DCASE2020-Task1

Jupyter notebooks for DCASE 2020 challenge Task 1

Training

Use create_extra_labels.ipynb to create new data labels for domain adaptation (Task 1a only).

Run DCASE2020_Task1x_training.ipynb to train the models.

The following block of script is used to save the model after training:

stamp = datetime.datetime.now().strftime('%y%m%d%H%M')
tag = stamp + '_' + WhichTask + '_' + MODE + '_'+ str(num_epochs)
savedir = os.path.join(os.getcwd(), tag)
print "Model path: %s" % savedir
try:
    os.makedirs(savedir)
except OSError:
    if not os.path.isdir(savedir):
        raise

Save checkpoints by specifying the weights of which epochs to be saved:

ckpt = ckpt(filepath=ckpt_path, ckpts=[70, 150])

The class ckpt is defined in DCASE_training_functions.py

Test

Run DCASE2020_Task1a_inference.ipynb to test the model and to print out the confusion matrices