kaldi-asr/kaldi

nnet3 : Reduce unnecessary outputs (file or email) during training

Open

#1,349 opened on Jan 17, 2017

View on GitHub
 (2 comments) (0 reactions) (0 assignees)Shell (15,392 stars) (5,359 forks)batch import
enhancementhelp wantednnet3stale

Description

(Created to ensure we keep track of a low priority issue.)

Status update emails are sent after every iteration if an email id is specified during training. However status information uses output of train/validation accuracy/log-probability computation jobs which are run in the background. In many cases these might be very slow to compute, depending on the model, as these are run on the CPU. This would mean that there are uninformative status mails which do not provide any info other than training time.

e.g.

 %Iter   duration        train_loss      valid_loss      difference
Total training time is 0:59:53

Contributor guide

Tech stack
cppshell
Domain
machine learning
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
Understanding of Kaldi training scriptsBasic shell scripting
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40
Research direction
Investigate the training scripts in 'steps/nnet3/' to understand how status emails are generated. Look for the logic that decides whether to send email and how the validation accuracy is computed. Consider adding a condition to skip sending if validation results are not ready or are uninformative. Check existing comments or related issues for previous discussion.
nnet3 : Reduce unnecessary outputs (file or email) during training · kaldi-asr/kaldi#1349 | Good First Issue