scoutbee / Pytorch Nlp Notebooks
Projects that are alternatives of or similar to Pytorch Nlp Notebooks
A Deep Dive into NLP with PyTorch
Learn how to use PyTorch to solve some common NLP problems with deep learning. View these notebooks on nbviewer.
-
1_BoW_text_classification.ipynb
: Train a bag-of-words model to predict the sentiment of IMDB reviews -
2_embeddings.ipynb
: Play around with different pretrained word embeddings -
3_rnn_text_classification.ipynb
: Train an RNN to predict the sentiment of IMDB movie reviews -
4_character_text_generation.ipynb
: Train a character-level RNN language model to generate weight loss articles -
5_seq2seq_attention_translation.ipynb
: Train an RNN-based Seq2Seq model with attention to translate from English to French -
6_transformer_translation.ipynb
: Train a pure self-attention based transformer Seq2Seq model to translate from English to French -
7_gpt2_finetuned_text_generation.ipynb
: Fine-tune the pretrained (small) GPT-2 model to generate weight loss articles
Tutorials
Events | Dates | Slides |
---|---|---|
PyData London 2019 | 12 Jul 2019 | link |
PyData Cambridge 2019 | 15 Nov 2019 | link |
DS Con Belgrade 2019 | 18 Nov 2019 | link |
Setup
Make sure you have a Google account and visit Google Colab. You should see a list of notebooks pop up:
If you have trouble with that, you can also save the notebook you want to run from this repo to your local filesystem, and then upload it to Google Colab with File -> Open Notebook -> Upload
.
Basic Navigation
You can run cells with <SHIFT> + <ENTER>.
Missing packages
If you find that you are missing a necessary package, you can prepend !
to a bash command. For example, to install googledrivedownloader
, you would run in a cell:
!pip install googledrivedownloader
Using a GPU
To use a GPU (for free!), select from the top menu from Colab Runtime -> Change Runtime Type -> Hardware Accelerator -> GPU
. Pay attention to how much memory the GPU is currently using by clicking Runtime -> Manage Sessions
.
Contributing
Feel free to submit a PR for cleanups, error-fixing, or adding new (relevant) content!