junte/ML
1
0
Fork 0
mirror of https://github.com/Hestia-Homes/ML.git synced 2026-06-08 11:17:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

add infernce limit

Browse source
This commit is contained in:
Michael Duong 2023-11-27 21:50:08 +00:00
parent ed4c1aebf6
commit 639ba9dd11
11 changed files with 8 additions and 105 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
deployment/handlers/prediction_app.py
View file

@ -69,9 +69,7 @@ def handler(event, context):
storage_filepath = f"s3://{PREDICTIONS_BUCKET}/{portfolio_id}/{property_id}/{created_at}.parquet"
logger.info("-------------------------")
logger.info(f"--- Initiate MLModel ---")
logger.info("-------------------------")
build_model_params = settings.build_model
client_params = settings.client
@ -80,17 +78,13 @@ def handler(event, context):
model = model_factory(build_model_params["model_type"])
logger.info("----------------------------")
logger.info(f"--- Initiate Input DataClient ---")
logger.info("----------------------------")
input_dataclient = dataclient_factory(
dataclient_type="aws-s3",
dataclient_config=client_params["aws-s3"],
)
logger.info("----------------------------")
logger.info(f"--- Initiate Output DataClient ---")
logger.info("----------------------------")
output_dataclient = dataclient_factory(
dataclient_type="aws-s3",
dataclient_config=client_params["aws-s3"],

10
modules/ml-pipeline/src/pipeline/0_startup_cleanup.py
View file

@ -16,13 +16,9 @@ def run_cleanup(artefacts_directory: str, metrics_directory: str) -> None:
Remove the directory where artefacts are stored
"""
logger.info("---------------------")
logger.info(f"--- Run Clean up ---")
logger.info("---------------------")
logger.info("-------------------------")
logger.info(f"--- Delete artefacts ---")
logger.info("-------------------------")
artefact_directory_path = Path(artefacts_directory)
@ -31,9 +27,7 @@ def run_cleanup(artefacts_directory: str, metrics_directory: str) -> None:
logger.info(f"Removing the directory: {artefacts_directory}")
shutil.rmtree(artefact_directory_path)
logger.info("-----------------------")
logger.info(f"--- Delete metrics ---")
logger.info("-----------------------")
metrics_directory_path = Path(metrics_directory)
@ -45,15 +39,11 @@ def run_cleanup(artefacts_directory: str, metrics_directory: str) -> None:
if __name__ == "__main__":
logger.info("----------------------------")
logger.info(f"--- {__file__} - Start! ---")
logger.info("----------------------------")
run_cleanup(
artefacts_directory=startup_cleanup_params["artefacts"],
metrics_directory=startup_cleanup_params["metrics"],
)
logger.info("-------------------------------")
logger.info(f"--- {__file__} - Complete! ---")
logger.info("-------------------------------")

20
modules/ml-pipeline/src/pipeline/1_prepare_data.py
View file

@ -17,9 +17,7 @@ from core.DataClient import dataclient_factory
from core.FeatureProcessor import feature_processor_factory
from config import settings
logger.info("----------------------------")
logger.info(f"--- Initiate Parameters ---")
logger.info("----------------------------")
RUNTIME_ENVIRONMENT = os.environ.get("RUNTIME_ENVIRONMENT", "local")
@ -33,9 +31,7 @@ output_train_filepath = prepare_data_params["output_train_filepath"]
output_test_filepath = prepare_data_params["output_test_filepath"]
feature_processor_config = feature_process_params["feature_processor_config"]
logger.info("----------------------------")
logger.info(f"--- Initiate DataClient ---")
logger.info("----------------------------")
input_dataclient_type = prepare_data_params["input_dataclient_type"]
output_dataclient_type = prepare_data_params["output_dataclient_type"]
@ -49,9 +45,7 @@ output_dataclient = dataclient_factory(
dataclient_config=client_params[output_dataclient_type],
)
logger.info("----------------------------------")
logger.info(f"--- Initiate FeatureProcessor ---")
logger.info("----------------------------------")
feature_processor = feature_processor_factory(
feature_process_params["feature_processor_type"]
@ -76,15 +70,11 @@ def prepare_data(
:param pipeline_mode: bool, Default False, this caches out the file for experimentation, objects returned in pipeline mode
"""
logger.info("--------------------")
logger.info("--- Loading data ---")
logger.info("--------------------")
data = input_dataclient.load_data(location=data_filepath, load_config={})
logger.info("--------------------------")
logger.info("--- Feature Processing ---")
logger.info("--------------------------")
data = feature_processor.feature_process(
data,
@ -93,9 +83,7 @@ def prepare_data(
new_feature_funcs=new_feature_funcs,
)
logger.info("----------------------")
logger.info("--- Splitting data ---")
logger.info("----------------------")
if train_proportion == 1:
train = data
@ -108,9 +96,7 @@ def prepare_data(
train = train.reset_index(drop=True)
logger.info("-----------------------")
logger.info("--- Outputting data ---")
logger.info("-----------------------")
output_dataclient.save_data(
obj=train, location=output_train_filepath, save_config=None
@ -126,13 +112,9 @@ def prepare_data(
if __name__ == "__main__":
logger.info("----------------------------")
logger.info(f"--- {__file__} - Start! ---")
logger.info("----------------------------")
logger.info("---------------------------")
logger.info(f"--- Prepare Data Stage ---")
logger.info("---------------------------")
prepare_data(
input_dataclient=input_dataclient,
@ -147,6 +129,4 @@ if __name__ == "__main__":
new_feature_funcs=new_feature_funcs,
)
logger.info("-------------------------------")
logger.info(f"--- {__file__} - Complete! ---")
logger.info("-------------------------------")

26
modules/ml-pipeline/src/pipeline/2_build_model.py
View file

@ -18,9 +18,7 @@ from core.MLMetrics import metrics_factory
from configs.post_prediction_logic import post_prediction_logic
from config import settings
logger.info("----------------------------")
logger.info(f"--- Initiate Parameters ---")
logger.info("----------------------------")
RUNTIME_ENVIRONMENT = os.environ.get("RUNTIME_ENVIRONMENT", "local")
@ -40,22 +38,16 @@ train_filepath = prepare_data_params["output_train_filepath"]
test_filepath = prepare_data_params["output_test_filepath"]
fit_metrics_filepath = build_model_params["fit_metrics_filepath"]
logger.info("----------------------------")
logger.info(f"--- Initiate DataClient ---")
logger.info("----------------------------")
# Output of previous prepare data step, will be where the data is
dataclient = dataclient_factory(prepare_data_params["output_dataclient_type"])
logger.info("-------------------------")
logger.info(f"--- Initiate MLModel ---")
logger.info("-------------------------")
model = model_factory(model_type)
logger.info("-------------------------")
logger.info(f"--- Initiate Metrics ---")
logger.info("-------------------------")
metrics = metrics_factory(generate_metrics_params["metrics_type"])
@ -75,9 +67,7 @@ def build_model(
test_data: Union[pd.DataFrame, None] = None,
pipeline_mode: bool = False,
):
logger.info("--------------------------------------")
logger.info("--- Loading Data for build process ---")
logger.info("--------------------------------------")
if train_data is None:
if train_filepath is None:
@ -89,9 +79,7 @@ def build_model(
raise ValueError(f"Need {test_filepath} if no data supplied")
test_data = dataclient.load_data(location=test_filepath, load_config=None)
logger.info("----------------------")
logger.info("--- Training model ---")
logger.info("----------------------")
model.train_model(
data=train_data.drop(columns=identifier_columns),
@ -99,32 +87,24 @@ def build_model(
model_hyperparameters=model_hyperparameters,
)
logger.info("----------------------------------")
logger.info("--- Generating fit predictions ---")
logger.info("----------------------------------")
fit_predictions = model.predict(
data=train_data, post_prediction_logic=post_prediction_logic
)
logger.info("------------------------------")
logger.info("--- Generating fit metrics ---")
logger.info("------------------------------")
metrics_output = metrics.generate_metrics(
target=train_data[target],
predictions=pd.Series(fit_predictions),
)
logger.info("--------------------")
logger.info("--- Saving model ---")
logger.info("--------------------")
model.save_model(path=Path(model_save_location))
logger.info("--------------------------")
logger.info("--- Saving fit metrics ---")
logger.info("--------------------------")
dataclient.save_data(
obj=metrics_output, location=fit_metrics_filepath, save_config=None
@ -133,13 +113,9 @@ def build_model(
if __name__ == "__main__":
logger.info("----------------------------")
logger.info(f"--- {__file__} - Start! ---")
logger.info("----------------------------")
logger.info("--------------------------")
logger.info(f"--- Build Model Stage ---")
logger.info("--------------------------")
build_model(
dataclient=dataclient,
@ -154,6 +130,4 @@ if __name__ == "__main__":
fit_metrics_filepath=fit_metrics_filepath,
)
logger.info("-------------------------------")
logger.info(f"--- {__file__} - Complete! ---")
logger.info("-------------------------------")

12
modules/ml-pipeline/src/pipeline/3_generate_predictions.py
View file

@ -10,9 +10,7 @@ from core.Logger import logger
from config import settings
from generate_predictions import generate_predictions
logger.info("----------------------------")
logger.info(f"--- Initiate Parameters ---")
logger.info("----------------------------")
RUNTIME_ENVIRONMENT = os.environ.get("RUNTIME_ENVIRONMENT", "local")
@ -33,15 +31,11 @@ model_filepath = build_model_params["model_save_filepath"]
predictions_output_filepath = generate_predictions_params["predictions_output_filepath"]
predictions_column_name = generate_predictions_params["predictions_column_name"]
logger.info("-------------------------")
logger.info(f"--- Initiate MLModel ---")
logger.info("-------------------------")
model = model_factory(build_model_params["model_type"])
logger.info("----------------------------")
logger.info(f"--- Initiate DataClient ---")
logger.info("----------------------------")
# We may have different locations of loading hence why we use one specified in generate_predictions.yaml
# I.e. for metric runs, this will be a local data client
@ -59,13 +53,9 @@ output_dataclient = dataclient_factory(
if __name__ == "__main__":
logger.info("----------------------------")
logger.info(f"--- {__file__} - Start! ---")
logger.info("----------------------------")
logger.info("----------------------------------")
logger.info(f"--- Generate Predictions Stage---")
logger.info("----------------------------------")
generate_predictions(
input_dataclient=input_dataclient,
@ -78,6 +68,4 @@ if __name__ == "__main__":
predictions_column_name=predictions_column_name,
)
logger.info("-------------------------------")
logger.info(f"--- {__file__} - Complete! ---")
logger.info("-------------------------------")

22
modules/ml-pipeline/src/pipeline/4_generate_metrics.py
View file

@ -16,9 +16,7 @@ from core.MLMetrics import metrics_factory
from core.Logger import logger
from config import settings
logger.info("----------------------------")
logger.info(f"--- Initiate Parameters ---")
logger.info("----------------------------")
RUNTIME_ENVIRONMENT = os.environ.get("RUNTIME_ENVIRONMENT", "local")
@ -36,15 +34,11 @@ predictions_column_name = generate_predictions_params["predictions_column_name"]
metrics_output_filepath = generate_metrics_params["metrics_output_filepath"]
logger.info("-------------------------")
logger.info(f"--- Initiate MLModel ---")
logger.info("-------------------------")
model = model_factory(build_model_params["model_type"])
logger.info("----------------------------")
logger.info(f"--- Initiate DataClient ---")
logger.info("----------------------------")
# Use data client for input and output, as we use dvc to cache later to the cloud
dataclient_type = generate_metrics_params["dataclient_type"]
@ -53,9 +47,7 @@ dataclient = dataclient_factory(
dataclient_config=client_params[dataclient_type],
)
logger.info("---------------------------")
logger.info(f"--- Initiate MLMetrics ---")
logger.info("---------------------------")
metrics = metrics_factory(generate_metrics_params["metrics_type"])
@ -75,34 +67,26 @@ def generate_metrics(
For a given model, we generate prediction and evaluate this against the true target
"""
logger.info("-------------------------")
logger.info("--- Loading test data ---")
logger.info("-------------------------")
test_data = input_dataclient.load_data(
location=test_data_filepath, load_config=None
)
logger.info("---------------------------")
logger.info("--- Loading predictions ---")
logger.info("---------------------------")
predictions = input_dataclient.load_data(
location=predictions_output_filepath, load_config=None
)
logger.info("--------------------------")
logger.info("--- Generating metrics ---")
logger.info("--------------------------")
metrics_output = metrics.generate_metrics(
target=test_data[target],
predictions=pd.Series(predictions[predictions_column_name]),
)
logger.info("----------------------")
logger.info("--- Saving metrics ---")
logger.info("----------------------")
output_dataclient.save_data(
obj=metrics_output, location=metrics_output_filepath, save_config=None
@ -111,13 +95,9 @@ def generate_metrics(
if __name__ == "__main__":
logger.info("----------------------------")
logger.info(f"--- {__file__} - Start! ---")
logger.info("----------------------------")
logger.info("------------------------------")
logger.info(f"--- Generate Metrics Stage---")
logger.info("------------------------------")
generate_metrics(
input_dataclient=dataclient,
@ -131,6 +111,4 @@ if __name__ == "__main__":
metrics_output_filepath=metrics_output_filepath,
)
logger.info("-------------------------------")
logger.info(f"--- {__file__} - Complete! ---")
logger.info("-------------------------------")

2
modules/ml-pipeline/src/pipeline/configs/build_model.yaml
View file

@ -16,3 +16,5 @@ default:
time_limit: 4000
presets: medium_quality
excluded_model_types: ['KNN', 'RF']
infer_limit: 0.05
infer_limit_batch_size: 10000

2
modules/ml-pipeline/src/pipeline/configs/settings.yaml
View file

@ -21,7 +21,7 @@ default:
# data_filepath: s3://retrofit-data-dev/sap_change_model/dataset_with_differencing.parquet
# data_filepath: s3://retrofit-data-dev/sap_change_model/floor_area_clean_test.parquet
# data_filepath: s3://retrofit-data-dev/sap_change_model/dataset_without_differencing.parquet
data_filepath: s3://retrofit-data-dev/sap_change_model/dataset.parquet
data_filepath: s3://retrofit-data-dev/sap_change_model/dataset_test.parquet
train_proportion: 0.9
output_train_filepath: ./data/prepared_data/train.parquet
output_test_filepath: ./data/prepared_data/test.parquet

1
modules/ml-pipeline/src/pipeline/core/Logger.py
View file

@ -21,6 +21,7 @@ def setup_logger():
# Add the stream handler to the logger
logger.addHandler(stream_handler)
logger.propagate = False
return logger

4
modules/ml-pipeline/src/pipeline/core/MLModels.py
View file

@ -149,6 +149,8 @@ class AutogluonAutoML:
"time_limit",
"presets",
"excluded_model_types",
"infer_limit",
"infer_limit_batch_size",
]
def load_model(self, path: Union[Path, str]) -> None:
@ -203,6 +205,8 @@ class AutogluonAutoML:
time_limit=model_hyperparameters["time_limit"],
presets=model_hyperparameters["presets"],
excluded_model_types=model_hyperparameters["excluded_model_types"],
infer_limit=model_hyperparameters["infer_limit"],
infer_limit_batch_size=model_hyperparameters["infer_limit_batch_size"],
)
def predict(

8
modules/ml-pipeline/src/pipeline/generate_predictions.py
View file

@ -20,23 +20,17 @@ def generate_predictions(
For a given model, we generate prediction and evaluate this against the true target
"""
logger.info("-------------------------")
logger.info("--- Loading test data ---")
logger.info("-------------------------")
test_data = input_dataclient.load_data(
location=test_data_filepath, load_config=None
)
logger.info("---------------------")
logger.info("--- Loading model ---")
logger.info("---------------------")
model.load_model(model_filepath)
logger.info("------------------------------")
logger.info("--- Generating predictions ---")
logger.info("------------------------------")
prediction_data = (
test_data.drop(columns=target) if target in test_data.columns else test_data
@ -46,9 +40,7 @@ def generate_predictions(
data=prediction_data, post_prediction_logic=post_prediction_logic
)
logger.info("--------------------------")
logger.info("--- Saving predictions ---")
logger.info("--------------------------")
predictions_df = pd.DataFrame(predictions)
predictions_df.columns = [predictions_column_name]