amansinha09 / bio-challenge

Team : Biobot

bio-challenge

Competition : Biocreative VII Official Website

Team : Biobot

Track 3: Automatic extraction of medication names in tweets

TODOs

• Build corpora from SMM4H18 trainset
• Make evaluation code ready
• Make src ready from notebook
• add gpu running feature
• Put it on grid
• Add inference code
• Add testing code
• Run initial baseline char-rnn - only biotrainset
• Birnn- attention-char
• Create word-rnn model
• Testing word model working
• Rnn- word
• Bi-rnn -word
• Bi-rnn attention -word
• Bert /scibert ---> to be resolved
• resolved bert inference
• word-char model
• Crf
• Custom loss w/ best char model
• Custom loss w/ best word model
• Use different/custom losses
• Think augmenation
• Add requirements.txt

Instructions

Getting Started

Clone the repository : git clone https://github.com/amansinha09/bio-challenge.git Create a virtual environment (venv) with requirements.txt using the commans below:

python3 -m venv .venv
. .venv/bin/activate
pip3 install --upgrade pip
cd bio-challenge
pip3 install -r requirements.txt

Directory Structure

bio-challenge/
	|-- data/
	|-- src/
		|-- data.py
		[-- utils.py
		|-- evaluation.py
		|-- model.py
		|-- run.py
	|-- ref/
	|-- res/
	|-- .logs/
	|-- .model/
	|--run_exp.sh
	|-- README.md

For running experiment

mkdir ref/ res/ .model/ .logs/
cd .logs/
sh ../run_exp.sh

Change the parameters for custom experiments :

usage: run.py [-h] [--device DEVICE] [--hs HS] [--epochs EPOCHS] [--bs BS]
              [--nl NL] [--bidir] [--inplen INPLEN] [--inpsize INPSIZE]
              [--vocabsize VOCABSIZE] [--lr LR] [--test_every TEST_EVERY]
              [--save_dir SAVE_DIR] [--model_id MODEL_ID] [--save_preds]
              [--save_model] [--stop_early]

Running ner...

optional arguments:
  -h, --help            show this help message and exit
  --device DEVICE       cpu or gpu
  --hs HS               Hidden layer size
  --epochs EPOCHS       Number of epochs
  --bs BS               batch size
  --nl NL               Number of layers
  --bidir               bi-directional
  --inplen INPLEN       sequence/sentence length
  --inpsize INPSIZE     embedding size
  --vocabsize VOCABSIZE
                        vocab size
  --lr LR               Learning rate of loss optimization
  --test_every TEST_EVERY
                        Testing after steps
  --save_dir SAVE_DIR   Model dir
  --model_id MODEL_ID   model name identifier
  --save_preds          whether to save the preditions
  --save_model          whether to save the model
  --stop_early          whether to use early stopping

For evalution

Add the prediction file into the res/ folder and test file in the ref/ folder, then run the evalution script: python src/evaluation.py ref/<gold_file> res/<pred_file>