Word Vector Apply

The node Word Vector Apply with the option Document Mean flagged, gives me often the error: Invalid shape, Requested INDArray[0,100]....  I haven't yet understood what causes this error.



Hi Francesco

we are aware of this problem. It seems to me that there is a bug in the library. The problem is that if the Word Vector Apply node can't find any matching tokens from an input sentence and the vocabulary of the word vector model this error will be thrown. The cause is that model does not contain all words it was learned on. E.g. if you learn a Doc2Vec model with three labels the resulting model may only contain two is some cases. If you then try to apply this model to a sentence containing only this single word it will fail. Furthermore, the problem seems not to be reliably reproducible. On Windows, it seems to work but on some Linux it fails.