Keras¶

The Snowflake ML Model Registry supports Keras 3 models (keras.Model with Keras version >= 3.0.0). Keras 3 is a multi-backend framework that supports TensorFlow, PyTorch, and JAX as backends.

Note

For Keras version < 3.0.0, use the TensorFlow handler.

The following additional options can be used in the options dictionary when you call log_model:


Option	Description
`target_methods`	A list of the names of the methods available on the model object. Keras models have `predict` as the default target method.
`cuda_version`	The version of the CUDA runtime to be used when deploying to a platform with GPU; defaults to 11.8. If manually set to `None`, the model cannot be deployed to a platform having a GPU.

You must specify either the sample_input_data or signatures parameter when logging a Keras model so that the registry knows the signatures of the target methods.

Note

Keras models can only have one target method.

Examples¶

These examples assume reg is an instance of snowflake.ml.registry.Registry.

Sequential Model¶

The following example demonstrates training a Keras 3 sequential model, logging it to the Snowflake ML Model Registry, and running inference.

import keras
from sklearn import datasets, model_selection

# Load dataset
iris = datasets.load_iris(as_frame=True)
X = iris.data
y = iris.target

# Rename columns for valid Snowflake identifiers
X.columns = [col.replace(' ', '_').replace('(', '').replace(')', '') for col in X.columns]

X_train, X_test, y_train, y_test = model_selection.train_test_split(X, y, test_size=0.2)

# Build Keras sequential model
model = keras.Sequential([
    keras.layers.Dense(64, activation='relu'),
    keras.layers.Dense(32, activation='relu'),
    keras.layers.Dense(3, activation='softmax')
])

model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy']
)

# Train the model
model.fit(X_train, y_train, epochs=50, verbose=0)

# Log the model
model_ref = reg.log_model(
    model=model,
    model_name="my_keras_classifier",
    version_name="v1",
    sample_input_data=X_test,
)

# Make predictions
result_df = model_ref.run(X_test[-10:], function_name="predict")

Functional API Model¶

The following example demonstrates creating a model using the Keras Functional API.

import keras
import numpy as np
import pandas as pd

# Create sample data
n_samples, n_features = 100, 10
X = pd.DataFrame(
    np.random.rand(n_samples, n_features),
    columns=[f"feature_{i}" for i in range(n_features)]
)
y = np.random.randint(0, 2, n_samples).astype(np.float32)

# Build model using Functional API
inputs = keras.Input(shape=(n_features,))
x = keras.layers.Dense(32, activation='relu')(inputs)
x = keras.layers.Dense(16, activation='relu')(x)
outputs = keras.layers.Dense(1, activation='sigmoid')(x)
model = keras.Model(inputs=inputs, outputs=outputs)

model.compile(
    optimizer=keras.optimizers.SGD(learning_rate=0.01),
    loss=keras.losses.MeanSquaredError()
)

# Train the model
model.fit(X, y, epochs=10, verbose=0)

# Log the model
model_ref = reg.log_model(
    model=model,
    model_name="my_functional_model",
    version_name="v1",
    sample_input_data=X,
)

# Make predictions
result_df = model_ref.run(X[-10:], function_name="predict")

Custom Subclass Model¶

The following example demonstrates creating a custom model by subclassing keras.Model.

import keras
import numpy as np
import pandas as pd

# Define custom model with serialization support
@keras.saving.register_keras_serializable()
class BinaryClassifier(keras.Model):
    def __init__(self, hidden_units: int, output_units: int) -> None:
        super().__init__()
        self.dense1 = keras.layers.Dense(hidden_units, activation="relu")
        self.dense2 = keras.layers.Dense(output_units, activation="sigmoid")

    def call(self, inputs):
        x = self.dense1(inputs)
        return self.dense2(x)

    def get_config(self):
        base_config = super().get_config()
        config = {
            "dense1": keras.saving.serialize_keras_object(self.dense1),
            "dense2": keras.saving.serialize_keras_object(self.dense2),
        }
        return {**base_config, **config}

    @classmethod
    def from_config(cls, config):
        dense1_config = config.pop("dense1")
        dense1 = keras.saving.deserialize_keras_object(dense1_config)
        dense2_config = config.pop("dense2")
        dense2 = keras.saving.deserialize_keras_object(dense2_config)
        obj = cls(1, 1)
        obj.dense1 = dense1
        obj.dense2 = dense2
        return obj

# Create sample data
n_samples, n_features = 100, 10
X = pd.DataFrame(
    np.random.rand(n_samples, n_features),
    columns=[f"feature_{i}" for i in range(n_features)]
)
y = np.random.randint(0, 2, n_samples).astype(np.float32)

# Create and train model
model = BinaryClassifier(hidden_units=32, output_units=1)
model.compile(
    optimizer=keras.optimizers.SGD(learning_rate=0.01),
    loss=keras.losses.MeanSquaredError()
)
model.fit(X, y, epochs=10, verbose=0)

# Log the model
model_ref = reg.log_model(
    model=model,
    model_name="my_custom_classifier",
    version_name="v1",
    sample_input_data=X,
)

# Make predictions
result_df = model_ref.run(X[-10:], function_name="predict")