Evaluator¶

class kospeech.evaluator.evaluator.Evaluator(dataset, vocab, batch_size=1, device=None, num_workers=1, print_every=100, decode='greedy', beam_size=None)[source]¶

Class to evaluate models with given datasets.

Parameters

dataset (kospeech.data.data_loader.SpectrogramDataset) – dataset for spectrogram & script matching
batch_size (int) – size of batch. recommended batch size is 1.
device (torch.device) – device - ‘cuda’ or ‘cpu’
num_workers (int) – the number of cpu cores used
print_every (int) – to determine whether to store training progress every N timesteps (default: 10)

evaluate(model)[source]¶: Evaluate a model on given dataset and return performance.