run_instructions.txt

If you wish to train additional models beyond the default --model_dir to `train.py`.
  I.e., `python3 train.py --model_dir experiments/{exp_name}/`
  This directory should contain a `params.json` file with the following attributes:
    "model_name" : Model name
    "model_type" : Model type, which model inputs to use
      Choices:
        "guf" : just visual branch,
        "combined" : visual branch plus one set of embeddings
        "all_three" : visual, doc2vec, and graph2vec
        "embeddings" : just either doc2vec or graph2vec
    "use_graph" : If true, uses graph2vec, else use doc2vec when there is a choice
    "lr" : learning rate
    "b1" : beta 1 parameter for Adam optimizer
    "b2" : beta 2 parameter for Adam optimizer
    "batch_size" : batch size
    "val_split" : percent to hold out for validation
    "num_epochs" : number of epochs of training
    "sigmoid_out" : clamp output of models with sigmoid
    "eval_every" : number of epochs to run training evaluation
    "plot_preds" : plot predictions at end of validation evaluation
    "visual_emb_size" : size of embedding generated by visual branch
    "conv_activation" : activation function between convolution layers
      Choices: "relu", "none", "sigmoid"
    "countries" : Which countries to run training split over