Compare SortaGrad vs Existing Curriculum Learning

[kdavis-mozilla]

No description provided.

[andi4191]

Updating the results:

DataSet Used: TED
Number of GPU instances: 4
Epochs : 10

I Test of Epoch 10 - WER: 0.417148, loss: 72.8541198836, mean edit distance: 0.192530
I --------------------------------------------------------------------------------
I WER: 0.200000, loss: 2.473928, mean edit distance: 0.043478
I - src: "it was a beautiful fish"
I - res: "i was a beautiful fish"
I --------------------------------------------------------------------------------
I WER: 0.250000, loss: 2.241259, mean edit distance: 0.052632
I - src: "if this doesnt work"
I - res: "if this doesnt wore"
I --------------------------------------------------------------------------------
I WER: 0.250000, loss: 2.501267, mean edit distance: 0.125000
I - src: "there is no feed"
I - res: "there e is no feed"
I --------------------------------------------------------------------------------
I WER: 0.333333, loss: 1.873822, mean edit distance: 0.076923
I - src: "no im serious"
I - res: "now im serious"
I --------------------------------------------------------------------------------
I WER: 0.333333, loss: 2.247723, mean edit distance: 0.076923
I - src: "no im serious"
I - res: "now im serious"
I --------------------------------------------------------------------------------
I WER: 0.333333, loss: 2.601931, mean edit distance: 0.111111
I - src: "and i say"
I - res: "and i saw"
I --------------------------------------------------------------------------------
I WER: 1.000000, loss: 1.610738, mean edit distance: 0.500000
I - src: "so"
I - res: "o"
I --------------------------------------------------------------------------------
I WER: 1.000000, loss: 1.964374, mean edit distance: 0.500000
I - src: "so"
I - res: "o"
I --------------------------------------------------------------------------------
I WER: 1.000000, loss: 2.448013, mean edit distance: 0.333333
I - src: "and"
I - res: "aand"
I --------------------------------------------------------------------------------
I WER: 1.000000, loss: 2.463090, mean edit distance: 0.333333
I - src: "and"
I - res: "aand"
I --------------------------------------------------------------------------------

@kdavis-mozilla : Need a code review for code changes as well

[kdavis-mozilla]

@andi4191 Just wondering why a code review is needed?

I'd guess we only need a detailed review if the particular test, SortaGrad in this case, was better than Curriculum Learning and we wanted to merge the test code in to master.

[andi4191]

@kdavis-mozilla : Yeah, Since the training time for 1st epoch took around 5 hrs, I wanted to confirm if the changes made for testing this feature were correct and I did not miss anything.

[kdavis-mozilla]

@andi4191 OK :-)

[kdavis-mozilla]

Closing for lack of activity.

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.