snipsco / ntm-lasagne

Neural Turing Machines library in Theano with Lasagne
https://medium.com/snips-ai/ntm-lasagne-a-library-for-neural-turing-machines-in-lasagne-2cdce6837315#.63t84s5r5
MIT License
300 stars 51 forks source link

examples are not working #30

Open mnemonicker opened 8 years ago

mnemonicker commented 8 years ago

Examples are not working. Python stops at this line: train_fn = theano.function([input_var, target_var], loss, updates=updates)

Error is the following (this is from cop-task.py): TypeError: (‘The following error happened while compiling the node’, forall_inplace,cpu,grad_of_scan_fn}(Shape_i{1}.0, Elemwise{sqr,no_inplace}.0, InplaceDimShuffle{0,1,3,2}.0, InplaceDimShuffle{0,1,3,2}.0, InplaceDimShuffle{0,1,2,3,x}.0, Subtensor{int64:int64:int64}.0, Alloc.0, Subtensor{int64:int64:int64}.0, Subtensor{int64:int64:int64}.0, Subtensor{int64:int64:int64}.0, Subtensor{int64:int64:int64}.0, Subtensor{int64:int64:int64}.0, Alloc.0, Subtensor{::int64}.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Alloc.0, Shape_i{1}.0, Shape_i{1}.0, Shape_i{1}.0, controller.W_in_to_hid, controller.W_reads_to_hid, InplaceDimShuffle{x,0}.0, InplaceDimShuffle{x,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, InplaceDimShuffle{x,0,1}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, MakeVector{dtype=’int64'}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, Reshape{2}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0, InplaceDimShuffle{1,0}.0), ‘\n’, ‘Non-unit value on shape on a broadcastable dimension.’, (0L, 0L, 0L), (False, True, False), ‘Container name “None”’)

The problem is about "updates"

tristandeleu commented 8 years ago

This looks a lot like #28 . I think that something changed with the bleeding edge version of Theano. Try to uninstall Theano and install the pip version (pip install Theano) like suggested in https://github.com/snipsco/ntm-lasagne/issues/28#issuecomment-238518018.

mnemonicker commented 8 years ago

I tried the pip version of thenao, still have the same problem..

andreasrobinson commented 7 years ago

I am getting the same error as the above user when I try to run the copy-task example.

andreasrobinson commented 7 years ago

I tried downgrading theano (per the suggestion above), rather than using the version pulled by the requirements.txt, but it still gave the same error (as w/ user mnemonicker). But then after I downgraded lasagne as well, it now seems to be working.

s3xton commented 7 years ago

Can you tell me which versions of each you are using? In the setup I have the dyck words example is producing NaN and the associative recall task is failing on train_fn = theano.function([input_var, target_var], loss, updates=updates) similar to above.

andreasrobinson commented 7 years ago

These are the versions that worked for me, but the only example I've tried so far is "copy". theano: 0.8.2 lasagne: 0.1

s3xton commented 7 years ago

I've tried running the examples with several different version combinations of theano and lasagne and copy has been the only one to work so far. Are the other examples working for you, or just the copy?