updated file reading

[emilyzfliu]

i used the code from here https://github.com/corticometrics/neuroimage-tensorflow/blob/master/genTFrecord.py to generate tf record files and modified it slightly so it also returns the input shape of the file. not entirely sure what exactly the input shape is going to be though + need to speed up the code

[satra]

@emilyzfliu - this looks like it contains all the changes. should i close the other PRs. technically you don't need to send new PRs, you can just push commits to one PR. PRs are used primarily to group commits for different components (new features, fixes, etc.,.)

[satra]

instead of the code from there, can you use nobrainer code itself. since we will eventually want to merge these changes into the nobrainer library.

[satra]

the next thing is to add a test script. assuming you can run the server in one terminal, can you run a test script that mimics two clients talking to the server in sequence:

client 1: get model from server (should be empty since no model has yet been sent to server)
client 1: train model on dataset1
client 1: send model to server (let's call this model1)

client 2: get updated model from server (check downloaded model == model1)
client 2: train model on dataset2
client 2: send model to server (model 2)

client 1: get updated model from server (check download model == distributed_consolidation(model1 + model2) )

@Aakanksha-Rana - can you please work with @emilyzfliu to setup this full evaluation?