113 lines
3.6 KiB
Python
113 lines
3.6 KiB
Python
"""
|
|
Script to train tune hyperparameters
|
|
Based on:
|
|
https://learn.microsoft.com/en-us/azure/machine-learning/how-to-train-scikit-learn
|
|
"""
|
|
from azure.ai.ml import Input, command
|
|
from azure.ai.ml.constants import AssetTypes
|
|
from azure.ai.ml.entities import Model
|
|
from azure.ai.ml.sweep import Choice
|
|
|
|
from compute_aml import create_or_load_aml
|
|
from data_tabular import create_tabular_dataset, name_dataset
|
|
from environment import create_docker_environment, custom_env_name
|
|
from initialize_constants import AML_COMPUTE_NAME
|
|
from ml_client import create_or_load_ml_client
|
|
|
|
experiment_folder = "diabetes_hyperdrive"
|
|
experiment_name = "mslearn-diabetes-hyperdrive"
|
|
script_name = "diabetes_training.py"
|
|
registered_model_name = "diabetes_model_hyper"
|
|
best_model_name = "best_diabetes_model"
|
|
|
|
|
|
def main():
|
|
# 1. Create or Load a ML client
|
|
ml_client = XXXX()
|
|
|
|
# 2. Create compute resources
|
|
XXXX()
|
|
|
|
# 3. Create and register a File Dataset
|
|
XXXX()
|
|
latest_version_dataset = XXXX()
|
|
|
|
# 4. Environment
|
|
environment_names = [env.name for XXXX in ml_client.environments.list()]
|
|
if custom_env_name not in environment_names:
|
|
create_docker_environment()
|
|
|
|
# 5. Run Job
|
|
job_for_sweep = command(
|
|
inputs=dict(
|
|
script_name=script_name,
|
|
data=Input(
|
|
type=AssetTypes.URI_FILE,
|
|
# @latest doesn't work with dataset paths
|
|
path=f"azureml:{name_dataset}:{latest_version_dataset}",
|
|
),
|
|
registered_model_name=registered_model_name,
|
|
learning_rate=XXXX(values= XXXX),
|
|
n_estimators=XXXX(values=XXXX),
|
|
),
|
|
code=experiment_folder,
|
|
command=(
|
|
"python XXXX"
|
|
+ " --data XXXX"
|
|
+ " --registered_model_name XXXX"
|
|
+ " --learning_rate XXXX"
|
|
+ " --n_estimators XXXX"
|
|
),
|
|
environment=XXXX,
|
|
compute=AML_COMPUTE_NAME,
|
|
experiment_name=experiment_name,
|
|
display_name=experiment_name,
|
|
)
|
|
|
|
# Configure hyperdrive settings
|
|
sweep_job = job_for_sweep.XXXX(
|
|
compute=AML_COMPUTE_NAME,
|
|
sampling_algorithm="grid",
|
|
primary_metric="AUC",
|
|
goal="Maximize",
|
|
max_total_trials=6,
|
|
max_concurrent_trials=2,
|
|
)
|
|
|
|
# submit the command
|
|
returned_sweep_job = ml_client.create_or_update(sweep_job)
|
|
|
|
# stream the output and wait until the job is finished
|
|
ml_client.jobs.stream(returned_sweep_job.name)
|
|
|
|
# refresh the latest status of the job after streaming
|
|
returned_sweep_job = ml_client.jobs.get(name=returned_sweep_job.name)
|
|
|
|
# Find and register the best model
|
|
if returned_sweep_job.status == "Completed":
|
|
# First let us get the run which gave us the best result
|
|
best_run = returned_sweep_job.properties["best_child_run_id"]
|
|
|
|
# lets get the model from this run
|
|
model = Model(
|
|
# the script stores the model as the given name
|
|
path=(
|
|
f"azureml://jobs/{best_run}/outputs/artifacts/paths/"
|
|
+ f"{registered_model_name}/"
|
|
),
|
|
name=best_model_name,
|
|
type="mlflow_model",
|
|
)
|
|
else:
|
|
print(
|
|
f"Sweep job status: {returned_sweep_job.status}. \
|
|
Please wait until it completes"
|
|
)
|
|
|
|
# Register best model
|
|
print(f"Registering Model {best_model_name}")
|
|
ml_client.models.XXXX(model=model)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|