len(rs) == future_seq_len
y_pred = pipeline.predict(test_df)
assert y_pred.shape == (test_sample_num - default_past_seq_len + 1,
future_seq_len + 1)
y_pred_df = pipeline.predict(test_df[:-future_seq_len])
columns = [
"{}_{}".format(target_col, i) for i in range(future_seq_len)
]
y_pred_value = y_pred_df[columns].values
y_df = test_df[default_past_seq_len:]
y_value = TimeSequenceFeatureTransformer()._roll_test(
y_df[target_col], future_seq_len)
mse_pred_eval, rs_pred_eval = [
Evaluator.evaluate(m, y_value, y_pred_value) for m in metrics
]
mse_eval, rs_eval = pipeline.evaluate(test_df, metrics)
assert_array_almost_equal(mse_pred_eval, mse_eval, decimal=2)
assert_array_almost_equal(rs_pred_eval, rs_eval, decimal=2)
def create_feature_transformer(self):
ft = TimeSequenceFeatureTransformer(self.future_seq_len,
self.dt_col,
self.target_col,
self.extra_features_col,
self.drop_missing)
return ft
class TestZouwuModelLSTMForecaster(TestCase):
def setUp(self):
tf.keras.backend.clear_session()
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
def tearDown(self):
pass
def create_data(self):
def gen_train_sample(data, past_seq_len, future_seq_len):
data = pd.DataFrame(data)
x, y = self.ft._roll_train(data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
return x, y
def gen_test_sample(data, past_seq_len):
test_data = pd.DataFrame(data)
x = self.ft._roll_test(test_data, past_seq_len=past_seq_len)
return x
self.long_num = 6
self.time_step = 2
look_back = (self.long_num + 1) * self.time_step
look_forward = 1
self.x_train, self.y_train = gen_train_sample(
data=np.random.randn(64, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_val, self.y_val = gen_train_sample(data=np.random.randn(16, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_test = gen_test_sample(data=np.random.randn(16, 4),
past_seq_len=look_back)
def test_forecast_lstm(self):
# TODO hacking to fix a bug
model = LSTMForecaster(target_dim=1,
feature_dim=self.x_train.shape[-1])
model.fit(self.x_train,
self.y_train,
validation_data=(self.x_val, self.y_val),
batch_size=8,
distributed=False)
model.evaluate(self.x_val, self.y_val)
model.predict(self.x_test)
def load_ts_pipeline(file):
feature_transformers = TimeSequenceFeatureTransformer()
model = TimeSequenceModel(check_optional_config=False)
all_config = restore_zip(file, feature_transformers, model)
ts_pipeline = TimeSequencePipeline(feature_transformers=feature_transformers,
model=model,
config=all_config)
print("Restore pipeline from", file)
return ts_pipeline
def test_save_restore(self):
train_data = pd.DataFrame(data=np.random.randn(64, 4))
test_data = pd.DataFrame(data=np.random.randn(16, 4))
future_seq_len = 1
past_seq_len = 6
# use roll method in time_sequence
tsft = TimeSequenceFeatureTransformer()
x_train, y_train = tsft._roll_train(train_data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
x_test = tsft._roll_test(test_data, past_seq_len=past_seq_len)
config = {
'epochs': 2,
"lr": 0.001,
"lstm_1_units": 16,
"dropout_1": 0.2,
"lstm_2_units": 10,
"dropout_2": 0.2,
"batch_size": 32,
}
dirname = tempfile.mkdtemp(prefix="automl_test_vanilla")
try:
model = VanillaLSTM(check_optional_config=False,
future_seq_len=future_seq_len)
model.fit_eval(x_train, y_train, **config)
predict_before = model.predict(x_test)
model_path = os.path.join(dirname, "testmodel.h5")
config_path = os.path.join(dirname, "local_config.json")
model.save(model_path=model_path, config_path=config_path)
local_config = load_config(config_path)
config.update(local_config)
model.restore(model_path=model_path, **config)
predict_after = model.predict(x_test)
assert np.allclose(predict_before, predict_after)
finally:
shutil.rmtree(dirname)
def setup_method(self, method):
# super().setup_method(method)
self.train_data = pd.DataFrame(data=np.random.randn(64, 4))
self.val_data = pd.DataFrame(data=np.random.randn(16, 4))
self.test_data = pd.DataFrame(data=np.random.randn(16, 4))
self.past_seq_len = 6
self.future_seq_len_1 = 1
self.future_seq_len_2 = 2
# use roll method in time_sequence
self.feat = TimeSequenceFeatureTransformer()
self.config = {'batch_size': 32, 'epochs': 1}
self.model_1 = LSTMSeq2Seq(check_optional_config=False,
future_seq_len=self.future_seq_len_1)
self.model_2 = LSTMSeq2Seq(check_optional_config=False,
future_seq_len=self.future_seq_len_2)
self.fitted = False
self.predict_1 = None
self.predict_2 = None
def test_dataframe_input_with_datetime(self):
train_df, validation_df, future_seq_len = get_ts_input()
dataframe_with_datetime = {'df': train_df, 'val_df': validation_df}
ft = TimeSequenceFeatureTransformer(future_seq_len=future_seq_len,
dt_col="datetime",
target_col="value")
searcher = prepare_searcher(
data=dataframe_with_datetime,
model_creator=LSTM_model_creator,
name='test_ray_dateframe_with_datetime_with_val',
feature_transformer=ft)
searcher.run()
best_trials = searcher.get_best_trials(k=1)
assert best_trials is not None
class TestSeq2Seq(ZooTestCase):
def setup_method(self, method):
# super().setup_method(method)
self.train_data = pd.DataFrame(data=np.random.randn(64, 4))
self.val_data = pd.DataFrame(data=np.random.randn(16, 4))
self.test_data = pd.DataFrame(data=np.random.randn(16, 4))
self.past_seq_len = 6
self.future_seq_len_1 = 1
self.future_seq_len_2 = 2
# use roll method in time_sequence
self.feat = TimeSequenceFeatureTransformer()
self.config = {'batch_size': 32, 'epochs': 1}
self.model_1 = LSTMSeq2Seq(check_optional_config=False,
future_seq_len=self.future_seq_len_1)
self.model_2 = LSTMSeq2Seq(check_optional_config=False,
future_seq_len=self.future_seq_len_2)
self.fitted = False
self.predict_1 = None
self.predict_2 = None
def teardown_method(self, method):
pass
def test_fit_eval_1(self):
x_train_1, y_train_1 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_1)
print("fit_eval_future_seq_len_1:",
self.model_1.fit_eval(x_train_1, y_train_1, **self.config))
assert self.model_1.past_seq_len == 6
assert self.model_1.feature_num == 4
assert self.model_1.future_seq_len == 1
assert self.model_1.target_col_num == 1
def test_fit_eval_2(self):
x_train_2, y_train_2 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_2)
print("fit_eval_future_seq_len_2:",
self.model_2.fit_eval(x_train_2, y_train_2, **self.config))
assert self.model_2.future_seq_len == 2
self.fitted = True
def test_evaluate_1(self):
x_train_1, y_train_1 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_1)
x_val_1, y_val_1 = self.feat._roll_train(
self.val_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_1)
self.model_1.fit_eval(x_train_1, y_train_1, **self.config)
print("evaluate_future_seq_len_1:",
self.model_1.evaluate(x_val_1, y_val_1, metric=['mse', 'r2']))
def test_evaluate_2(self):
x_train_2, y_train_2 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_2)
x_val_2, y_val_2 = self.feat._roll_train(
self.val_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_2)
self.model_2.fit_eval(x_train_2, y_train_2, **self.config)
print("evaluate_future_seq_len_2:",
self.model_2.evaluate(x_val_2, y_val_2, metric=['mse', 'r2']))
def test_predict_1(self):
x_train_1, y_train_1 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_1)
x_test_1 = self.feat._roll_test(self.test_data,
past_seq_len=self.past_seq_len)
self.model_1.fit_eval(x_train_1, y_train_1, **self.config)
predict_1 = self.model_1.predict(x_test_1)
assert predict_1.shape == (x_test_1.shape[0], self.future_seq_len_1)
def test_predict_2(self):
x_train_2, y_train_2 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_2)
x_test_2 = self.feat._roll_test(self.test_data,
past_seq_len=self.past_seq_len)
self.model_2.fit_eval(x_train_2, y_train_2, **self.config)
predict_2 = self.model_2.predict(x_test_2)
assert predict_2.shape == (x_test_2.shape[0], self.future_seq_len_2)
def test_save_restore_1(self):
x_train_1, y_train_1 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_1)
x_test_1 = self.feat._roll_test(self.test_data,
past_seq_len=self.past_seq_len)
self.model_1.fit_eval(x_train_1, y_train_1, **self.config)
predict_1_before = self.model_1.predict(x_test_1)
new_model_1 = LSTMSeq2Seq(check_optional_config=False)
dirname = tempfile.mkdtemp(prefix="automl_test_feature")
try:
save(dirname, model=self.model_1)
restore(dirname, model=new_model_1, config=self.config)
predict_1_after = new_model_1.predict(x_test_1)
assert_array_almost_equal(predict_1_before, predict_1_after, decimal=2), \
"Prediction values are not the same after restore: " \
"predict before is {}, and predict after is {}".format(predict_1_before,
predict_1_after)
new_config = {'epochs': 1}
new_model_1.fit_eval(x_train_1, y_train_1, **new_config)
finally:
shutil.rmtree(dirname)
def test_save_restore_2(self):
x_train_2, y_train_2 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_2)
x_test_2 = self.feat._roll_test(self.test_data,
past_seq_len=self.past_seq_len)
self.model_2.fit_eval(x_train_2, y_train_2, **self.config)
predict_2_before = self.model_2.predict(x_test_2)
new_model_2 = LSTMSeq2Seq(check_optional_config=False)
dirname = tempfile.mkdtemp(prefix="automl_test_feature")
try:
save(dirname, model=self.model_2)
restore(dirname, model=new_model_2, config=self.config)
predict_2_after = new_model_2.predict(x_test_2)
assert_array_almost_equal(predict_2_before, predict_2_after, decimal=2), \
"Prediction values are not the same after restore: " \
"predict before is {}, and predict after is {}".format(predict_2_before,
predict_2_after)
new_config = {'epochs': 2}
new_model_2.fit_eval(x_train_2, y_train_2, **new_config)
finally:
shutil.rmtree(dirname)
def test_predict_with_uncertainty(self, ):
x_train_2, y_train_2 = self.feat._roll_train(
self.train_data,
past_seq_len=self.past_seq_len,
future_seq_len=self.future_seq_len_2)
x_test_2 = self.feat._roll_test(self.test_data,
past_seq_len=self.past_seq_len)
self.model_2.fit_eval(x_train_2, y_train_2, mc=True, **self.config)
prediction, uncertainty = self.model_2.predict_with_uncertainty(
x_test_2, n_iter=2)
assert prediction.shape == (x_test_2.shape[0], self.future_seq_len_2)
assert uncertainty.shape == (x_test_2.shape[0], self.future_seq_len_2)
assert np.any(uncertainty)
new_model_2 = LSTMSeq2Seq(check_optional_config=False)
dirname = tempfile.mkdtemp(prefix="automl_test_feature")
try:
save(dirname, model=self.model_2)
restore(dirname, model=new_model_2, config=self.config)
prediction, uncertainty = new_model_2.predict_with_uncertainty(
x_test_2, n_iter=2)
assert prediction.shape == (x_test_2.shape[0],
self.future_seq_len_2)
assert uncertainty.shape == (x_test_2.shape[0],
self.future_seq_len_2)
assert np.any(uncertainty)
finally:
shutil.rmtree(dirname)
class TestMTNetKeras(ZooTestCase):
def setup_method(self, method):
tf.keras.backend.clear_session()
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
self.model = MTNetKeras()
self.config = {"long_num": self.long_num,
"time_step": self.time_step,
"ar_window": np.random.randint(1, 3),
"cnn_height": np.random.randint(1, 3),
"epochs": 1}
def teardown_method(self, method):
pass
def create_data(self):
def gen_train_sample(data, past_seq_len, future_seq_len):
data = pd.DataFrame(data)
x, y = self.ft._roll_train(data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len
)
return x, y
def gen_test_sample(data, past_seq_len):
test_data = pd.DataFrame(data)
x = self.ft._roll_test(test_data, past_seq_len=past_seq_len)
return x
self.long_num = 6
self.time_step = 2
look_back = (self.long_num + 1) * self.time_step
look_forward = 1
self.x_train, self.y_train = gen_train_sample(data=np.random.randn(
64, 4), past_seq_len=look_back, future_seq_len=look_forward)
self.x_val, self.y_val = gen_train_sample(data=np.random.randn(16, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_test = gen_test_sample(data=np.random.randn(16, 4),
past_seq_len=look_back)
def test_fit_evaluate(self):
self.model.fit_eval(self.x_train, self.y_train,
validation_data=(self.x_val, self.y_val),
**self.config)
self.model.evaluate(self.x_val, self.y_val)
def test_save_restore(self):
self.model.fit_eval(self.x_train, self.y_train,
validation_data=(self.x_val, self.y_val),
**self.config)
y_pred = self.model.predict(self.x_test)
assert y_pred.shape == (self.x_test.shape[0], self.y_train.shape[1])
dirname = "tmp"
restored_model = MTNetKeras()
try:
save(dirname, model=self.model)
restore(dirname, model=restored_model, config=self.config)
predict_after = restored_model.predict(self.x_test)
assert_array_almost_equal(y_pred, predict_after, decimal=2), \
"Prediction values are not the same after restore: " \
"predict before is {}, and predict after is {}".format(y_pred, predict_after)
restored_model.fit_eval(self.x_train, self.y_train, epochs=1)
restored_model.evaluate(self.x_val, self.y_val)
finally:
shutil.rmtree("tmp")
def test_predict_with_uncertainty(self):
self.model.fit_eval(self.x_train, self.y_train,
validation_data=(self.x_val, self.y_val),
mc=True,
**self.config)
pred, uncertainty = self.model.predict_with_uncertainty(self.x_test, n_iter=2)
assert pred.shape == (self.x_test.shape[0], self.y_train.shape[1])
assert uncertainty.shape == pred.shape
assert np.any(uncertainty)
def setUp(self):
tf.keras.backend.clear_session()
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
class TestZouwuModelForecast(ZooTestCase):
def setup_method(self, method):
tf.keras.backend.clear_session()
# super(TestZouwuModelForecast, self).setup_method(method)
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
def teardown_method(self, method):
pass
def create_data(self):
def gen_train_sample(data, past_seq_len, future_seq_len):
data = pd.DataFrame(data)
x, y = self.ft._roll_train(data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
return x, y
def gen_test_sample(data, past_seq_len):
test_data = pd.DataFrame(data)
x = self.ft._roll_test(test_data, past_seq_len=past_seq_len)
return x
self.long_num = 6
self.time_step = 2
look_back = (self.long_num + 1) * self.time_step
look_forward = 1
self.x_train, self.y_train = gen_train_sample(
data=np.random.randn(64, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_val, self.y_val = gen_train_sample(data=np.random.randn(16, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_test = gen_test_sample(data=np.random.randn(16, 4),
past_seq_len=look_back)
def test_forecast_lstm(self):
# TODO hacking to fix a bug
model = LSTMForecaster(target_dim=1,
feature_dim=self.x_train.shape[-1])
model.fit(self.x_train,
self.y_train,
validation_data=(self.x_val, self.y_val),
batch_size=8,
distributed=False)
model.evaluate(self.x_val, self.y_val)
model.predict(self.x_test)
def test_forecast_mtnet(self):
# TODO hacking to fix a bug
model = MTNetForecaster(target_dim=1,
feature_dim=self.x_train.shape[-1],
lb_long_steps=self.long_num,
lb_long_stepsize=self.time_step)
x_train_long, x_train_short = model.preprocess_input(self.x_train)
x_val_long, x_val_short = model.preprocess_input(self.x_val)
x_test_long, x_test_short = model.preprocess_input(self.x_test)
model.fit([x_train_long, x_train_short],
self.y_train,
validation_data=([x_val_long, x_val_short], self.y_val),
batch_size=32,
distributed=False)
model.evaluate([x_val_long, x_val_short], self.y_val)
model.predict([x_test_long, x_test_short])
def test_forecast_tcmf(self):
from zoo.zouwu.model.forecast import TCMFForecaster
model = TCMFForecaster(max_y_iterations=1,
init_XF_epoch=1,
max_FX_epoch=1,
max_TCN_epoch=1,
alt_iters=2)
x = np.random.rand(300, 480)
model.fit(x)
model.predict(x=None, horizon=24)
target_value = np.random.rand(300, 24)
model.evaluate(x=None, target_value=target_value, metric=['mse'])
class TimeSequencePipeline(Pipeline):
def __init__(self,
feature_transformers=None,
model=None,
config=None,
name=None):
"""
initialize a pipeline
:param model: the internal model
:param feature_transformers: the feature transformers
"""
self.feature_transformers = feature_transformers
self.model = model
self.config = config
self.name = name
self.time = time.strftime("%Y%m%d-%H%M%S")
def describe(self):
init_info = [
'future_seq_len', 'dt_col', 'target_col', 'extra_features_col',
'drop_missing'
]
print("**** Initialization info ****")
for info in init_info:
print(info + ":", self.config[info])
print("")
def fit(self, input_df, validation_df=None, mc=False, epoch_num=20):
x, y = self.feature_transformers.transform(input_df, is_train=True)
if validation_df is not None and not validation_df.empty:
validation_data = self.feature_transformers.transform(
validation_df)
else:
validation_data = None
new_config = {'epochs': epoch_num}
self.model.fit_eval(x,
y,
validation_data,
mc=mc,
verbose=1,
**new_config)
print('Fit done!')
def _is_val_df_valid(self, validation_df):
df_not_empty = isinstance(validation_df,
pd.DataFrame) and not validation_df.empty
df_list_not_empty = isinstance(validation_df, list) \
and validation_df and not all([d.empty for d in validation_df])
if validation_df is not None and (df_not_empty or df_list_not_empty):
return True
else:
return False
def _check_configs(self):
required_configs = {'future_seq_len'}
if not self.config.keys() & required_configs:
raise ValueError("Missing required parameters in configuration. " +
"Required parameters are: " +
str(required_configs))
default_config = {
'dt_col': 'datetime',
'target_col': 'value',
'extra_features_col': None,
'drop_missing': True,
'past_seq_len': 2,
'batch_size': 64,
'lr': 0.001,
'dropout': 0.2,
'epochs': 10,
'metric': 'mse'
}
for config, value in default_config.items():
if config not in self.config:
print('Config: \'{}\' is not specified. '
'A default value of {} will be used.'.format(
config, value))
def get_default_configs(self):
default_configs = {
'dt_col': 'datetime',
'target_col': 'value',
'extra_features_col': None,
'drop_missing': True,
'future_seq_len': 1,
'past_seq_len': 2,
'batch_size': 64,
'lr': 0.001,
'dropout': 0.2,
'epochs': 10,
'metric': 'mean_squared_error'
}
print("**** default config: ****")
for config in default_configs:
print(config + ":", default_configs[config])
print(
"You can change any fields in the default configs by passing into "
"fit_with_fixed_configs(). Otherwise, the default values will be used."
)
return default_configs
def fit_with_fixed_configs(self,
input_df,
validation_df=None,
mc=False,
**user_configs):
"""
Fit pipeline with fixed configs. The model will be trained from initialization
with the hyper-parameter specified in configs. The configs contain both identity configs
(Eg. "future_seq_len", "dt_col", "target_col", "metric") and automl tunable configs
(Eg. "past_seq_len", "batch_size").
We recommend calling get_default_configs to see the name and default values of configs you
you can specify.
:param input_df: one data frame or a list of data frames
:param validation_df: one data frame or a list of data frames
:param user_configs: you can overwrite or add more configs with user_configs. Eg. "epochs"
:return:
"""
# self._check_configs()
if self.config is None:
self.config = self.get_default_configs()
if user_configs is not None:
self.config.update(user_configs)
ft_id_config_set = {
'future_seq_len', 'dt_col', 'target_col', 'extra_features_col',
'drop_missing'
}
ft_id_configs = {a: self.config[a] for a in ft_id_config_set}
self.feature_transformers = TimeSequenceFeatureTransformer(
**ft_id_configs)
model_id_config_set = {'future_seq_len'}
ft_id_configs = {a: self.config[a] for a in model_id_config_set}
self.model = TimeSequenceModel(check_optional_config=False,
**ft_id_configs)
all_available_features = self.feature_transformers.get_feature_list(
input_df)
self.config.update({"selected_features": all_available_features})
(x_train, y_train) = self.feature_transformers.fit_transform(
input_df, **self.config)
if self._is_val_df_valid(validation_df):
validation_data = self.feature_transformers.transform(
validation_df)
else:
validation_data = None
self.model.fit_eval(x_train,
y_train,
validation_data=validation_data,
mc=mc,
verbose=1,
**self.config)
def evaluate(self, input_df, metrics=["mse"], multioutput='raw_values'):
"""
evaluate the pipeline
:param input_df:
:param metrics: subset of ['mean_squared_error', 'r_square', 'sMAPE']
:param multioutput: string in ['raw_values', 'uniform_average']
'raw_values' :
Returns a full set of errors in case of multioutput input.
'uniform_average' :
Errors of all outputs are averaged with uniform weight.
:return:
"""
if isinstance(metrics, str):
metrics = [metrics]
# if not isinstance(metrics, list):
# raise ValueError("Expected metrics to be a list!")
x, y = self.feature_transformers.transform(input_df, is_train=True)
y_pred = self.model.predict(x)
if y_pred.shape[1] == 1:
multioutput = 'uniform_average'
y_unscale, y_pred_unscale = self.feature_transformers.post_processing(
input_df, y_pred, is_train=True)
return [
Evaluator.evaluate(m,
y_unscale,
y_pred_unscale,
multioutput=multioutput) for m in metrics
]
def predict(self, input_df):
"""
predict test data with the pipeline fitted
:param input_df:
:return:
"""
x, _ = self.feature_transformers.transform(input_df, is_train=False)
y_pred = self.model.predict(x)
y_output = self.feature_transformers.post_processing(input_df,
y_pred,
is_train=False)
return y_output
def predict_with_uncertainty(self, input_df, n_iter=100):
x, _ = self.feature_transformers.transform(input_df, is_train=False)
y_pred, y_pred_uncertainty = self.model.predict_with_uncertainty(
x=x, n_iter=n_iter)
y_output = self.feature_transformers.post_processing(input_df,
y_pred,
is_train=False)
y_uncertainty = self.feature_transformers.unscale_uncertainty(
y_pred_uncertainty)
return y_output, y_uncertainty
def save(self, ppl_file=None):
"""
save pipeline to file, contains feature transformer, model, trial config.
:param ppl_file:
:return:
"""
ppl_file = ppl_file or os.path.join(
DEFAULT_PPL_DIR, "{}_{}.ppl".format(self.name, self.time))
save_zip(ppl_file, self.feature_transformers, self.model, self.config)
print("Pipeline is saved in", ppl_file)
return ppl_file
def config_save(self, config_file=None):
"""
save all configs to file.
:param config_file:
:return:
"""
config_file = config_file or os.path.join(
DEFAULT_CONFIG_DIR, "{}_{}.json".format(self.name, self.time))
save_config(config_file, self.config, replace=True)
return config_file
def _hp_search(self, input_df, validation_df, metric):
# features
# feature_list = ["WEEKDAY(datetime)", "HOUR(datetime)",
# "PERCENTILE(value)", "IS_WEEKEND(datetime)",
# "IS_AWAKE(datetime)", "IS_BUSY_HOURS(datetime)"
# # "DAY(datetime)","MONTH(datetime)", #probabaly not useful
# ]
# target_list = ["value"]
# ft = TimeSequenceFeatures(self.future_seq_len, self.dt_col, self.target_col, self.extra_features_col)
# ft = DummyTimeSequenceFeatures(file_path='../../../../data/nyc_taxi_rolled_split.npz')
ft = TimeSequenceFeatureTransformer(self.future_seq_len, self.dt_col,
self.target_col,
self.extra_features_col,
self.drop_missing)
feature_list = ft.get_feature_list(input_df)
# model
model = VanillaLSTM(check_optional_config=False,
future_seq_len=self.future_seq_len)
search_space = {
# -------- feature related parameters
"selected_features":
RandomSample(lambda spec: np.random.choice(
feature_list,
size=np.random.randint(low=3, high=len(feature_list), size=1),
replace=False)),
# --------- model related parameters
# 'input_shape_x': x_train.shape[1],
# 'input_shape_y': x_train.shape[-1],
'out_units':
self.future_seq_len,
"lr":
0.001,
"lstm_1_units":
GridSearch([16, 32]),
"dropout_1":
0.2,
"lstm_2_units":
10,
"dropout_2":
RandomSample(lambda spec: np.random.uniform(0.2, 0.5)),
"batch_size":
1024,
}
stop = {"reward_metric": -0.05, "training_iteration": 10}
searcher = RayTuneSearchEngine(logs_dir=self.logs_dir,
ray_num_cpus=6,
resources_per_trial={"cpu": 2})
searcher.compile(
input_df,
search_space=search_space,
stop=stop,
# feature_transformers=TimeSequenceFeatures,
feature_transformers=ft, # use dummy features for testing the rest
model=model,
validation_df=validation_df,
metric=metric)
# searcher.test_run()
trials = searcher.run()
best = searcher.get_best_trials(
k=1)[0] # get the best one trial, later could be n
pipeline = self._make_pipeline(
best,
feature_transformers=ft,
# feature_transformers=TimeSequenceFeatures(
# file_path='../../../../data/nyc_taxi_rolled_split.npz'),
model=VanillaLSTM(check_optional_config=False))
return pipeline
class TestZouwuModelForecast(ZooTestCase):
def setup_method(self, method):
tf.keras.backend.clear_session()
# super(TestZouwuModelForecast, self).setup_method(method)
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
def teardown_method(self, method):
pass
def create_data(self):
def gen_train_sample(data, past_seq_len, future_seq_len):
data = pd.DataFrame(data)
x, y = self.ft._roll_train(data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
return x, y
def gen_test_sample(data, past_seq_len):
test_data = pd.DataFrame(data)
x = self.ft._roll_test(test_data, past_seq_len=past_seq_len)
return x
self.long_num = 6
self.time_step = 2
look_back = (self.long_num + 1) * self.time_step
look_forward = 1
self.x_train, self.y_train = gen_train_sample(
data=np.random.randn(64, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_val, self.y_val = gen_train_sample(data=np.random.randn(16, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_test = gen_test_sample(data=np.random.randn(16, 4),
past_seq_len=look_back)
def test_forecast_lstm(self):
# TODO hacking to fix a bug
model = LSTMForecaster(horizon=1, feature_dim=self.x_train.shape[-1])
model.fit(self.x_train,
self.y_train,
validation_data=(self.x_val, self.y_val),
batch_size=8,
distributed=False)
model.evaluate(self.x_val, self.y_val)
model.predict(self.x_test)
def test_forecast_mtnet(self):
# TODO hacking to fix a bug
model = MTNetForecaster(horizon=1,
feature_dim=self.x_train.shape[-1],
lb_long_steps=self.long_num,
lb_long_stepsize=self.time_step)
x_train_long, x_train_short = model.preprocess_input(self.x_train)
x_val_long, x_val_short = model.preprocess_input(self.x_val)
x_test_long, x_test_short = model.preprocess_input(self.x_test)
model.fit([x_train_long, x_train_short],
self.y_train,
validation_data=([x_val_long, x_val_short], self.y_val),
batch_size=32,
distributed=False)
model.evaluate([x_val_long, x_val_short], self.y_val)
model.predict([x_test_long, x_test_short])
def setup_method(self, method):
tf.keras.backend.clear_session()
# super(TestZouwuModelForecast, self).setup_method(method)
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
class TestZouwuModelForecast(ZooTestCase):
def setup_method(self, method):
tf.keras.backend.clear_session()
# super(TestZouwuModelForecast, self).setup_method(method)
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
def teardown_method(self, method):
pass
def create_data(self):
def gen_train_sample(data, past_seq_len, future_seq_len):
data = pd.DataFrame(data)
x, y = self.ft._roll_train(data,
past_seq_len=past_seq_len,
future_seq_len=future_seq_len)
return x, y
def gen_test_sample(data, past_seq_len):
test_data = pd.DataFrame(data)
x = self.ft._roll_test(test_data, past_seq_len=past_seq_len)
return x
self.long_num = 6
self.time_step = 2
look_back = (self.long_num + 1) * self.time_step
look_forward = 1
self.x_train, self.y_train = gen_train_sample(
data=np.random.randn(64, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_val, self.y_val = gen_train_sample(data=np.random.randn(16, 4),
past_seq_len=look_back,
future_seq_len=look_forward)
self.x_test = gen_test_sample(data=np.random.randn(16, 4),
past_seq_len=look_back)
def test_forecast_lstm(self):
# TODO hacking to fix a bug
model = LSTMForecaster(target_dim=1,
feature_dim=self.x_train.shape[-1])
model.fit(self.x_train,
self.y_train,
validation_data=(self.x_val, self.y_val),
batch_size=8,
distributed=False)
model.evaluate(self.x_val, self.y_val)
model.predict(self.x_test)
def test_forecast_mtnet(self):
# TODO hacking to fix a bug
model = MTNetForecaster(target_dim=1,
feature_dim=self.x_train.shape[-1],
long_series_num=self.long_num,
series_length=self.time_step)
x_train_long, x_train_short = model.preprocess_input(self.x_train)
x_val_long, x_val_short = model.preprocess_input(self.x_val)
x_test_long, x_test_short = model.preprocess_input(self.x_test)
model.fit([x_train_long, x_train_short],
self.y_train,
validation_data=([x_val_long, x_val_short], self.y_val),
batch_size=32,
distributed=False)
model.evaluate([x_val_long, x_val_short], self.y_val)
model.predict([x_test_long, x_test_short])
def test_forecast_tcmf(self):
from zoo.zouwu.model.forecast import TCMFForecaster
import tempfile
model = TCMFForecaster(max_y_iterations=1,
init_FX_epoch=1,
max_FX_epoch=1,
max_TCN_epoch=1,
alt_iters=2)
horizon = np.random.randint(1, 50)
# construct data
id = np.arange(300)
data = np.random.rand(300, 480)
input = dict({'data': data})
with self.assertRaises(Exception) as context:
model.fit(input)
self.assertTrue("key `y` doesn't exist in x" in str(context.exception))
input = dict({'id': id, 'y': data})
with self.assertRaises(Exception) as context:
model.is_distributed()
self.assertTrue('You should run fit before calling is_distributed()' in
str(context.exception))
model.fit(input)
assert not model.is_distributed()
with self.assertRaises(Exception) as context:
model.fit(input)
self.assertTrue('This model has already been fully trained' in str(
context.exception))
with self.assertRaises(Exception) as context:
model.fit(input, incremental=True)
self.assertTrue(
'NotImplementedError' in context.exception.__class__.__name__)
with tempfile.TemporaryDirectory() as tempdirname:
model.save(tempdirname)
loaded_model = TCMFForecaster.load(tempdirname, distributed=False)
yhat = model.predict(x=None, horizon=horizon)
yhat_loaded = loaded_model.predict(x=None, horizon=horizon)
yhat_id = yhat_loaded["id"]
assert (yhat_id == id).all()
yhat = yhat["prediction"]
yhat_loaded = yhat_loaded["prediction"]
assert yhat.shape == (300, horizon)
assert (yhat == yhat_loaded).all()
target_value = np.random.rand(300, horizon)
target_value = dict({"y": target_value})
model.evaluate(x=None, target_value=target_value, metric=['mse'])
def test_forecast_tcmf_without_id(self):
from zoo.zouwu.model.forecast import TCMFForecaster
import tempfile
model = TCMFForecaster(max_y_iterations=1,
init_FX_epoch=1,
max_FX_epoch=1,
max_TCN_epoch=1,
alt_iters=2)
horizon = np.random.randint(1, 50)
# construct data
id = np.arange(200)
data = np.random.rand(300, 480)
input = dict({'y': "abc"})
with self.assertRaises(Exception) as context:
model.fit(input)
self.assertTrue(
"the value of y should be an ndarray" in str(context.exception))
input = dict({'id': id, 'y': data})
with self.assertRaises(Exception) as context:
model.fit(input)
self.assertTrue(
"the length of the id array should be equal to the number of" in
str(context.exception))
input = dict({'y': data})
model.fit(input)
assert not model.is_distributed()
with self.assertRaises(Exception) as context:
model.fit(input)
self.assertTrue('This model has already been fully trained' in str(
context.exception))
with tempfile.TemporaryDirectory() as tempdirname:
model.save(tempdirname)
loaded_model = TCMFForecaster.load(tempdirname, distributed=False)
yhat = model.predict(x=None, horizon=horizon)
yhat_loaded = loaded_model.predict(x=None, horizon=horizon)
assert "id" not in yhat_loaded
yhat = yhat["prediction"]
yhat_loaded = yhat_loaded["prediction"]
assert yhat.shape == (300, horizon)
assert (yhat == yhat_loaded).all()
target_value = np.random.rand(300, horizon)
target_value_fake = dict({"data": target_value})
with self.assertRaises(Exception) as context:
model.evaluate(x=None,
target_value=target_value_fake,
metric=['mse'])
self.assertTrue("key y doesn't exist in y" in str(context.exception))
target_value = dict({"y": target_value})
model.evaluate(x=None, target_value=target_value, metric=['mse'])
def test_forecast_tcmf_xshards(self):
from zoo.zouwu.model.forecast import TCMFForecaster
from zoo.orca import OrcaContext
import zoo.orca.data.pandas
import tempfile
OrcaContext.pandas_read_backend = "pandas"
def preprocessing(df, id_name, y_name):
id = df.index
data = df.to_numpy()
result = dict({id_name: id, y_name: data})
return result
def postprocessing(pred_results, output_dt_col_name):
id_arr = pred_results["id"]
pred_results = pred_results["prediction"]
pred_results = np.concatenate(
(np.expand_dims(id_arr, axis=1), pred_results), axis=1)
final_df = pd.DataFrame(pred_results,
columns=["id"] + output_dt_col_name)
final_df.id = final_df.id.astype("int")
final_df = final_df.set_index("id")
final_df.columns.name = "datetime"
final_df = final_df.unstack().reset_index().rename(
{0: "prediction"}, axis=1)
return final_df
def get_pred(d):
return d["prediction"]
model = TCMFForecaster(max_y_iterations=1,
init_FX_epoch=1,
max_FX_epoch=1,
max_TCN_epoch=1,
alt_iters=2)
with tempfile.NamedTemporaryFile() as temp:
data = np.random.rand(300, 480)
df = pd.DataFrame(data)
df.to_csv(temp.name)
shard = zoo.orca.data.pandas.read_csv(temp.name)
shard.cache()
shard_train = shard.transform_shard(preprocessing, 'id', 'data')
with self.assertRaises(Exception) as context:
model.fit(shard_train)
self.assertTrue("key `y` doesn't exist in x" in str(context.exception))
shard_train = shard.transform_shard(preprocessing, 'cid', 'y')
with self.assertRaises(Exception) as context:
model.fit(shard_train)
self.assertTrue(
"key `id` doesn't exist in x" in str(context.exception))
with self.assertRaises(Exception) as context:
model.is_distributed()
self.assertTrue('You should run fit before calling is_distributed()' in
str(context.exception))
shard_train = shard.transform_shard(preprocessing, 'id', 'y')
model.fit(shard_train)
assert model.is_distributed()
with self.assertRaises(Exception) as context:
model.fit(shard_train)
self.assertTrue('This model has already been fully trained' in str(
context.exception))
with self.assertRaises(Exception) as context:
model.fit(shard_train, incremental=True)
self.assertTrue(
'NotImplementedError' in context.exception.__class__.__name__)
with tempfile.TemporaryDirectory() as tempdirname:
model.save(tempdirname + "/model")
loaded_model = TCMFForecaster.load(tempdirname + "/model",
distributed=True)
horizon = np.random.randint(1, 50)
yhat_shard_origin = model.predict(x=None, horizon=horizon)
yhat_list_origin = yhat_shard_origin.collect()
yhat_list_origin = list(map(get_pred, yhat_list_origin))
yhat_shard = loaded_model.predict(x=None, horizon=horizon)
yhat_list = yhat_shard.collect()
yhat_list = list(map(get_pred, yhat_list))
yhat_origin = np.concatenate(yhat_list_origin)
yhat = np.concatenate(yhat_list)
assert yhat.shape == (300, horizon)
assert (yhat == yhat_origin).all()
output_dt_col_name = pd.date_range(start='2020-05-01',
periods=horizon,
freq='H').to_list()
yhat_df_shards = yhat_shard.transform_shard(postprocessing,
output_dt_col_name)
final_df_list = yhat_df_shards.collect()
final_df = pd.concat(final_df_list)
final_df.sort_values("datetime", inplace=True)
assert final_df.shape == (300 * horizon, 3)
OrcaContext.pandas_read_backend = "spark"
def setup_method(self, method):
self.ft = TimeSequenceFeatureTransformer()
self.create_data()
class TimeSequencePipeline(Pipeline):
def __init__(self, feature_transformers=None, model=None, config=None):
"""
initialize a pipeline
:param model: the internal model
:param feature_transformers: the feature transformers
"""
if feature_transformers is None:
assert model is None and config is None
self.feature_transformers = TimeSequenceFeatureTransformer()
self.model = VanillaLSTM(check_optional_config=False)
print("Initialize new time sequence pipeline.")
else:
self.feature_transformers = feature_transformers
self.model = model
self.config = config
def evaluate(self, input_df, metric=["mean_squared_error"]):
"""
evaluate the pipeline
:param input_df:
:param metric:
:return:
"""
x, y = self.feature_transformers.transform(input_df, is_train=True)
return self.model.evaluate(x, y, metric)
def predict(self, input_df):
# there might be no y in the data, TODO needs to fix in TimeSquenceFeatures
x = self.feature_transformers.transform(input_df, is_train=False)
y_pred = self.model.predict(x)
y_output = self.feature_transformers.post_processing(y_pred)
return y_output
def save(self, file):
"""
save pipeline to file, contains feature transformer, model, trial config.
:param file:
:return:
"""
if not os.path.isdir(file):
os.mkdir(file)
model_path = os.path.join(file, "weights_tune.h5")
config_path = os.path.join(file, "all_config.json")
self.feature_transformers.save(config_path, replace=True)
self.model.save(model_path, config_path)
# check if ** is needed
save_config(config_path, self.config)
def restore(self, file):
"""
restore pipeline from file
:param file:
:param config:
:return:
"""
model_path = os.path.join(file, "weights_tune.h5")
config_path = os.path.join(file, "all_config.json")
all_config = load_config(config_path)
self.model.restore(model_path, **all_config)
self.feature_transformers.restore(**all_config)
