https://github.com/danieleds/qc_with_untrained_recurrent_embeddings
Question Classification with Untrained Recurrent Embeddings
https://github.com/danieleds/qc_with_untrained_recurrent_embeddings
echo-state-networks machine-learning neural-networks question-classification reservoir-computing
Last synced: 12 months ago
JSON representation
Question Classification with Untrained Recurrent Embeddings
- Host: GitHub
- URL: https://github.com/danieleds/qc_with_untrained_recurrent_embeddings
- Owner: danieleds
- Created: 2019-05-22T09:14:53.000Z (about 7 years ago)
- Default Branch: master
- Last Pushed: 2022-10-26T16:56:53.000Z (over 3 years ago)
- Last Synced: 2024-10-18T21:56:27.985Z (almost 2 years ago)
- Topics: echo-state-networks, machine-learning, neural-networks, question-classification, reservoir-computing
- Language: Python
- Homepage:
- Size: 29.3 KB
- Stars: 3
- Watchers: 2
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
This repository contains the code for the following paper:
"[Question Classification with Untrained Recurrent Embeddings](http://for.unipi.it/daniele_di_sarli/files/2019/11/paper_54.pdf)",
D. Di Sarli, C. Gallicchio, A. Micheli,
AI*IA 2019 – Advances in Artificial Intelligence.
## Running
The folder structure is organized on two levels.
At the top level we have the dataset (QC), and inside that folder
we have the models applied to that dataset.
To launch an experiment, for example a leaky ESN on the QC task, run this command
from the root of the project:
$ python3 QC/leaky_esn # ...args...
## Reproducing results
To reproduce the results in the paper, just run the following commands:
$ python3 QC/leaky_esn --searches 0 --final-trials 10
$ python3 QC/leaky_esn_ensemble --searches 0 --final-trials 10
$ python3 QC/leaky_esn_attn --searches 0 --final-trials 10
$ python3 QC/mygru --searches 0 --final-trials 10
## Hardware
It is advisable to run the code on a GPU with at least 16GB of memory. Alternatively, the code
can easily be modified to accumulate the gradients of smaller minibatches in the training loop.