добавить логирование прогонов в MLFlow

mlflow/.gitignore

@@ -0,0 +1,2 @@
+mlruns.sqlite
+mlartifacts/

requirements.txt

@@ -2,7 +2,9 @@ bokeh >=3.7.2,<4
 ipykernel >=6.30.1,<7
 ipympl ~=0.9.6
 matplotlib >=3.10.1,<4
-numpy >=2.3.1,<3
+mlflow>=2.16,<2.22
+numpy >=2.2.6,<3
 pandas >=2.3.1,<3
-scipy >=1.16.1,<2
+scipy >=1.15.3,<2
+scikit-learn >=1.7.2,<2
 seaborn ~=0.13.2

research/research.py

@@ -0,0 +1,230 @@
+# ---
+# jupyter:
+#   jupytext:
+#     formats: py:percent,ipynb
+#     text_representation:
+#       extension: .py
+#       format_name: percent
+#       format_version: '1.3'
+#       jupytext_version: 1.17.3
+#   kernelspec:
+#     display_name: .venv
+#     language: python
+#     name: python3
+# ---
+
+# %%
+from typing import Optional
+
+# %%
+# %% tags=["parameters"]
+data_path: Optional[str] = None
+# Полный путь к файлу (CSV) с исходным датасетом. Если не установлен, ищется файл в `data/<data_relpath>`.
+data_relpath: str = 'cars.csv'
+# Путь к файлу (CSV) с исходным датасетом относительно директории данных `data`. Игнорируется, если установлен data_path.
+
+data_aug_pickle_path: Optional[str] = None
+# Полный путь к файлу (pickle) для сохранения очищенного датасета. Если не установлен, используется `data/<data_aug_pickle_relpath>`.
+data_aug_pickle_relpath: str = 'cars.aug.pickle'
+# Путь к файлу (pickle) для сохранения очищенного датасета относительно директории данных `data`. Игнорируется, если установлен data_aug_pickle_path.
+
+mlflow_tracking_server_uri: str = 'http://localhost:5000'
+mlflow_registry_uri: Optional[str] = None
+
+mlflow_experiment_name: str = 'Current price predicion for used cars'
+mlflow_experiment_new: bool = False
+mlflow_run_name: str = 'Baseline model'
+
+# %%
+import os
+import pathlib
+import pickle
+
+# %%
+import mlflow
+import mlflow.models
+import mlflow.sklearn
+import sklearn.compose
+import sklearn.ensemble
+import sklearn.metrics
+import sklearn.model_selection
+import sklearn.pipeline
+import sklearn.preprocessing
+
+# %%
+BASE_PATH = pathlib.Path('..')
+
+# %%
+mlflow.set_tracking_uri(mlflow_tracking_server_uri)
+if mlflow_registry_uri is not None:
+    mlflow.set_registry_uri(mlflow_registry_uri)
+
+# %%
+DATA_PATH = (
+    pathlib.Path(os.path.dirname(data_path))
+    if data_path is not None
+    else (BASE_PATH / 'data')
+)
+
+# %%
+with open(
+    (
+    data_aug_pickle_path
+    if data_aug_pickle_path is not None
+    else (DATA_PATH / data_aug_pickle_relpath)
+    ),
+    'rb',
+) as input_file:
+    df_orig = pickle.load(input_file)
+
+# %%
+df_orig.head(0x10)
+
+# %%
+len(df_orig)
+
+# %%
+df_orig.info()
+
+# %%
+feature_columns = (
+    'selling_price',
+    'driven_kms',
+    'fuel_type',
+    'selling_type',
+    'transmission',
+    #'owner',
+    'age',
+)
+
+target_columns = (
+    'present_price',
+)
+
+# %%
+features_to_scale_to_standard_columns = (
+    'selling_price',
+    'driven_kms',
+    'age',
+)
+assert all(
+    (col in df_orig.select_dtypes(('number',)).columns)
+    for col in features_to_scale_to_standard_columns
+)
+
+features_to_encode_one_hot_columns = (
+    'fuel_type',
+    'selling_type',
+    'transmission',
+    #'owner',
+)
+assert all(
+    (col in df_orig.select_dtypes(('category', 'object')).columns)
+    for col in features_to_encode_one_hot_columns
+)
+
+# %%
+df_orig_features = df_orig[list(feature_columns)]
+df_target = df_orig[list(target_columns)]
+
+# %%
+DF_TEST_PORTION = 0.25
+
+# %%
+df_orig_features_train, df_orig_features_test, df_target_train, df_target_test = (
+    sklearn.model_selection.train_test_split(
+        df_orig_features, df_target, test_size=DF_TEST_PORTION, random_state=0x7AE6,
+    )
+)
+
+# %%
+tuple(map(len, (df_target_train, df_target_test)))
+
+# %%
+preprocess_transformer = sklearn.compose.ColumnTransformer(
+    [
+        ('scale_to_standard', sklearn.preprocessing.StandardScaler(), features_to_scale_to_standard_columns),
+        (
+            #'encode_categoricals_one_hot',
+            'encode_categoricals_wrt_target',
+            #sklearn.preprocessing.OneHotEncoder(),
+            sklearn.preprocessing.TargetEncoder(
+                target_type='continuous', smooth='auto', cv=3, shuffle=True, random_state=0x2ED6,
+            ),
+            features_to_encode_one_hot_columns,
+        ),
+    ],
+    remainder='drop',
+)
+
+# %%
+regressor = sklearn.ensemble.RandomForestRegressor(
+    10, criterion='squared_error', max_features='sqrt', random_state=0x016B,
+)
+
+# %%
+pipeline = sklearn.pipeline.Pipeline([
+    ('preprocess', preprocess_transformer),
+    ('regress', regressor),
+])
+
+# %%
+pipeline
+
+# %%
+_ = pipeline.fit(df_orig_features_train, df_target_train.iloc[:, 0])
+
+# %%
+target_test_predicted = pipeline.predict(df_orig_features_test)
+
+# %%
+metrics = {
+    'mse': sklearn.metrics.mean_squared_error(df_target_test, target_test_predicted),
+    'mae': sklearn.metrics.mean_absolute_error(df_target_test, target_test_predicted),
+    'mape': sklearn.metrics.mean_absolute_percentage_error(df_target_test, target_test_predicted),
+}
+
+# %%
+metrics
+
+# %%
+MODEL_PIP_REQUIREMENTS_PATH = BASE_PATH / 'requirements.txt'
+MODEL_COMMENTS_FILE_PATH = BASE_PATH / 'comment.txt'
+
+# %%
+MODEL_INOUT_EXAMPLE_SIZE = 0x10
+
+# %%
+model_inout_example = (df_orig_features.head(MODEL_INOUT_EXAMPLE_SIZE), df_target.head(MODEL_INOUT_EXAMPLE_SIZE))
+
+# %%
+mlflow_model_signature = mlflow.models.infer_signature(model_input=model_inout_example[0], model_output=model_inout_example[1])
+
+# %%
+mlflow_model_signature
+
+# %%
+model_params = pipeline.get_params()
+
+# %%
+model_params
+
+# %%
+if mlflow_experiment_new:
+    experiment = mlflow.get_experiment(mlflow.create_experiment(mlflow_experiment_name))
+else:
+    experiment = mlflow.set_experiment(experiment_name=mlflow_experiment_name)
+
+# %%
+with mlflow.start_run(experiment_id=experiment.experiment_id, run_name=mlflow_run_name):
+    _ = mlflow.sklearn.log_model(
+        pipeline,
+        'model',
+        signature=mlflow_model_signature,
+        input_example=model_inout_example[0],
+        pip_requirements=str(MODEL_PIP_REQUIREMENTS_PATH),
+    )
+    _ = mlflow.log_params(model_params)
+    _ = mlflow.log_metrics({k: float(v) for k, v in metrics.items()})
+    if MODEL_COMMENTS_FILE_PATH.exists():
+        mlflow.log_artifact(str(MODEL_COMMENTS_FILE_PATH))

run_mlflow_server.sh

@@ -0,0 +1,13 @@
+# Must be a relative path to be used in an option for mlflow server.
+BACKEND_STORE_DB_PATH="./mlflow/mlruns.sqlite"
+DEFAULT_ARTIFACTS_ROOT="./mlflow/"
+
+if [ ! -e "$BACKEND_STORE_DB_PATH" ]; then
+    printf '%s\n' "Error: '$BACKEND_STORE_DB_PATH' does not exist." >&2
+    exit 1
+fi
+
+mlflow server \
+    --backend-store-uri="sqlite:///$BACKEND_STORE_DB_PATH"  \
+    --default-artifacts-root="$DEFAULT_ARTIFACTS_ROOT" \
+    -p 5000