def save(self, filepath: str, model_name: str):
return Util.dump(self.model, filepath + model_name + ".pkl")
def run_blocks(input_df: pd.DataFrame,
blocks: List,
y=None,
preprocess_block=None,
logger=None,
filepath: str = "./",
task: str = "train",
save_feature: bool = False) -> pd.DataFrame:
"""
Args:
input_df (pd.DataFrame): original DataFrame
blocks (List): function block
y (_type_, optional): _description_. Defaults to None.
preprocess_block (_type_, optional): if need preprocessing for example fillna, you need set function of preporcessing
logger (_type_, optional): if is not None, output log fie
filepath (str, optional): output feature block as pkl. Defaults to "./".
task (str, optional): _description_. Defaults to "train".
save_feature; create feature as pkl. default=False
Returns:
pd.DataFrame: feature engined feature
"""
out_df = pd.DataFrame()
if preprocess_block is not None:
input_df = preprocess_block(input_df)
_input_df = input_df.copy()
if save_feature and not os.path.isdir(filepath + "features/"):
os.makedirs(filepath + "features")
print(decorate(f"start create block for {task}"))
with Timer(logger=logger, prefix=f'create {task} block'):
for block in blocks:
if save_feature:
if hasattr(block, "cols"):
if hasattr(block, "name") and hasattr(
block, "n_components"):
file_name = os.path.join(
filepath + "features/",
f"{task}_{block.__class__.__name__}_{str(block.cols)}_{str(block.name)}_{str(block.n_components)}.pkl"
)
else:
file_name = os.path.join(
filepath + "features/",
f"{task}_{block.__class__.__name__}_{str(block.cols)}.pkl"
)
else:
file_name = os.path.join(
filepath + "features/",
f"{task}_{block.__class__.__name__}.pkl")
with Timer(logger=logger, prefix='\t- {}'.format(str(block))):
if save_feature and os.path.isfile(file_name):
out_i = Util.load(file_name)
else:
if task == "train":
out_i = block.fit(_input_df)
if save_feature:
Util.dump(out_i, file_name)
else:
out_i = block.transform(_input_df)
if save_feature:
Util.dump(out_i, file_name)
assert len(input_df) == len(out_i), block
name = block.__class__.__name__
out_df = pd.concat([out_df, out_i.add_suffix(f'@{name}')], axis=1)
return out_df
