jcjohnson/torch-rnn

Large data sets for training and select of best options

Open

#166 opened on Mar 4, 2017

View on GitHub
 (2 comments) (4 reactions) (0 assignees)Lua (509 forks)github user discovery
help wanted

Repository metrics

Stars
 (2,557 stars)
PR merge metrics
 (No merged PRs in 30d)

Description

Hi all!

Maybe someone will be interested in getting large clean sets of data to test the training of the torch-rnn? And certainly there will be those who can also share their data sets.

I parsed and cleaned the news headlines from the WSJ.com (English text) and Lenta.ru (Russian text). All this is in the repository and is available for download to everyone: https://github.com/saippuakauppias/training-data

At the moment meaningful texts are obtained using the following settings: -num_layers 4 -rnn_size 850 -dropout 0.1 -batch_size 200 -seq_length 200

I propose to discuss and choose the best settings for these (and other) data sets to get the best results. How do you look at this?

Contributor guide

Research direction
Try the suggested settings (e.g., num layers 4 rnn size 850) on the provided datasets and compare results. Discuss alternative hyperparameters.
Tech stack
lua
Domain
machine learningai
Issue type
Research
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
1
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Under 1 hour
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
torchlua
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
90