tsabl

Twitter Sentiment Analysis

Setup

1. Make sure you have python 3, pip and virtualenv installed.
2. Go to https://apps.twitter.com/, log in, and create a new app to get a consumer key and a consumer secret.
3. Perform the following steps (remember to insert your key and secret):

git clone https://github.com/frederikgdl/tsabl.git
cd tsabl
virtualenv venv -p python3
echo 'export TSABL_CONSUMER_KEY="<YOUR CONSUMER KEY HERE>"' >> venv/bin/activate
echo 'export TSABL_CONSUMER_SECRET="<YOUR CONSUMER KEY SECRET>"' >> venv/bin/activate
echo 'export PYTHONPATH="$PYTHONPATH:$PWD"' >> venv/bin/activate
echo 'export KERAS_BACKEND=theano' >> venv/bin/activate
source venv/bin/activate
pip install -r requirements.txt

Run

Before running any commands make sure the virtual environment is activated by running:

source venv/bin/activate

Embeddings

• Edit embeddings/config.py to use the data files you want.
• Run python embeddings/main.py to train and save the embeddings using CPU.

Classifiers

• Edit classifiers/config.py to use the data files you want.
• Run python classifiers/train_and_test.py to train and test the models and save the results.

Scripts

• Edit scripts/config.py to use the data files you want.
• Run python scripts/test_all_epochs to test all epochs in selected folder and write results and print graph.

Train embeddings on GPU

Embeddings can be trained on GPU using either Theano or TensorFlow backend.

• Make sure CUDA is installed if using NVIDIA GPU.

Theano

1. Make sure libgpuarray is installed.
2. Edit run_gpu.sh: the KERAS_BACKEND variable must be set to theano and paths to libgpuarray and CUDA must be specified.
3. Run by typing (may need to give permission first):

./run_gpu.sh

TensorFlow

1. Make sure GPU version of TensorFlow is installed.
2. Edit run_gpu.sh: the KERAS_BACKEND variable must be set to tensorflow.
3. Run by typing (may need to give permission first):

./run_gpu.sh

Tests

python -m unittest