npow / memnn Goto Github PK

@npow Thank you.

https://github.com/npow/MemNN/blob/master/main.py#L138
print "i: ", i, " nwrong: ", n_wrong

https://github.com/npow/MemNN/blob/master/main.py#L140
refs = line['refs']

Hi! Thanks for sharing this. I tried to run this, but got some errors. I'm using Theano 0.7.0 and scikit-learn 0.16.1.

I have had success with other Theano toolkits, and the data files are correctly loaded as far as I can tell.

After printing the error below, instead of exiting the program it continues running and printing this message every now and then.

kyle:MemNN kyle$ python main.py 
Using gpu device 0: GeForce GT 750M
args:  Namespace(embedding_size=50, gamma=1, lr=0.1, n_epochs=10, task=1, test_file='', train_file='')
********************************************************************************
epoch:  0
/usr/local/lib/python2.7/site-packages/theano/scan_module/scan_perform_ext.py:133: RuntimeWarning: numpy.ndarray size changed, may indicate binary incompatibility
  from scan_perform.scan_perform import *
<<!! BUG IN FGRAPH.REPLACE OR A LISTENER !!>> <type 'exceptions.TypeError'> ('The type of the replacement must be compatible with the type of the original Variable.', Sum{acc_dtype=float64}.0, HostFromGpu.0, TensorType(float64, scalar), TensorType(float32, scalar), 'local_gpu_careduce') local_gpu_careduce
ERROR (theano.gof.opt): Optimization failure due to: local_gpu_careduce
ERROR (theano.gof.opt): TRACEBACK:
ERROR (theano.gof.opt): Traceback (most recent call last):
  File "/usr/local/lib/python2.7/site-packages/theano/gof/opt.py", line 1527, in process_node
    fgraph.replace_all_validate(repl_pairs, reason=lopt)
  File "/usr/local/lib/python2.7/site-packages/theano/gof/toolbox.py", line 259, in replace_all_validate
    fgraph.replace(r, new_r, reason=reason, verbose=False)
  File "/usr/local/lib/python2.7/site-packages/theano/gof/fg.py", line 474, in replace
    str(reason))
TypeError: ('The type of the replacement must be compatible with the type of the original Variable.', Sum{acc_dtype=float64}.0, HostFromGpu.0, TensorType(float64, scalar), TensorType(float32, scalar), 'local_gpu_careduce')

This looks really interesting, could you post the data files so I can run it locally?

I run on a intel i5820 CPU about 10 days, not done yet

/home/work/.jumbo/lib/python2.7/site-packages/sklearn/externals/joblib/_multiprocessing_helpers.py:29: UserWarning: This platform lacks a functioning sem_open implementation, therefore, the required synchronization primitives needed will not function, see issue 3770.. joblib will operate in serial mode
warnings.warn('%s. joblib will operate in serial mode' % (e,))
/home/work/.jumbo/lib/python2.7/site-packages/theano/scan_module/scan_perform_ext.py:85: RuntimeWarning: numpy.ndarray size changed, may indicate binary incompatibility
from scan_perform.scan_perform import *
Traceback (most recent call last):
File "main.py", line 245, in
main()
File "main.py", line 242, in main
model.train(args.n_epochs)
File "main.py", line 153, in train
self.create_train(lenW, len(f))
File "main.py", line 118, in create_train
updates=updates)
File "/home/work/.jumbo/lib/python2.7/site-packages/theano/compile/function.py", line 223, in function
profile=profile)
File "/home/work/.jumbo/lib/python2.7/site-packages/theano/compile/pfunc.py", line 490, in pfunc
no_default_updates=no_default_updates)
File "/home/work/.jumbo/lib/python2.7/site-packages/theano/compile/pfunc.py", line 217, in rebuild_collect_shared
raise TypeError(err_msg, err_sug)
TypeError: ('An update must have the same type as the original shared variable (shared_var=<TensorType(float32, matrix)>, shared_var.type=TensorType(float32, matrix), update_val=Elemwise{sub,no_inplace}.0, update_val.type=TensorType(float64, matrix)).', 'If the difference is related to the broadcast pattern, you can call the tensor.unbroadcast(var, axis_to_unbroadcast[, ...]) function to remove broadcastable dimensions.')

cost_arr[2*i], _ = theano.scan(lambda f_bar, t: T.switch(T.or_(T.eq(t, f[i]), T.eq(t, T.shape(L)[0]-1)), 0, T.largest(gamma - s_Ot(T.stack(*m[:i+1]), f[i], t, L), 0)),
                                          sequences=[L, T.arange(T.shape(L)[0])])

It seem to do nothing?

https://github.com/npow/MemNN/blob/master/main.py#L150
m and f as two inputs, Why they are equal?
Thanks @npow