Monitoring and visualization

End-to-End Machine Learning

Joshua Stapleton

Machine Learning Engineer

What's next?

  • Trained, optimized, deployed, predicted... what next?
  • Monitoring
    • Logging results
    • Visualizing performance

Monitoring phase in the machine learning lifecycle

End-to-End Machine Learning

Logging with python

import logging
import matplotlib.pyplot as plt

# Setting up basic logging configuration
logging.basicConfig(filename='predictions.log', level=logging.INFO)

# Make predictions on the test set and log the results
for i in range(X_test.shape[0]):
    instance = X_test[i,:].reshape(1, -1)
    prediction = model.predict(instance)
    logging.info(f'Inst. {i} - PredClass: {prediction[0]}, RealClass: {y_test[i]}')
End-to-End Machine Learning

Logging with python (cont.)

# Function to visualize the predictions from log
with open(logfile, 'r') as f:
    lines = f.readlines()
    predicted_classes = [int(line.split("Predicted Class: ")[1].split(",")[0]) \
        for line in lines]

    # Perform data analysis, visualization, etc.
    ...
  • Use Python logging to trace model performance
End-to-End Machine Learning

Visualization

  • Inspect performance over time
  • Transform raw data of inputs / predictions into insights
import matplotlib.pyplot as plt


# Sample data: Random accuracy values for 12 months
months = ["Jan", "Feb", "Mar", ...]
accuracies = [0.86, 0.91, 0.74, ...]

plt.plot(months, accuracies, '-o')
plt.title("Model Accuracy Over Months")
plt.xlabel("Months")
plt.ylabel("Accuracy")
plt.show()
End-to-End Machine Learning

Visualization example

A line graph showing hypothetical model accuracy over a 12 month period

End-to-End Machine Learning

Logging

  • Recording of events

    • Tracking variable values, Function calls
    • Information that informs execution + performance

  • Monitoring helps track:

    • Usage, Performance, Errors/anomalies
2023-08-04 09:15:20 [INFO] Model version 1.2.7 started

2023-08-04 09:15:45 [INFO] Preprocessing input data for prediction

2023-08-04 09:15:47 [DEBUG] Input data shape: (1, 12)

2023-08-04 09:15:48 [INFO] Making prediction

2023-08-04 09:15:50 [DEBUG] Output prediction: [0.78]

...
End-to-End Machine Learning

Visualization examples

  • Helpful metric for our model: balanced accuracy over time
  • Spot trends, see if performance degrades
  • See if retraining is necessary
  • Choose helpful metrics for our use-case

Example:

  • Balanced accuracy changes relative to expected, real-world rate
  • Potentially indicative of problem
  • Choose and evaluate

Visualization of model's balanced accuracy

End-to-End Machine Learning

Let's practice!

End-to-End Machine Learning

Preparing Video For Download...