Cannot find custom model handler

[deadsoul44]

We exactly followed this doc to save a custom model:

https://docs.snowflake.com/en/developer-guide/snowflake-ml/model-registry/custom-models

But we get a handler error as described in this issue: #135

TypeError("Your input type to custom model is not currently supported")

Could you please help us to find a way to define a custom handler?

[sfc-gh-xjiang]

Hey @deadsoul44 , thanks for reporting this issue and we removed this assertion in the next release.

Apart from the the assertion, is it possible for you to provide a minimal reproducible example so that we can help investigate the issue?

Thanks,
Xinyi

[deadsoul44]

We were trying to save a model that was trained using PerpetualBooster:

https://github.com/perpetual-ml/perpetual

Perpetual is still not available on the Snowflake conda channel, although we made a request almost two months ago.

It should be stated in docs that the package should be available on the Snowflake Anaconda channel.

Or there should be a way to define a custom handler.

[sfc-gh-xjiang]

Got it, I wrote an example that may work in your use case, can you try?

import pickle
import os

from snowflake.ml.model import custom_model
from sklearn.svm import SVC
import pandas as pd


# Step 1: Train the model
model = SVC() <<please_replace_with_your_model>> # Example model, please replace with your own model
data = pd.DataFrame({'feature1': [1, 2, 3], 'feature2': [4, 5, 6]})
target = [1, 2, 3]
model.fit(data, target)

# Step 2: Save the model to a file
os.makedirs('path_to_save', exist_ok=True)

with open('path_to_save/model.pkl', 'wb') as f:
    pickle.dump(model, f)

# Step 3: Define a ModelContext with the path to the saved model
mc = custom_model.ModelContext(
    model_file='path_to_save/model.pkl',
)

# Step 4: Define a custom model class that loads the model and makes prediction
class ExampleBringYourOwnModel(custom_model.CustomModel):
    def __init__(self, context: custom_model.ModelContext) -> None:
        super().__init__(context)
        with open(context['model_file'], 'rb') as f:
            self.model = pickle.load(f)

    @custom_model.inference_api
    def predict(self, input: pd.DataFrame) -> pd.DataFrame:
        # Return predictions and probabilities as a DataFrame
        return pd.DataFrame({'output': self.model.predict(input)})

# Example usage:
# Initialize the custom model and run predictions
bring_your_own_model = ExampleBringYourOwnModel(context=mc)
result = bring_your_own_model.predict(data)
print(result)


# Snowpark for Python
from snowflake.ml.registry import Registry
session = <<create_your_session>>

# Step 5: Log the custom model to the registry
reg = Registry(session=session, database_name=session.get_current_database(), schema_name=session.get_current_schema())

mv = reg.log_model(bring_your_own_model,
            model_name="my_model",
            version_name="version1",
            conda_dependencies=["scikit-learn"],  #<<please_include_dependency_here>>
            comment="My Custom ML Model Pipeline",
            sample_input_data=data)

# Step 6: Run the model in the registry
output_df = mv.run(data)
print(output_df)

Please let me know if you've met any issues
Xinyi

[sfc-gh-xjiang]

Hey @deadsoul44 ,

Just following up on this issue. Has there been any progress or updates regarding the fix above? If there's anything I can help with, let me know!

Thanks in advance for your time.

Xinyi