Closed HunterHantao closed 7 years ago
Closed. After I revise all the float32 to float64
Hi, thanks for your interest in our project, and I'm glad you've found a solution.
For speed considerations, I would however recommend setting floatx
in keras.json
or .theanorc
to float32
instead. Hope this helps.
I try to run this code with command as below python2.7 tsnet_cli.py -d mnist -n conv:0/16 relu:0 flat:0 sfmx:0/10 -e 10 -lrnparam 1e-3 1e-3 0.9 -v 1
Please suggest me how to fix it. Thanks a lot.
Traceback (most recent call last): File "tsnet_cli.py", line 4, in
run(sys.argv[1:])
File "/home/kowalski/tensor/tsnet/tsnet/launcher.py", line 38, in run
hst = net.fit(dataset, settings)
File "/home/kowalski/tensor/tsnet/tsnet/core_numpy/network.py", line 218, in fit
ly = th-4; trn += [process(XT, YT )]; self.solve()
File "/home/kowalski/tensor/tsnet/tsnet/core_numpy/network.py", line 195, in process
rep = self.backward(Yb).update(settings.lrnalg, settings.lrnparam) if trn else None
File "/home/kowalski/tensor/tsnet/tsnet/core_numpy/network.py", line 100, in backward
for B in self.blocks[::-1]: Y = B.backward(Y)
File "/home/kowalski/tensor/tsnet/tsnet/core_numpy/network.py", line 43, in backward
for L in self.layers[::-1]: Y = L.backward(Y, mode='XG') if Y is not None else Y
File "/home/kowalski/tensor/tsnet/tsnet/core_numpy/layers.py", line 156, in backward
O = unexpand(O)
File "/home/kowalski/tensor/tsnet/tsnet/core_numpy/layers.py", line 73, in unexpand
for y, x in product(xrange(T.shape[4]), xrange(T.shape[5])): neadd(O[:,:,y:y+T.shape[2],x:x+T.shape[3]], T[:,:,:,:,y,x])
File "/home/kowalski/tensor/tsnet/tsnet/core_numpy/layers.py", line 12, in neadd
def neadd(Y, X): ne.evaluate('Y + X', out=Y)
File "/usr/local/lib/python2.7/dist-packages/numexpr/necompiler.py", line 802, in evaluate
return compiled_ex(*arguments, **kwargs)
TypeError: Iterator requested dtype could not be cast from dtype('float64') to dtype('float32'), the operand 0 dtype, according to the rule 'safe'