Alan Bootstrap

Active Learning

Setup Environment

1. Create a Conda Environment for Python 3.7.3:

conda create --name <EVN_NAME> python=3.7.3

2. Git Clone:

git clone https://<USR_NAME>@bitbucket.org/rbcmllab/data_bootstrap.git

3. Install Python Dependencies

cd data_bootstrap/alanbootstrap
pip install -r requirements.txt
pip install alanbootstrap

Run

> python bin/ast_al_bin.py -h

usage: ast_al_bin.py [-h] -i INIT -p POOL -o OUTPUT [-d DEVSET]

Parameters for AST classification active learner. The model will be persisted
into the output folder at every iteration. If a dev dataset was provided, the
classification accuracy score would be calculated on this dataset at each
iteration and a performance plot would be saved into the output folder.

optional arguments:
  -h, --help            show this help message and exit
  -i INIT, --init INIT  Shared-AST json file path; ASTs in this file will be
                        used to initialize the active learner. Every AST in
                        this file must contain a target value and a list of
                        complexity_features.
  -p POOL, --pool POOL  Shared-AST json file path; Active learner will sample
                        ASTs from this file Every AST in this file must
                        contain a list of complexity_features.
  -o OUTPUT, --output OUTPUT
                        Absolute path of the model persisting folder.
  -d DEVSET, --devset DEVSET
                        Shared-AST json file path; ASTs in this file will be
                        used to evaluate the active learner. Every AST in this
                        file must contain a target value and a list of
                        complexity_features.

Example:

> python bin/ast_al_bin.py \
-i $DATA_DIR/ast_train.json \
-p $DATA_DIR/ast_pool.json \
-o $DATA_DIR \
-d $DATA_DIR/ast_dev.json