Geno to Pheno Repo

Quick explainer for the fodler structure:

• cur_* files are model tensorflow definitions, some of them outdated/experimental
  • VAE specific files are:
    • cur_encoder: encoder model
    • cur_decoder: decoder model
    • cur_autoencoder: model stiching together encoder + decoder, additionally defines train/test_step
    • custom_lr_on_callback: custom implementation of ReduceLROnPlateau with additional start_epoch parameter
    • cur_helpers: helper functions such as custom model.fit function, metric visualisations and loss function
    • var_autoencoder_parental.ipynb: notebook where actual VAE training & data extraction takes place
    • data/var_autoencoder/: Different VAE implementation backups (definitions, weights & data used for training)
  • Geno to trait specific files are:
    • geno_to_trait_model.py: tf model definitions
    • geno_to_trait_helpers.py: helper functions such as sampleing layers and functions used to record metrics
    • geno_to_trait.ipynb: notebook where actual training & data extraction takes place
  • reg VAE specific files are:
    • reg_vae_helpers.py: helper functions such as custom model.fit function, metric visualisations and loss function
    • cur_encoder.py: encoder model
    • cur_decoder.py: decoder model
    • reg_autoencoder.py: trash, to be deleted
    • geno_to_trait_model.py: trait prediction model
    • regression_vae.py: model stiching together encoder + decoder + geno_to_trait, additionally defines train/test_step
    • data/reg_vae/: Different reg VAE implementation backups (definitions, weights & data used for training)

Other folders than those mentioned previously in /data/ mostly contain raw data files, actual data used for training are in the .csv files in /data. Note that .keras weight files are to large for github, therefore i cant upload them :(