BUG-Fixed: [CT][MS][Document] correct the description of the operator sequence_mask

pull/10733/head
hedongdong 5 years ago
parent e817019141
commit f78df961fe

@ -113,11 +113,12 @@ def sequence_mask(lengths, maxlen):
[d_1, d_2, ..., d_n, maxlen], with mask[i_1, i_2, ..., i_n, j] = (j < lengths[i_1, i_2, ..., i_n])
Inputs:
- **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor must be
less than or equal to `maxlen`. Must be type int32 or int64.
- **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor should be
less than or equal to `maxlen`. Values greater than `maxlen` will be treated as `maxlen`.
Must be type int32 or int64.
- **maxlen** (int) - size of the last dimension of returned tensor. Must be positive and same
type as elements in `lengths`. Default is the maximum value in lengths.
type as elements in `lengths`.
Outputs:
One mask tensor of shape lengths.shape + (maxlen,).

@ -689,11 +689,12 @@ class SequenceMask(PrimitiveWithCheck):
[d_1, d_2, ..., d_n, maxlen], with mask[i_1, i_2, ..., i_n, j] = (j < lengths[i_1, i_2, ..., i_n])
Inputs:
- **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor must be
less than or equal to `maxlen`. Must be type int32 or int64.
- **lengths** (Tensor) - Tensor to calculate the mask for. All values in this tensor should be
less than or equal to `maxlen`. Values greater than `maxlen` will be treated as `maxlen`.
Must be type int32 or int64.
- **maxlen** (int) - size of the last dimension of returned tensor. Must be positive and same
type as elements in `lengths`. Default is the maximum value in lengths.
type as elements in `lengths`.
Outputs:
One mask tensor of shape lengths.shape + (maxlen,).

Loading…
Cancel
Save