README.md

AI System for Primer Design

Autoencoder

1. Train encder and decoder.

train.py [-h] [-k K_FOLDS]
        [-e NUM_EPOCHS]
        [-b BATCH_SIZE]
        [-l LEARNING_RATE]
        [--embedding_dim EMBEDDING_DIM]
        [--hidden_dim HIDDEN_DIM]
        [--plot_every PLOT_EVERY]
        [--data_path DATA_PATH]
        [--word_dict WORD_DICT]
        [--debug DEBUG

2. Regress CT values.

regress.py [-h] [-k K_FOLDS]
          [-e NUM_EPOCHS]
          [-b BATCH_SIZE]
          [-l LEARNING_RATE]
          [--embedding_dim EMBEDDING_DIM]
          [--hidden_dim HIDDEN_DIM]
          [--plot_every PLOT_EVERY]
          [--data_path DATA_PATH]
          [--word_dict WORD_DICT]
          [--debug DEBUG]

3. Inference CT values.

inference.py [-h] [-k K_FOLDS]
            [-b BATCH_SIZE]
            [--embedding_dim EMBEDDING_DIM]
            [--hidden_dim HIDDEN_DIM]
            [--data_path DATA_PATH]
            [--word_dict WORD_DICT]
            [--debug DEBUG]

Multi-input CNN

Two-step classifier-regressor

0. Generate binary 'label' for data with NaN
1. Train classifier

   python train_cnn_multi.py \
       --data_path='./data/train_df_with_label.csv' \
       --loss_function='bce_loss' \
       -e 1000 \
       --target_name='label' \
       --patience 10
   

2. Inference with classifier (you may use '--model_path') -> produce 'train/test_df_wtih_label_no_nan.csv'

   python inference_cnn_classifier.py \
       --data_path='./data/train_df_with_label.csv' \
       --target_name='label'    
   python inference_cnn_classifier.py \
       --data_path='./data/test_df_with_label.csv' \
       --target_name='label'    
   

3. Train regressor without NaN (predicted) data

   python train_cnn_multi.py --data_path='./data/train_df_with_label_no_nan.csv' \
       -e 1000 \
       --patience 20
   

4. Inference with regressor on train set for qualitative analysis (you may use '--model_path')

   python inference_cnn_multi.py --data_path='./data/train_df_with_label_no_nan.csv'
   

5. Inference with regressor on test set for predicting ct value (you may use '--model_path')

   python inference_cnn_multi.py --data_path='./data/test_df_with_label_no_nan.csv'