sentiment-classification

Results

Algorithm	Accuracy
CNN	84.59%
CNN-SBN-ST	86.44%
CNN-DBN-ST	87.54%

Prerequisite

- Anaconda (python 3.x)

Instructions to run

- First setup conda environment: run "conda env create -f sentclass.yml"
- Activate the environment with "conda activate sentclass"
- Run the main.py with python main.py --arguments

Arguments specification

- epochs [Amount of epochs to train the model]
- embed_dim [Size of the the embeddings used in the model]
- batch_size [Size of the batch to use during training]
- n_filters [Number of filters/output channels]
- filter_sizes [A comma separated string of filter sizes, such as "2,3,4,5"]
- output_dim [The number of outputs, which is 1 in the case of the sentiment classification task]
- dropout_rate [The probability to which dropout should be applied]
- binary_neuron [Type of binary neuron to use, has to be D-ST or S-ST]

Example

To run the program for 10 epochs with an embeddings dimension of 100 the batch size set to 64 and the D-ST trained Binary Neuron 
run "python main.py --epochs 10 --embed_dim 100 --batch_size 64 --binary_neuron D-ST