• matplotlib
• keras
• tensorflow==1.14
• tensorflow_probability=0.7.0

The main scripts used are TMC.py and IWAE_forward.py, the results presented in the report were obtained by running these two scripts. The hyper parameter search can be done by configuring TMC_hyper_param_search.py and using the makefile in a similar manner. In both TMC.py and IWAE_forward.py there is the option to restore a model from a previously saved checkpoint and plot reconstructed data. To do this set restore_and_recon=True in the scripts. To train regularly set it to False.

Examples of how to run these two files:

python3 TMC.py -k 20 --epochs 400 --batch_size 128 --model_type small
python3 TMC.py -k 20 --epochs 400 --batch_size 128 --model_type large

python3 IWAE_forward.py -k 20 --epochs 400 --batch_size 128 --model_type small
python3 IWAE_forward.py -k 20 --epochs 400 --batch_size 128 --model_type large

The hyper parameter search can be done by configuring TMC_hyper_param_search.py or running the makefile with

make tmc-hyper-param-srch EPOCHS=400 BATCHSIZE=128 MODELTYPE=small FILE=models/IWAE_forward.py

See LICENCE.txt file.

https://doi.org/10.5281/zenodo.3707783

@software{linus_nilsson_2020_3707783,
  author       = {Linus Nilsson and
                  Martin Larsson and
                  Oskar Kviman},
  title        = {TMC - Tensor Monte Carlo - Reproducibility Report},
  month        = mar,
  year         = 2020,
  publisher    = {Zenodo},
  version      = {v1.0.0},
  doi          = {10.5281/zenodo.3707783},
  url          = {https://doi.org/10.5281/zenodo.3707783}
}