Snowpark ML Model Registryから最新バージョンのモデルを利用する

# 以下ガイドより2024/10/26に引用
# https://docs.snowflake.com/developer-guide/snowflake-ml/model-registry/overview#registering-models-and-versions

mv = reg.log_model(clf,
                   model_name="my_model",
                   version_name="v1",
                   conda_dependencies=["scikit-learn"],
                   comment="My awesome ML model",
                   metrics={"score": 96},
                   sample_input_data=train_features,
                   task=type_hints.Task.TABULAR_BINARY_CLASSIFICATION)

import pandas as pd
import numpy as np

from snowflake.snowpark.context import get_active_session
session = get_active_session()

import snowflake.ml.modeling.preprocessing as snowml
from snowflake.ml.modeling.pipeline import Pipeline
from snowflake.ml.modeling.model_selection import GridSearchCV
from snowflake.ml.modeling.xgboost import XGBRegressor

# Snowflakeのテーブルからのデータの取得
# sessionはあらかじめ作成しておく
DEMO_TABLE = 'DIAMONDS'
input_tbl = f"{session.get_current_database()}.{session.get_current_schema()}.{DEMO_TABLE}"
diamonds_df = session.table(input_tbl)

# データセットの分割
diamonds_train_df, diamonds_test_df = diamonds_df.random_split(weights=[0.9, 0.1], seed=0)

# 前処理・学習・推論のためのカラム定義
CATEGORICAL_COLUMNS = ["CUT", "COLOR", "CLARITY"]
CATEGORICAL_COLUMNS_OE = ["CUT_OE", "COLOR_OE", "CLARITY_OE"]
NUMERICAL_COLUMNS = ["CARAT", "DEPTH", "TABLE_PCT", "X", "Y", "Z"]
categories = {
    "CUT": np.array(["Ideal", "Premium", "Very Good", "Good", "Fair"]),
    "CLARITY": np.array(["IF", "VVS1", "VVS2", "VS1", "VS2", "SI1", "SI2", "I1", "I2", "I3"]),
    "COLOR": np.array(['D', 'E', 'F', 'G', 'H', 'I', 'J']),
}
LABEL_COLUMNS = ['PRICE']
OUTPUT_COLUMNS = ['PREDICTED_PRICE']

# 機械学習パイプラインの定義
pipeline = Pipeline(
    steps=[
            (
                "OE",
                snowml.OrdinalEncoder(
                    input_cols=CATEGORICAL_COLUMNS,
                    output_cols=CATEGORICAL_COLUMNS_OE,
                    categories=categories,
                )
            ),
            (
                "MMS",
                snowml.MinMaxScaler(
                    clip=True,
                    input_cols=NUMERICAL_COLUMNS,
                    output_cols=NUMERICAL_COLUMNS,
                )
            ),
            (
                "GridSearchCV",
                GridSearchCV(
                    estimator=XGBRegressor(random_state=42),
                        param_grid={
                                        "n_estimators":[300, 400],
                                        "learning_rate":[0.1, 0.2],
                                        },
                        n_jobs = -1,
                        scoring="neg_mean_squared_error",
                        input_cols=CATEGORICAL_COLUMNS_OE+NUMERICAL_COLUMNS,
                        label_cols=LABEL_COLUMNS,
                        output_cols=OUTPUT_COLUMNS
                )
            )
    ]
)

# モデルのトレーニング
pipeline.fit(diamonds_train_df)

from snowflake.ml.registry import Registry

reg = Registry(session=session)

# V_1の登録
reg.log_model(
    model_name="diamond_xgboost_regressor",
    version_name="V_1",
    model=pipeline,
    comment="version 1"
)

# V_2の登録
reg.log_model(
    model_name="diamond_xgboost_regressor",
    version_name="V_2",
    model=pipeline,
    comment="version 2"
)

m = reg.get_model("diamond_xgboost_regressor")
mv = m.version("LAST")

print(mv.comment)
# version 2

WITH LATEST AS MODEL DIAMOND_XGBOOST_REGRESSOR VERSION LAST
    SELECT LATEST!PREDICT(*):PREDICTED_PRICE::FLOAT AS PREDICTED_PRICE
    FROM 
    (
        SELECT * EXCLUDE PRICE 
        FROM DIAMONDS
        LIMIT 10
    );