plot_filters.py

import numpy as np
import matplotlib.pyplot as plt


def plot_filters(filters, shape, num_rows, num_cols):
    """
    Assumes filters are rows (so transpose it that way if necessary).

    matplotlib isn't the speediest horse in town, I wouldn't draw
    more than a few dozen this way if you need it to frequently update.
    An alternative would be to pack stuff into one big array rather
    than using subplots (this is what pylearn2's show_filters script
    does, but in a somewhat inflexible way).
    """
    # Speed things up a bit by drawing only when we're done.
    interactive = plt.isinteractive()
    try:
        if interactive:
            plt.ioff()
        filters = np.asarray(filters)
        assert filters.ndim == 2
        num_filters_displayed = min(num_rows * num_cols, filters.shape[0])
        min_value, max_value = filters.min(), filters.max()
        for i in range(num_filters_displayed):
            plt.subplot(num_rows, num_cols, i + 1)
            plt.imshow(filters[i].reshape(shape),
                       cmap=plt.cm.gray, interpolation='nearest')
            plt.xticks([])
            plt.yticks([])
            plt.clim(min_value, max_value)
    finally:
        # Restore interactive state, if necessary.
        if interactive:
            plt.show()  # Do queued up plotting commands.
            plt.ion()