You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I find batchseq quite useful to reshape data in preparation for training a recurrent model in Flux.
Unfortunately, looks like batchseq only works for sequences of vectors (i.e.1-feature inputs) not sequences of matrices (multi-feature inputs). Consider
# ex.jlusing MLUtils: batchseq
seqlen =2
seq_a = [[1, 2, 3] for i in1:seqlen]
@show seq_a
batched_seq_a =batchseq(seq_a)
@show batched_seq_a
seq_B = [[123; 456] for i in1:seqlen]
@show seq_B
batched_seq_B =batchseq(seq_B)
Is the desired behavior the following?
Given an input list xs containing B arrays of size * x Li, where * is common to all arrays and Li can vary,
return a list of length Lmax where the elements have size * x B.
I find
batchseq
quite useful to reshape data in preparation for training a recurrent model in Flux.Unfortunately, looks like
batchseq
only works for sequences of vectors (i.e.1-feature inputs) not sequences of matrices (multi-feature inputs). Considerwith
MLUtils@v0.2.10
this givesclick here to see results
batchseq
work for sequences of matrices something in the plans forMLUtils
?Thank you very much!
The text was updated successfully, but these errors were encountered: