def search_space(self):
return {
"model":
"MTNet",
"lr":
0.001,
"batch_size":
16,
"epochs":
1,
"cnn_dropout":
0.2,
"rnn_dropout":
0.2,
"time_step":
hp.choice([3, 4]),
"cnn_height":
2,
"long_num":
hp.choice([3, 4]),
"ar_size":
hp.choice([2, 3]),
"past_seq_len":
hp.sample_from(lambda spec:
(spec.config.long_num + 1) * spec.config.time_step),
}
def search_space(self, all_available_features):
return {
"selected_features":
hp.sample_from(lambda spec: json.dumps(
list(
np.random.choice(all_available_features,
size=np.random.randint(
low=3,
high=len(all_available_features)),
replace=False)))),
"model":
"MTNet",
"lr":
self.lr,
"batch_size":
self.batch_size,
"epochs":
self.epochs,
"cnn_dropout":
self.cnn_dropout,
"rnn_dropout":
self.rnn_dropout,
"time_step":
self.time_step,
"long_num":
self.long_num,
"ar_size":
self.ar_size,
"past_seq_len":
self.past_seq_len,
"cnn_hid_size":
self.cnn_hid_size,
"cnn_height":
self.cnn_height
}
def search_space(self, all_available_features):
return {
# -------- feature related parameters
"selected_features":
hp.sample_from(lambda spec: json.dumps(
list(
np.random.choice(all_available_features,
size=np.random.randint(
low=3,
high=len(all_available_features) + 1),
replace=False)))),
"model":
"Seq2Seq",
"latent_dim":
self.latent_dim,
"dropout":
self.dropout_config,
# ----------- optimization parameters
"lr":
self.lr,
"batch_size":
self.batch_size,
"epochs":
self.epochs,
"past_seq_len":
self.past_seq_config,
}
def search_space(self, all_available_features):
return {
# -------- feature related parameters
"selected_features":
hp.sample_from(lambda spec: json.dumps(
list(
np.random.choice(all_available_features,
size=np.random.randint(
low=3,
high=len(all_available_features) + 1),
replace=False)))),
"model":
"LSTM",
# --------- Vanilla LSTM model parameters
"lstm_1_units":
self.lstm_1_units_config,
"dropout_1":
0.2,
"lstm_2_units":
self.lstm_2_units_config,
"dropout_2":
self.dropout_2_config,
# ----------- optimization parameters
"lr":
self.lr,
"batch_size":
self.batch_size,
"epochs":
self.epochs,
"past_seq_len":
self.past_seq_config,
}
def _prepare_tune_config(self, space):
tune_config = {}
for k, v in space.items():
if isinstance(v, RandomSample):
tune_config[k] = hp.sample_from(v.func)
elif isinstance(v, GridSearch):
tune_config[k] = hp.grid_search(v.values)
else:
tune_config[k] = v
return tune_config
def __init__(self,
num_rand_samples=1,
epochs=5,
training_iteration=10,
time_step=[3, 4],
long_num=[3, 4],
cnn_height=[2, 3],
cnn_hid_size=[32, 50, 100],
ar_size=[2, 3],
batch_size=[32, 64]):
"""
__init__()
Constructor.
:param num_rand_samples: number of hyper-param configurations sampled randomly
:param training_iteration: no. of iterations for training (n epochs) in trials
:param epochs: no. of epochs to train in each iteration
:param time_step: random search candidates for model param "time_step"
:param long_num: random search candidates for model param "long_num"
:param ar_size: random search candidates for model param "ar_size"
:param batch_size: grid search candidates for batch size
:param cnn_height: random search candidates for model param "cnn_height"
:param cnn_hid_size: random search candidates for model param "cnn_hid_size"
"""
super(self.__class__, self).__init__()
# -- run time params
self.num_samples = num_rand_samples
self.training_iteration = training_iteration
# -- optimization params
self.lr = hp.uniform(0.001, 0.01)
self.batch_size = hp.grid_search(batch_size)
self.epochs = epochs
# ---- model params
self.cnn_dropout = hp.uniform(0.2, 0.5)
self.rnn_dropout = hp.uniform(0.2, 0.5)
self.time_step = hp.choice(time_step)
self.long_num = hp.choice(long_num, )
self.cnn_height = hp.choice(cnn_height)
self.cnn_hid_size = hp.choice(cnn_hid_size)
self.ar_size = hp.choice(ar_size)
self.past_seq_len = hp.sample_from(
lambda spec: (spec.config.long_num + 1) * spec.config.time_step)
def search_space(self, all_available_features):
return {
# -------- feature related parameters
"selected_features":
hp.sample_from(lambda spec: json.dumps(
list(
np.random.choice(all_available_features,
size=np.random.randint(
low=3,
high=len(all_available_features)),
replace=False)))),
# -------- model selection TODO add MTNet
"model":
hp.choice(["LSTM", "Seq2seq"]),
# --------- Vanilla LSTM model parameters
"lstm_1_units":
hp.grid_search([16, 32]),
"dropout_1":
0.2,
"lstm_2_units":
hp.grid_search([16, 32]),
"dropout_2":
hp.uniform(0.2, 0.5),
# ----------- Seq2Seq model parameters
"latent_dim":
hp.grid_search([32, 64]),
"dropout":
hp.uniform(0.2, 0.5),
# ----------- optimization parameters
"lr":
hp.uniform(0.001, 0.01),
"batch_size":
hp.choice([32, 64]),
"epochs":
self.epochs,
"past_seq_len":
self.past_seq_config,
}
def test_data_creator(self):
train_data_creator, val_data_creator = get_data_creators()
auto_xgb_reg = AutoXGBRegressor(cpus_per_trial=2,
name="auto_xgb_regressor",
tree_method='hist')
model_search_space = get_xgb_search_space()
# todo: change to hp.choice_n
search_space = {
"features":
hp.sample_from(
lambda spec: np.random.choice(["f1", "f2", "f3"], size=2))
}
search_space.update(model_search_space)
auto_xgb_reg.fit(data=train_data_creator,
epochs=1,
validation_data=val_data_creator,
metric="logloss",
metric_mode="min",
search_space=search_space,
n_sampling=2)
best_config = auto_xgb_reg.get_best_config()
assert all(k in best_config.keys() for k in search_space.keys())
assert len(best_config["features"]) == 2
