Skip to main content Link Menu Expand (external link) Document Search Copy Copied

Use Custom-Trained Model

After training your own model you can use it to generate predictions by passing the model checkpoint path to the script as follows:

python --pre-processed-dir <pre-processed-dir> --predictions-dir <predictions-dir> --model-checkpoint-path <checkpoint-dir>    

If you change the default tuning parameters during training (e.g., bi-lstm-window-size), you also need to set the same values for by using the respective directives (e.g., --bi-lstm-window-size).

Complete usage details of make_predictions.pu script with all overiding configuration values are as follows:

usage: [-h] --pre-processed-dir PRE_PROCESSED_DIR
                       [--predictions-dir PREDICTIONS_DIR] [--no-segment]
                       [--model-checkpoint-path MODEL_CHECKPOINT_PATH]
                       [--cnn-window-size CNN_WINDOW_SIZE]
                       [--bi-lstm-window-size BI_LSTM_WINDOW_SIZE]
                       [--down-sample-frequency DOWN_SAMPLE_FREQUENCY]
                       [--gt3x-frequency GT3X_FREQUENCY]
                       [--activpal-label-map ACTIVPAL_LABEL_MAP]
                       [--silent] [--padding {drop,zero,wrap}]

Argument parser for generating model predictions.

required arguments:
--pre-processed-dir PRE_PROCESSED_DIR
                        Pre-processed data directory

optional arguments:
-h, --help            show this help message and exit
                        Pre-trained prediction model name (default:
--predictions-dir PREDICTIONS_DIR
                        Predictions output directory (default: ./predictions)
--no-segment          Do not output segment number
--output-label        Whether to output the actual label
--model-checkpoint-path MODEL_CHECKPOINT_PATH
                        Path where the custom trained model checkpoint is
--cnn-window-size CNN_WINDOW_SIZE
                        CNN window size of the model in seconds on which the
                        predictions to be made (default: 10).
--bi-lstm-window-size BI_LSTM_WINDOW_SIZE
                        BiLSTM window size in minutes (default: 7).
--down-sample-frequency DOWN_SAMPLE_FREQUENCY
                        Downsample frequency in Hz for GT3X data (default:
--gt3x-frequency GT3X_FREQUENCY
                        GT3X device frequency in Hz (default: 30)
--activpal-label-map ACTIVPAL_LABEL_MAP
                        ActivPal label vocabulary (default: {"sitting": 0,
                        "not-sitting": 1, "no-label": -1})
--silent              Whether to hide info messages
--padding {drop,zero,wrap}
                Padding scheme for the last part of data that does not
                fill a whole lstm window (default: drop)