import ibis
import ibisml as ml

# A recipe for a feature engineering pipeline that:
# - imputes missing values in numeric columns with their mean
# - applies standard scaling to all numeric columns
# - one-hot-encodes all nominal columns
recipe = ml.Recipe(
    ml.ImputeMean(ml.numeric()),
    ml.ScaleStandard(ml.numeric()),
    ml.OneHotEncode(ml.nominal()),
)

# Use the recipe inside of a larger Scikit-Learn pipeline
from sklearn.pipeline import Pipeline
pipeline = Pipeline([("recipe", recipe), ("model", LinearSVC())])

# Fit the recipe against some local training data,
# just as you would with any other scikit-learn model
X, y = load_training_data()
pipeline.fit(X, y)

# Evaluate the model against some local testing data.
X_test, y_test = load_testing_data()
pipeline.score(X_test, y_test)

# Now apply the same preprocessing pipeline against any of ibis's
# supported backends
con = ibis.connect(...)
X_remote = con.table["mytable"]
for batch in recipe.to_pyarrow_batches(X_remote):
    ...