janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-441164957
I have resolved my doubt by initial the probility alphas with 'exp(x)' ,
rather than 'x'. this two method seem achieve the same .but the
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-441164957
I have resolved my doubt by initial the probility alphas with 'exp(x)' ,
rather than 'x'. this two method seem achieve the same .but the
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-441164957
I have resolved my doubt by initial the probility alphas with 'exp(x)' ,
rather than 'x'. the new result is numerical stability, this two
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-441164957
I have resolved my doubt by initial the begian probility alphas with
'exp(x)' , rather than 'x'. the new result is numerical stability, this
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-441164957
I have resolved my doubt by initial the begian probility alphas with 'x' ,
rather than 'exp(x)'. the new result is numerical stability, this
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-443061810
```
def _forward_alg(self, feats):
alphas = [[0.] * self.tagset_size]
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-441164957
I have resolved my doubt by initial the begian probility alphas with
'exp(x)' , rather than 'x'. the new result is numerical stability, this
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-441164957
I have resolved my doubt by initial the begian probility alphas with 'x' ,
rather than 'exp(x)'. the new result is numerical stability, this
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum" and the second one should be
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum" and the second one should be
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum" and the second one should be
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum" and the second one should be
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum" ,and the second one is "log_sum"
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum" in theory .but it may causs a
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum_exp" ,and the second one is
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) , I think the first "log_sum_exp" in the red
pane may be replaced with "sum_exp" ,and the second one is
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) ，but I think the first "log_sum_exp" in the
red pane may be replaced with "sum_exp" ,and the second one is
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) ，but I think the first "log_sum_exp" in the
red pane should be replaced with "sum_exp"
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) ，but I think the first "logsumexp" may be
replaced with "sumexp"
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440304581
it's difficult to compute Z(t) ，but I think the first "logsumexp" should be
replaced with "sumexp"
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440134211
here are the codes
https://github.com/apache/incubator-mxnet/blob/master/example/gluon/lstm_crf/lstm_crf.py
janelu9 edited a comment on issue #13329: problem about lstm_crf
URL:
https://github.com/apache/incubator-mxnet/issues/13329#issuecomment-440133690
Should the " log_sum_exp " in the red pane be replaced by " nd.sum "? this
is my doubt
22 matches
Mail list logo
