def get_hill_climbing_params(self, current_models):
# second, hill climbing
for _ in range(self._hill_climbing_steps):
# get models orderer by loss
# TODO: refactor this callbacks.callbacks[0]
scores = [m.get_final_loss() for m in current_models]
model_types = [m.get_type() for m in current_models]
df_models = pd.DataFrame({
"model": current_models,
"score": scores,
"model_type": model_types
})
# do group by for debug reason
df_models = df_models.groupby("model_type").apply(
lambda x: x.sort_values("score"))
unique_model_types = np.unique(df_models.model_type)
for m_type in unique_model_types:
if m_type in [
"Baseline",
"Decision Tree",
"Linear",
"Nearest Neighbors",
]:
# dont tune Baseline and Decision Tree
continue
models = df_models[df_models.model_type == m_type]["model"]
for i in range(min(self._top_models_to_improve, len(models))):
m = models[i]
for p in HillClimbing.get(
m.params.get("learner"),
self._ml_task,
len(current_models) + self._seed,
):
model_indices = [
int(m.get_name().split("_")[0])
for m in current_models
]
model_max_index = np.max(model_indices)
logger.info(
"Hill climbing step, for model #{0}".format(
model_max_index + 1))
if p is not None:
all_params = copy.deepcopy(m.params)
all_params["learner"] = p
all_params["name"] = self.get_model_name(
all_params["learner"]["model_type"],
model_max_index + 1)
unique_params_key = MljarTuner.get_params_key(
all_params)
if unique_params_key not in self._unique_params_keys:
self._unique_params_keys += [unique_params_key]
yield all_params
def get_hill_climbing_params(self, current_models):
# second, hill climbing
for _ in range(self._hill_climbing_steps):
# get models orderer by loss
# TODO: refactor this callbacks.callbacks[0]
models = sorted(
[(m.callbacks.callbacks[0].final_loss, m)
for m in current_models],
key=lambda x: x[0],
)
for i in range(min(self._top_models_to_improve, len(models))):
m = models[i][1]
for p in HillClimbing.get(
m.params.get("learner"),
self._ml_task,
len(current_models) + self._seed,
):
logger.info("Hill climbing step, for model #{0}".format(
len(current_models) + 1))
if p is not None:
all_params = copy.deepcopy(m.params)
all_params["learner"] = p
all_params["name"] = f"model_{len(current_models) + 1}"
unique_params_key = MljarTuner.get_params_key(
all_params)
if unique_params_key not in self._unique_params_keys:
self._unique_params_keys += [unique_params_key]
yield all_params
def hill_climbing_step(self, X, y):
for hill_climbing in range(self._hill_climbing_steps):
# get models orderer by loss
models = []
for m in self._models:
models += [(m.callbacks.callbacks[0].final_loss, m)]
models = sorted(models, key=lambda x: x[0])
for i in range(min(self._top_models_to_improve, len(models))):
m = models[i][1]
for p in HillClimbing.get(m.params.get("learner")):
if p is not None:
all_params = copy.deepcopy(m.params)
all_params["learner"] = p
new_model = self.train_model(all_params, X, y)
if new_model is not None:
self._models += [new_model]
self.verbose_print(
"Learner {} final loss {} time {}".format(
new_model.get_name(),
new_model.get_final_loss(),
new_model.get_train_time(),
)
)
self.log_train_time(
new_model.get_name(), new_model.get_train_time()
)
def get_hill_climbing_params(self, current_models):
df_models, algorithms = self.df_models_algorithms(current_models)
generated_params = []
counts = {model_type: 0 for model_type in algorithms}
for i in range(df_models.shape[0]):
model_type = df_models["model_type"].iloc[i]
counts[model_type] += 1
if counts[model_type] > self._top_models_to_improve:
continue
m = df_models["model"].iloc[i]
for p in HillClimbing.get(
m.params.get("learner"), self._ml_task, len(current_models) + self._seed
):
model_indices = [
int(m.get_name().split("_")[0]) for m in current_models
]
model_max_index = np.max(model_indices)
logger.info(
"Hill climbing step, for model #{0}".format(model_max_index + 1)
)
if p is not None:
all_params = copy.deepcopy(m.params)
all_params["learner"] = p
all_params["name"] = self.get_model_name(
all_params["learner"]["model_type"],
model_max_index + 1 + len(generated_params),
)
if "golden_features" in all_params["preprocessing"]:
all_params["name"] += "_GoldenFeatures"
if "drop_features" in all_params["preprocessing"] and len(
all_params["preprocessing"]["drop_features"]
):
all_params["name"] += "_SelectedFeatures"
all_params["status"] = "initialized"
all_params["final_loss"] = None
all_params["train_time"] = None
unique_params_key = MljarTuner.get_params_key(all_params)
if unique_params_key not in self._unique_params_keys:
generated_params += [all_params]
return generated_params
def hill_climbing_step(self, X, y):
for hill_climbing in range(self._hill_climbing_steps):
# get models orderer by loss
models = []
for m in self._models:
models += [(m.callbacks.callbacks[0].final_loss, m)]
models = sorted(models, key=lambda x: x[0])
for i in range(min(self._top_models_to_improve, len(models))):
m = models[i][1]
for p in HillClimbing.get(
m.params.get("learner"), len(self._models) + self._seed
):
if p is not None:
all_params = copy.deepcopy(m.params)
all_params["learner"] = p
new_model = self.train_model(all_params, X, y)
self.keep_model(new_model)
else:
self._progress_bar.update(1)
def get_hill_climbing_params(self, current_models):
# second, hill climbing
# for _ in range(self._hill_climbing_steps):
# just do one step
# get models orderer by loss
# TODO: refactor this callbacks.callbacks[0]
scores = [m.get_final_loss() for m in current_models]
model_types = [m.get_type() for m in current_models]
df_models = pd.DataFrame(
{"model": current_models, "score": scores, "model_type": model_types}
)
# do group by for debug reason
df_models = df_models.groupby("model_type").apply(
lambda x: x.sort_values("score")
)
unique_model_types = np.unique(df_models.model_type)
generated_params = []
for m_type in unique_model_types:
if m_type in ["Baseline", "Decision Tree", "Linear", "Nearest Neighbors"]:
# dont tune Baseline and Decision Tree
continue
models = df_models[df_models.model_type == m_type]["model"]
for i in range(min(self._top_models_to_improve, len(models))):
m = models[i]
for p in HillClimbing.get(
m.params.get("learner"),
self._ml_task,
len(current_models) + self._seed,
):
model_indices = [
int(m.get_name().split("_")[0]) for m in current_models
]
model_max_index = np.max(model_indices)
logger.info(
"Hill climbing step, for model #{0}".format(model_max_index + 1)
)
if p is not None:
all_params = copy.deepcopy(m.params)
all_params["learner"] = p
all_params["name"] = self.get_model_name(
all_params["learner"]["model_type"],
model_max_index + 1 + len(generated_params),
)
if "golden_features" in all_params["preprocessing"]:
all_params["name"] += "_GoldenFeatures"
if "drop_features" in all_params["preprocessing"] and len(
all_params["preprocessing"]["drop_features"]
):
all_params["name"] += "_SelectedFeatures"
all_params["status"] = "initialized"
all_params["final_loss"] = None
all_params["train_time"] = None
unique_params_key = MljarTuner.get_params_key(all_params)
if unique_params_key not in self._unique_params_keys:
self._unique_params_keys += [unique_params_key]
generated_params += [all_params]
return generated_params