Report incremental updates to "Classify Many"

[lukeyeager]

Problem

There are several problems with trying to use "Classify Many" at scale.

1. There is no indication in the browser that the server is working on anything. You just click "Classify Many" and the page hangs until you get a response.
2. As reported in Output the testing results of many images as text file #70 (comment), Flask can't deal returning the response of classifying 1,000,000 images.
3. As reported in Classify many, uploads all images to ram? #115, the server can't deal with storing 400,000 images in memory.

Solution

1. Load images in batches
   • Avoids running out of memory
2. Report the results of each batch classification as a SocketIO update
   • Gives user intermediate results rather than waiting for everything to complete
   • Allows many small responses instead of one huge response

[dcmartin]

I think it would be better to store the results of the classification in an external repository, e.g. Cloudant, indexed by the image submitted and the classifier (job id) utilized. You might also include additional data, e.g. execution seconds, date-time processed, ... In addition, specification of size of result set, would be good (e.g. TOP5 appears to be current default), or even specialization of output for TOP1, e.g. indicating the population statistics for the resulting scores: max (TOP1), min, median, stdev, sum, count.
I implemented a batch script to existing API at ~dcmartin/age-at-home/bin/dodigit