diogo149 / pal

Pal for Active Learning

pal

Pal for Active Learning

pal is a tool for easily testing the effectiveness of active learning algoithms. See examples/ for some usage.

This chart shows a bunch of naive strategies trying for finding the best point to label on MNIST. The model starts with only 10 points in total, and we show the results of labeling 90 more points.

Dependencies

• numpy
• scipy
• scikit-learn
• pandas
• matplotlib
• seaborn (optional)

Running

Evaluate one of:

1. python examples/mnist.py
2. python -m examples.mnist
3. ipython -i examples/mnist.py (my preference)

to download mnist data and generate a chart of the results (mnist.png).