Source code for my senior thesis.
- clone this repo and move to the directory.
$ git clone --recursive https://github.com/wedddy0707/noisyEGG.git
$ cd noisyEGG
- move to EGG directory, git checkout, and install EGG.
$ cd EGG
$ git checkout v1.0
$ pip install --editable .
$ cd - # back to the previous directory
$ cd noisy_env_soft_maxlen
$ python train.py [options]
- Noises
--sender_noise_loc # mean of Gaussian noise on a speaker (sender)
--sender_noise_scale # standard deviation of Gaussian noise on a speaker (sender)
--receiver_noise_loc # mean of Gaussian noise on a listener (receiver)
--receiver_noise_scale # standard deviation of Gaussian noise on a listener (receiver)
--channel_repl_prob # channel replacement probability
- DER (Decayed Entropy Regularizer)
--sender_entropy_coeff # $\lambda_{\mathcal{H}}$ in the paper
--sender_entropy_common_ratio # $\rho_{\mathcal{H}}$ in the paper
- SML (Soft Max Len)
--length_cost # $\lambda_{sml}$ in the paper
--max_len # ${\rm max_len}$ in the paper
--effective_max_len # ${\rm eff_max_len}$ in the paper
For more information, please refer to EGG.