3. Caching

It is very simple to cache the output from parameter sweeps. First, enable caching:

ps.enable_cache()

Second, you need to have the function return the data you want to cache. In this case, we want to cache the pixels from the matplotlib plot, instead of writing file to disk through savefig(). Although a little involved, and could be done this way:

from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas
def create_graph(length, width, position):
    graph(length, width, position)
    canvas = FigureCanvas(gcf())
    canvas.draw()
    image = np.frombuffer(canvas.tostring_rgb(), dtype='uint8')
    f = gcf()
    plt.close()
    return image.reshape((int(f.get_figheight()*f.get_dpi()), int(f.get_figwidth()*f.get_dpi()), 3))

A simpler option could be writing the file to disk and then just reading it back, but takes up double the disk space and would not be suitable if the data is large, as the cache would be a second file written to disk.

The returned data from this function are the graph pixels:

im = create_graph(1,3,0.3)

import IPython.display
import PIL.Image
IPython.display.display(PIL.Image.fromarray(im))

You can run the sweep with:

ps.sweep(create_graph)

You will find that if you interrupt the sweep after enabling cache, the next time you start a sweep it will continue where the previous sweep left off.

You can get specific data from the cache thorugh the run_process() function:

l.value = 3
w.value = 7
pos.value = 0.5
im = ps.run_process(save_graph3)
plt.axis('off')
imshow(im)

If cache is enabled, the first time the function is executed through run_process() will produce cache and take longer. Note that sweep() calls run_process() internally, so any cache produced by the sweep will be available through run_process()

Also note that the parameter callbacks are not called when there is cache available. In some instances, you might want to force the parameters to actually change, you will need to add force_recompute=True when calling run_process().

You can clear the disk cache using clear_cache():

ps.clear_cache()

Next: 4. Notebook Widgets