/MissingDataVAEs

Primary LanguageJupyter Notebook

VAEs in the Presence of Missing Data

Code for the paper VAEs in the Presence of Missing Data, Mark Collier, Alfredo Nazabal, Christopher K.I. Williams, 2020.

The notebook VAEs_in_the_Presence_of_Missing_Data.ipynb was used to produce the results in the paper.

In order to run the MNIST MCAR experiments, set:

MISSINGNESS_TYPE = 'MCAR'
DATASET = 'MNIST'
LIKELIHOOD = 'BERNOULLI'

MNIST MNAR:

MISSINGNESS_TYPE = 'MNAR'
DATASET = 'MNIST'
LIKELIHOOD = 'BERNOULLI'

SVHN MCAR:

MISSINGNESS_TYPE = 'MCAR'
DATASET = 'SVHN'
LIKELIHOOD = 'LOGISTIC_MIXTURE'

SVHN MNAR:

MISSINGNESS_TYPE = 'MNAR'
DATASET = 'SVHN'
LIKELIHOOD = 'LOGISTIC_MIXTURE'

Please adjust NUM_RUNS to set the number of replicates to run.