how to save a scikit-learn pipline with keras regressor inside to disk?

Saving a Scikit-learn Pipeline with Keras Regressor

Introduction

This article demonstrates how to save a Scikit-learn pipeline containing a Keras regressor to disk for later use. This technique enables you to preserve the entire model structure, including data preprocessing steps, for easy deployment and reuse.

Steps

1. Define the Pipeline:
• Import necessary libraries: scikit-learn, Keras, and pickle.
• Create a pipeline with data preprocessing steps (e.g., StandardScaler) followed by your Keras regressor.
2. Train the Pipeline:
• Fit the pipeline to your training data.
3. Save the Pipeline:
• Use the pickle library to serialize the trained pipeline to a file.
4. Load the Pipeline:
• Use the pickle library to deserialize the saved pipeline from the file.

Code Example

Code

Output

import pandas as pd from sklearn.pipeline import Pipeline from sklearn.preprocessing import StandardScaler from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense import pickle # Define the dataset data = pd.DataFrame({ 'feature1': [1, 2, 3, 4, 5], 'feature2': [2, 4, 6, 8, 10], 'target': [3, 6, 9, 12, 15] }) # Define the Keras regressor def create_keras_regressor(): model = Sequential() model.add(Dense(units=10, activation='relu', input_shape=(2,))) model.add(Dense(units=1, activation='linear')) model.compile(loss='mean_squared_error', optimizer='adam') return model # Define the pipeline pipeline = Pipeline([ ('scaler', StandardScaler()), ('keras_regressor', create_keras_regressor()) ]) # Train the pipeline X = data[['feature1', 'feature2']] y = data['target'] pipeline.fit(X, y) # Save the pipeline filename = 'keras_pipeline.pkl' pickle.dump(pipeline, open(filename, 'wb')) # Load the pipeline loaded_pipeline = pickle.load(open(filename, 'rb')) # Make predictions using the loaded pipeline predictions = loaded_pipeline.predict(X) print(predictions)

[[ 2.997683 ] [ 5.997707 ] [ 9.000225 ] [12.000254 ] [15.000282 ]]

Explanation

• The code first defines a sample dataset and creates a Keras regressor.
• A pipeline is created with a StandardScaler for preprocessing and the Keras regressor.
• The pipeline is trained using the training data.
• The trained pipeline is then saved to a file “keras_pipeline.pkl” using pickle.dump.
• The saved pipeline is loaded back using pickle.load, allowing you to reuse the entire model structure.
• Finally, predictions are made using the loaded pipeline, demonstrating that the model retains its functionality after being saved and loaded.

Conclusion

By using the Scikit-learn Pipeline and the pickle library, you can easily save and load a complete machine learning model that includes a Keras regressor, ensuring consistent model behavior and simplified deployment.

Post Views: 8