Save and Load a Model

In this tutorial, we illustrate the process of saving and loading a model using CapyMOA.

• We use the SEA synthetic generator as the data source, and the AdaptiveRandomForestClassifier as the learner.

• The trained model is saved to a file, specifically ‘capymoa_model.pkl’.

• Subsequently, we reload the model from the file and resume training and evaluating its performance on the SEA data.

• As a final step, we delete the model file.

1. Training and saving the model

• We train the model on 5k instances from SEA using the evaluate_prequential function

• We proceed to save the model with save_model(learner, "capymoa_ARF_model.pkl")

from capymoa.classifier import AdaptiveRandomForestClassifier
from capymoa.evaluation import prequential_evaluation
from capymoa.stream.generator import SEA
from capymoa.misc import save_model, load_model

stream = SEA()
learner = AdaptiveRandomForestClassifier(schema=stream.get_schema(), ensemble_size=10)

results = prequential_evaluation(stream=stream, learner=learner, max_instances=5000)

print(f"Accuracy: {results['cumulative'].accuracy():.2f}")
save_model(learner, "capymoa_ARF_model.pkl")  # Save model to capymoa_model.pkl

Accuracy: 87.96

2. Loading and resuming training

• We use os.path.getsize() to inspect the size (KB) of the saved file.

• We don’t restart the synthetic stream, we just continue processing it through another call to prequential_evaluation

• Finally, we observe the accuracy

import os

model_file = "capymoa_ARF_model.pkl"

model_size = os.path.getsize(model_file)
print(f"The saved model size: {model_size / 1024:.2f} KB")

restored_learner = load_model(
    "capymoa_ARF_model.pkl"
)  # Load model from capymoa_model.pkl

# Train for more 50k instances on the restored model
results = prequential_evaluation(
    stream=stream, learner=restored_learner, max_instances=5000
)

print(f"Updated accuracy: {results['cumulative'].accuracy():.2f}")

The saved model size: 618.27 KB
Updated accuracy: 89.30

3. Cleanup

• As a last step, we delete the model

if os.path.exists(model_file):
    os.remove(model_file)
    print(f"File {model_file} has been deleted.")
else:
    print(f"File {model_file} not found.")

File capymoa_ARF_model.pkl has been deleted.