machine learning

References

Summary

Experience training, evaluating, and deploying models end-to-end (data preprocessing → training → validation → inference → monitoring)

Packages

Pandas

Scikit-Learn

PySpark

Open-source Big data processing and analytics

TensorFlow

Open-source

For large-scale production Static computation graphs

Declarative

Structure

Works with multidimensional immutable arrays (which they call tensors)

Layers are objects with the following methods

Layers

Some standard layers are

Pre-processing

If you're working with images, normalize the data, so that the pixels are all given values between 0 and 1

Compiling

  1. Loss Function

  2. Optimization

  3. Metrics

Training

Feed in the data

model.fit(train_images, train_labels, epochs = 20)

Evaluate performance

test_loss, test_acc = model.evaluate(test_images, test_labels, verbose = 2)

print('\nTest accuracy:', test_acc)
  1. Softmax

    Converts the logits to probabilities

    Then use

    probability_model.predict(test_images)
    

    To see how confident the model is

PyTorch

Open-source ML library

Dynamic computation graphs

Imperative

JAX

Open-source numerical computation libraryn

Just-in-time compilation

Natural Language Processing