|
|
|
@@ -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,). |
|
|
|
|