SIENTIAPDE-1252: Implement artifact generation and MLflow logging for model training results
This commit introduces artifact generation and MLflow logging capabilities to the model training process. It includes the following changes: - Added methods to generate reports, save data files, and log model parameters, metrics, models, and artifacts to MLflow. - Implemented error handling for various scenarios, such as missing files, invalid data, and MLflow connection errors. - Created a new 'header.html' file for report styling and navigation. - Modified the 'model_repository.py' file to include the new artifact generation and MLflow logging methods. - Added comprehensive unit tests to ensure the functionality and robustness of the new features.
This commit is contained in:
Bruno Domingues
@@ -11,16 +11,21 @@ By Monitoring we mean the evaluation of the performance of models, the generatio
|
||||
|
||||
"""
|
||||
|
||||
import shutil
|
||||
import traceback
|
||||
from datetime import datetime
|
||||
from os import makedirs, path, remove
|
||||
|
||||
import mlflow
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
from sientia.ModelServing import ModelServing # type: ignore[import-untyped]
|
||||
from sientia.reports import Reports # type: ignore[import-untyped]
|
||||
from sientia_do.observability.logger import Logger
|
||||
from sientia_do.temporal.constants import DATETIME_FORMAT_WITH_TZ
|
||||
|
||||
from model_manager.utils.models.train_model_result import TrainModelResult
|
||||
|
||||
|
||||
class MLFlowRepository:
|
||||
def __init__(self, host, username, password, logger: Logger):
|
||||
@@ -455,3 +460,375 @@ class MLFlowRepository:
|
||||
metadata['mlflow_experiment_id'] = experiment_id
|
||||
|
||||
return metadata
|
||||
|
||||
def get_next_run_name_new(self, experiment_name: str) -> str:
|
||||
"""
|
||||
Generates the next run name for a given experiment.
|
||||
|
||||
Args:
|
||||
experiment_name (str): The name of the experiment for which the next run name is being generated.
|
||||
|
||||
Returns:
|
||||
str: A unique run name in the format "<experiment_name>-<next_run_number>".
|
||||
"""
|
||||
runs = self.model_serving.search_runs_by_name(
|
||||
experiment_names=[experiment_name], order_by=['start_time desc']
|
||||
)
|
||||
|
||||
next_run_number = len(runs) + 1
|
||||
return f'{experiment_name}-{next_run_number}'
|
||||
|
||||
def generate_artifacts(self, data: TrainModelResult) -> TrainModelResult:
|
||||
"""
|
||||
Generates and organizes artifacts related to the training process, such as reports and data files.
|
||||
|
||||
Args:
|
||||
data: The training model result containing the datasets, model, and parameters.
|
||||
|
||||
Returns:
|
||||
The updated result object with paths to the generated artifacts.
|
||||
|
||||
Raises:
|
||||
FileNotFoundError: If the reports directory or header.html file does not exist.
|
||||
ValueError: If run_name is not set.
|
||||
"""
|
||||
# Validate that run_name is set
|
||||
if not data.run_name:
|
||||
error_msg = 'run_name must be set before generating artifacts'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
reference_data, current_data = self._init_artifacts_data(data)
|
||||
base_path = self._get_reports_directory()
|
||||
|
||||
# Validate that reports directory exists
|
||||
if not path.exists(base_path):
|
||||
error_msg = f'Reports directory does not exist: {base_path}'
|
||||
self.logger.error(error_msg)
|
||||
raise FileNotFoundError(error_msg)
|
||||
|
||||
data.run_dir = self._create_run_directory(base_path, data.run_name)
|
||||
header_file_path = path.join(base_path, 'header.html')
|
||||
|
||||
# Validate that header.html exists
|
||||
if not path.exists(header_file_path):
|
||||
error_msg = f'Header file does not exist: {header_file_path}'
|
||||
self.logger.error(error_msg)
|
||||
raise FileNotFoundError(error_msg)
|
||||
|
||||
self._setup_run_directory(data.run_dir, header_file_path)
|
||||
return self._generate_report(reference_data, current_data, data)
|
||||
|
||||
def save_run(self, data: TrainModelResult):
|
||||
"""
|
||||
Logs the details of a machine learning run, including parameters, metrics, models, and artifacts,
|
||||
to the Sientia tracking system.
|
||||
|
||||
Args:
|
||||
data: The training model result containing the datasets, model, parameters,
|
||||
and evaluation metrics.
|
||||
|
||||
Raises:
|
||||
ValueError: If required metrics or artifacts are missing.
|
||||
Exception: If MLflow logging fails for any reason.
|
||||
"""
|
||||
# Validate that required artifacts exist before attempting to log
|
||||
if not data.report_path or not path.exists(data.report_path):
|
||||
error_msg = f'Report file does not exist: {data.report_path}'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
if not data.train_data_path or not path.exists(data.train_data_path):
|
||||
error_msg = f'Training data file does not exist: {data.train_data_path}'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
if not data.test_data_path or not path.exists(data.test_data_path):
|
||||
error_msg = f'Test data file does not exist: {data.test_data_path}'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
# Validate that metrics are present
|
||||
if data.mse_val is None or data.r2_val is None or data.mae_val is None:
|
||||
error_msg = 'One or more metrics (MSE, R2, MAE) are None'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
try:
|
||||
# Prepare parameters
|
||||
train_test_split = f'{data.params.train_size}-{100 - data.params.train_size}'
|
||||
interval_strs = [
|
||||
(str(interval[0]), str(interval[1]))
|
||||
for interval in (data.params.removed_intervals or [])
|
||||
]
|
||||
|
||||
# Set experiment and create run
|
||||
self.model_serving.set_experiment(data.params.experiment_name)
|
||||
self.logger.info(
|
||||
f"Logging run '{data.run_name}' to experiment '{data.params.experiment_name}'"
|
||||
)
|
||||
|
||||
with self.model_serving.save_experiment(
|
||||
run_name=data.run_name, description=data.params.experiment_name
|
||||
):
|
||||
# Log model parameters
|
||||
self.model_serving.log_param('model_type', 'Linear Regression')
|
||||
self.model_serving.log_param('target_variable', data.params.target_variable)
|
||||
self.model_serving.log_param('input_variables', data.params.variable_columns)
|
||||
self.model_serving.log_param('lag_train', data.params.lag_train)
|
||||
self.model_serving.log_param('lag_val', data.params.lag_val)
|
||||
self.model_serving.log_param('ma', data.params.window)
|
||||
self.model_serving.log_param('low_lim', data.params.low_lim)
|
||||
self.model_serving.log_param('upp_lim', data.params.upp_lim)
|
||||
self.model_serving.log_param('normalized', data.scaler_dict)
|
||||
self.model_serving.log_param('ar', data.params.include_ar)
|
||||
self.model_serving.log_param('Train_test_split', train_test_split)
|
||||
self.model_serving.log_param('Removed_intervals', interval_strs)
|
||||
self.model_serving.log_param('Retrain', False)
|
||||
|
||||
# Log evaluation metrics
|
||||
self.model_serving.log_metric('MSE', data.mse_val)
|
||||
self.model_serving.log_metric('R2', data.r2_val)
|
||||
self.model_serving.log_metric('MAE', data.mae_val)
|
||||
|
||||
# Log models
|
||||
self.model_serving.log_model(data.process_data, 'data_model')
|
||||
self.model_serving.log_model(data.regr, 'prediction_model')
|
||||
|
||||
# Log artifacts
|
||||
self.model_serving.log_artifact(data.report_path)
|
||||
self.model_serving.log_artifact(data.train_data_path)
|
||||
self.model_serving.log_artifact(data.test_data_path)
|
||||
|
||||
self.logger.info(
|
||||
f"Successfully logged run '{data.run_name}' with metrics: MSE={data.mse_val:.4f}, R2={data.r2_val:.4f}, MAE={data.mae_val:.4f}"
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
error_msg = f"Failed to save run '{data.run_name}' to MLflow: {str(e)}"
|
||||
self.logger.error(error_msg)
|
||||
raise Exception(error_msg) from e
|
||||
|
||||
def _init_artifacts_data(self, data: TrainModelResult) -> tuple[pd.DataFrame, pd.DataFrame]:
|
||||
"""
|
||||
Prepares the reference and current datasets for artifact generation.
|
||||
|
||||
Args:
|
||||
data: The training model result containing the datasets and model.
|
||||
|
||||
Returns:
|
||||
tuple: A tuple containing:
|
||||
- reference_data: The training dataset with predictions added.
|
||||
- current_data: The testing dataset with predictions added.
|
||||
|
||||
Raises:
|
||||
ValueError: If training or test datasets are empty or invalid.
|
||||
AttributeError: If required attributes are missing from the data object.
|
||||
"""
|
||||
# Validate that required DataFrames are not empty
|
||||
# Note: x_train, y_train, x_test, y_test, and regr are required fields in TrainModelResult
|
||||
# so we only check if they are empty, not None
|
||||
if data.x_train.empty:
|
||||
error_msg = 'Training features (x_train) are empty'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
if data.y_train.empty:
|
||||
error_msg = 'Training target (y_train) is empty'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
if data.x_test.empty:
|
||||
error_msg = 'Test features (x_test) are empty'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
if data.y_test.empty:
|
||||
error_msg = 'Test target (y_test) is empty'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
# Validate that predictions exist (y_pred is optional, so check for None)
|
||||
if data.y_pred is None:
|
||||
error_msg = 'Test predictions (y_pred) are None'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
# Prepare reference data (training set)
|
||||
reference_data = pd.concat([data.x_train, data.y_train], axis=1)
|
||||
reference_data = reference_data.rename(columns={data.params.target_variable: 'target'})
|
||||
reference_data['prediction'] = data.regr.predict(data.x_train)
|
||||
|
||||
# Prepare current data (test set)
|
||||
current_data = pd.concat([data.x_test, data.y_test], axis=1)
|
||||
current_data = current_data.rename(columns={data.params.target_variable: 'target'})
|
||||
current_data['prediction'] = data.y_pred
|
||||
|
||||
return reference_data, current_data
|
||||
|
||||
def _create_run_directory(self, base_path: str, run_name: str) -> str:
|
||||
"""
|
||||
Creates a directory inside the 'reports' folder with the run name and a timestamp.
|
||||
|
||||
Uses microsecond precision in timestamp to minimize collision probability
|
||||
in high-concurrency scenarios.
|
||||
|
||||
Args:
|
||||
base_path (str): The path to the 'reports' folder.
|
||||
run_name (str): The name of the run.
|
||||
|
||||
Returns:
|
||||
str: The path to the created directory.
|
||||
|
||||
Raises:
|
||||
PermissionError: If there are insufficient permissions to create the directory.
|
||||
OSError: If directory creation fails for any other reason.
|
||||
"""
|
||||
# Use microsecond precision to reduce collision probability
|
||||
timestamp = datetime.now().strftime('%Y%m%d_%H%M%S_%f')
|
||||
run_dir = path.join(base_path, f'{run_name}_{timestamp}')
|
||||
|
||||
try:
|
||||
makedirs(run_dir, exist_ok=True)
|
||||
self.logger.info(f'Created run directory: {run_dir}')
|
||||
return run_dir
|
||||
except PermissionError as e:
|
||||
error_msg = f'Permission denied when creating directory: {run_dir}'
|
||||
self.logger.error(error_msg)
|
||||
raise PermissionError(error_msg) from e
|
||||
except OSError as e:
|
||||
error_msg = f'Failed to create directory {run_dir}: {str(e)}'
|
||||
self.logger.error(error_msg)
|
||||
raise OSError(error_msg) from e
|
||||
|
||||
def _setup_run_directory(self, run_dir: str, header_file_path: str):
|
||||
"""
|
||||
Creates empty files and copies a header file into the specified run directory.
|
||||
|
||||
Note: Lock removed as each run has its own unique directory, so no synchronization
|
||||
is needed between different runs. File operations within the same directory are
|
||||
atomic at the OS level.
|
||||
|
||||
Args:
|
||||
run_dir (str): The path to the run directory where the files will be created.
|
||||
header_file_path (str): The path to the header.html file to be copied.
|
||||
|
||||
Raises:
|
||||
FileNotFoundError: If the header file does not exist.
|
||||
PermissionError: If there are insufficient permissions to create files.
|
||||
OSError: If file creation or copying fails for any other reason.
|
||||
"""
|
||||
empty_files = ['data_drift.html', 'data_quality.html', 'regression.html']
|
||||
|
||||
try:
|
||||
# Create empty placeholder files
|
||||
for file_name in empty_files:
|
||||
file_path = path.join(run_dir, file_name)
|
||||
with open(file_path, 'w'):
|
||||
pass # Create empty file
|
||||
|
||||
# Copy header file to run directory
|
||||
header_dest = path.join(run_dir, 'header.html')
|
||||
shutil.copy(header_file_path, header_dest)
|
||||
|
||||
self.logger.info(f'Run directory setup completed successfully in: {run_dir}')
|
||||
|
||||
except FileNotFoundError as e:
|
||||
error_msg = f'Header file not found: {header_file_path}'
|
||||
self.logger.error(error_msg)
|
||||
raise FileNotFoundError(error_msg) from e
|
||||
except PermissionError as e:
|
||||
error_msg = f'Permission denied when setting up directory: {run_dir}'
|
||||
self.logger.error(error_msg)
|
||||
raise PermissionError(error_msg) from e
|
||||
except OSError as e:
|
||||
error_msg = f'Failed to setup run directory {run_dir}: {str(e)}'
|
||||
self.logger.error(error_msg)
|
||||
raise OSError(error_msg) from e
|
||||
|
||||
def _generate_report(
|
||||
self, reference_data: pd.DataFrame, current_data: pd.DataFrame, data: TrainModelResult
|
||||
) -> TrainModelResult:
|
||||
"""
|
||||
Generates a comprehensive report summarizing data quality, data drift, and regression analysis.
|
||||
|
||||
Args:
|
||||
reference_data (pd.DataFrame): The training dataset with predictions added.
|
||||
current_data (pd.DataFrame): The testing dataset with predictions added.
|
||||
data: The training model result containing the datasets, model, and parameters.
|
||||
|
||||
Returns:
|
||||
The updated result object with paths to the generated report and data files.
|
||||
|
||||
Raises:
|
||||
ValueError: If data conversion to float64 fails or DataFrames are invalid.
|
||||
PermissionError: If there are insufficient permissions to write files.
|
||||
OSError: If file writing fails for any other reason.
|
||||
"""
|
||||
try:
|
||||
# Convert data to float64 for report generation
|
||||
# This may raise ValueError if data contains non-numeric values
|
||||
reference_data_float = reference_data.astype(np.float64)
|
||||
current_data_float = current_data.astype(np.float64)
|
||||
|
||||
# Initialize report generator
|
||||
report = Reports(
|
||||
reference_data=reference_data_float,
|
||||
current_data=current_data_float,
|
||||
base_path=data.run_dir,
|
||||
)
|
||||
|
||||
# Generate report sections
|
||||
report.add_data_quality_section(columns=data.params.variable_columns + ['target'])
|
||||
report.add_data_drift_section(columns=data.params.variable_columns + ['target'])
|
||||
report.add_regression_section()
|
||||
|
||||
# Validate that run_dir is set (should be set by _create_run_directory)
|
||||
if not data.run_dir:
|
||||
error_msg = 'run_dir is not set after directory creation'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg)
|
||||
|
||||
# Save HTML report
|
||||
data.report_path = path.join(data.run_dir, 'report.html')
|
||||
report.save_all_sections_html(data.report_path)
|
||||
self.logger.info(f'Generated HTML report: {data.report_path}')
|
||||
|
||||
# Save training data CSV
|
||||
data.train_data_path = path.join(data.run_dir, 'train_data.csv')
|
||||
reference_data.to_csv(data.train_data_path, index=False)
|
||||
self.logger.info(f'Saved training data: {data.train_data_path}')
|
||||
|
||||
# Save test data CSV
|
||||
data.test_data_path = path.join(data.run_dir, 'test_data.csv')
|
||||
current_data.to_csv(data.test_data_path, index=False)
|
||||
self.logger.info(f'Saved test data: {data.test_data_path}')
|
||||
|
||||
return data
|
||||
|
||||
except ValueError as e:
|
||||
error_msg = f'Failed to convert data to float64 for report generation: {str(e)}'
|
||||
self.logger.error(error_msg)
|
||||
raise ValueError(error_msg) from e
|
||||
except PermissionError as e:
|
||||
error_msg = f'Permission denied when writing report files to: {data.run_dir}'
|
||||
self.logger.error(error_msg)
|
||||
raise PermissionError(error_msg) from e
|
||||
except OSError as e:
|
||||
error_msg = f'Failed to generate report in {data.run_dir}: {str(e)}'
|
||||
self.logger.error(error_msg)
|
||||
raise OSError(error_msg) from e
|
||||
|
||||
def _get_reports_directory(self) -> str:
|
||||
"""
|
||||
Get the absolute path to the reports directory.
|
||||
|
||||
Returns:
|
||||
str: Absolute path to model_manager/reports directory.
|
||||
"""
|
||||
# Get the directory where this file is located (model_manager/utils/repository/)
|
||||
current_file_dir = path.dirname(path.abspath(__file__))
|
||||
# Navigate up to model_manager/ and then to reports/
|
||||
model_manager_dir = path.dirname(path.dirname(current_file_dir))
|
||||
reports_dir = path.join(model_manager_dir, 'reports')
|
||||
return reports_dir
|
||||
|
||||
Reference in New Issue
Block a user