# CSV file is tabular --data_format=tabular # numpy is the default (so this argument is unnecessary here) --data_representation=numpy # Single file --data_files dataset/train_1501_fold1.csv # The CSV file does not include a header. Here, we define the column names --tabular_header_names string binding class # Split the data randomly into 10 folds --data_n_folds=10 --data_fold_split=random # Cross-validation for training and evaluation --data_set_type=holistic-cross-validation # Inputs: string column --data_inputs string # Outputs: binding affinity column --data_outputs binding # Preprocessing # We are using a tokenizer to translate the strings into lists of # integer tokens --tokenizer --tokenizer_max_tokens=24 --tokenizer_split=character --tokenizer_output_sequence_length=30 # Transform the integer tokens into 15-dimensional vectors # (this is a learned transformation) --embedding_dimensions=15