|
Nabu-asr
|
contains functions to compute the training loss More...
Functions | |
| def | nabu.neuralnetworks.trainers.loss_functions.factory (loss_function) |
| factory method for the loss function More... | |
| def | nabu.neuralnetworks.trainers.loss_functions.marigin_loss (targets, logits, logit_seq_length, target_seq_length) |
| marigin loss More... | |
| def | nabu.neuralnetworks.trainers.loss_functions.cross_entropy (targets, logits, seq_length) |
| compute the cross entropy for all sequences in the batch More... | |
| def | nabu.neuralnetworks.trainers.loss_functions.sigmoid_cross_entropy (targets, logits, seq_length) |
| compute the sigmnoid cross entropy for all sequences in the batch More... | |
| def | nabu.neuralnetworks.trainers.loss_functions.sum_cross_entropy (targets, logits, logit_seq_length, target_seq_length) |
| cross entropy summed over timesteps | |
| def | nabu.neuralnetworks.trainers.loss_functions.average_cross_entropy (targets, logits, logit_seq_length, target_seq_length) |
| cross entropy averaged over timesteps | |
| def | nabu.neuralnetworks.trainers.loss_functions.average_sigmoid_cross_entropy (targets, logits, logit_seq_length, target_seq_length) |
| sigmoid cross entropy averaged over timesteps | |
| def | nabu.neuralnetworks.trainers.loss_functions.CTC (targets, logits, logit_seq_length, target_seq_length) |
| CTC loss. More... | |
contains functions to compute the training loss
| def nabu.neuralnetworks.trainers.loss_functions.cross_entropy | ( | targets, | |
| logits, | |||
| seq_length | |||
| ) |
compute the cross entropy for all sequences in the batch
| targets | a dictionary of [batch_size x time x ...] tensor containing the targets |
| logits | a dictionary of [batch_size x time x ...] tensor containing the logits |
| seq_length | a dictionary of [batch_size] vectors containing the sequence lengths |
| def nabu.neuralnetworks.trainers.loss_functions.CTC | ( | targets, | |
| logits, | |||
| logit_seq_length, | |||
| target_seq_length | |||
| ) |
CTC loss.
| targets | a dictionary of [batch_size x time x ...] tensor containing the targets |
| logits | a dictionary of [batch_size x time x ...] tensor containing the logits |
| logit_seq_length | a dictionary of [batch_size] vectors containing the logit sequence lengths |
| target_seq_length | a dictionary of [batch_size] vectors containing the target sequence lengths |
| def nabu.neuralnetworks.trainers.loss_functions.factory | ( | loss_function | ) |
factory method for the loss function
| loss_function | the required loss function |
| def nabu.neuralnetworks.trainers.loss_functions.marigin_loss | ( | targets, | |
| logits, | |||
| logit_seq_length, | |||
| target_seq_length | |||
| ) |
marigin loss
| targets | a dictionary of [batch_size x time x ...] tensor containing the targets |
| logits | a dictionary of [batch_size x time x ...] tensor containing the logits |
| logit_seq_length | a dictionary of [batch_size] vectors containing the logit sequence lengths |
| target_seq_length | a dictionary of [batch_size] vectors containing the target sequence lengths |
| def nabu.neuralnetworks.trainers.loss_functions.sigmoid_cross_entropy | ( | targets, | |
| logits, | |||
| seq_length | |||
| ) |
compute the sigmnoid cross entropy for all sequences in the batch
| targets | a dictionary of [batch_size x time x ...] tensor containing the targets |
| logits | a dictionary of [batch_size x time x ...] tensor containing the logits |
| seq_length | a dictionary of [batch_size] vectors containing the sequence lengths |
1.8.12