Hi,
You are indeed still working with a symbolic graph, the error basically
says that it is not possible to get a subgraph going from the inputs you
provided that can compute the outputs you request.
In that case, it is likely happening when trying to build the `predict`
function:
predict = theano.function(inputs=[X], outputs=p_y_x_max,
updates=updates,
allow_input_downcast=True,
mode='FAST_RUN')
Here, your outputs are:
- `p_y_x_max`, and
- all the update values in `updates`
However, to get those update values, you need the cost, and for that you
need Y, which is not provided as input.
If you want a prediction function, you should probably remove
"updates=updates," from that call.
On Thu, Sep 08, 2016, [email protected] wrote:
> Hello theano community,
>
> I've been having this missing input error recently on small piece of theano
> example.
>
> The full stack error is the following:
>
> MissingInputError: A variable that is an input to the graph was neither
> provided as an input to the function nor given a value.
> A chain of variables leading from this input to an output is
> [ Y,
> Elemwise{mul}.0,
> Elemwise{true_div}.0,
> Elemwise{true_div}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{mul}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{Switch}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{add,no_inplace}.0,
> dot.0,
> Reshape{4}.0,
> Elemwise{Switch}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{add,no_inplace}.0,
> AbstractConv2d_gradInputs{border_mode='valid', subsample=(1, 1),
> filter_flip=True, imshp=(None, None, None, None),
> kshp=(None, None, None, None)}.0,
> Elemwise{Switch}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{add,no_inplace}.0,
> AbstractConv2d_gradInputs{border_mode='valid', subsample=(1, 1),
> filter_flip=True, imshp=(None, None, None, None),
> kshp=(None, None, None, None)}.0,
> Elemwise{Switch}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{add,no_inplace}.0,
> AbstractConv2d_gradInputs{border_mode='valid', subsample=(1, 1),
> filter_flip=True, imshp=(None, None, None, None),
> kshp=(None, None, None, None)}.0, Split{2}.0,
> IncSubtensor{Inc;::, ::, :int64:, :int64:}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{Switch}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{add,no_inplace}.0,
> AbstractConv2d_gradInputs{border_mode='valid', subsample=(1, 1),
> filter_flip=True, imshp=(None, None, None, None),
> kshp=(None, None, None, None)}.0,
> Elemwise{Switch}.0, Elemwise{add,no_inplace}.0,
> Elemwise{add,no_inplace}.0,
> AbstractConv2d_gradWeights{border_mode='valid', subsample=(1, 1),
> filter_flip=True, imshp=(None, None, None, None),
> kshp=(None, None, None, None)}.0,
> Elemwise{mul,no_inplace}.0,
> Elemwise{add,no_inplace}.0,
> Elemwise{true_div,no_inplace}.0,
> Elemwise{mul,no_inplace}.0,
> Elemwise{sub,no_inplace}.0 ].
>
> This chain may not be unique
> Backtrace when the variable is created:
> File "<decorator-gen-57>", line 2, in run
> File
> "/home/user/anaconda2/lib/python2.7/site-packages/IPython/core/magic.py",
> line 188, in <lambda>
> call = lambda f, *a, **k: f(*a, **k)
> File
> "/home/user/anaconda2/lib/python2.7/site-packages/IPython/core/magics/execution.py"
> , line 742, in run
> run()
> File
> "/home/user/anaconda2/lib/python2.7/site-packages/IPython/core/magics/execution.py"
> , line 728, in run
> exit_ignore=exit_ignore)
> File
> "/home/user/anaconda2/lib/python2.7/site-packages/IPython/core/interactiveshell.py"
> , line 2481, in safe_execfile
> self.compile if kw['shell_futures'] else None)
> File
> "/home/user/anaconda2/lib/python2.7/site-packages/IPython/utils/py3compat.py"
> , line 289, in execfile
> builtin_mod.execfile(filename, *where)
> File "/home/user/projects/python/theano/unet.py", line 291, in <module>
> training()
> File "/home/user/projects/python/theano/unet.py", line 259, in training
> Y = tt.fmatrix(name='Y').astype(dtype='float32')
>
>
>
> The minimal example I was trying to run is the following:
>
> def training():
>
>
> W = []
> for weight in range(len(weights)):
> W.append(init_weights(weights[weight]))
>
>
> X = tt.ftensor4(name='X').astype(dtype='float32')
> Y = tt.fmatrix(name='Y').astype(dtype='float32')
>
> p_y_x = model(X, W)
>
> p_y_x_max = tt.argmax(p_y_x, axis=1)
>
> cost = tt.mean(tt.nnet.categorical_crossentropy(p_y_x, Y))
>
> updates = optimize(cost, W)
>
> train = theano.function(inputs=[X, Y], outputs=cost,
> updates=updates,
> allow_input_downcast=True,
> mode='FAST_RUN')
>
>
> predict = theano.function(inputs=[X], outputs=p_y_x_max,
> updates=updates,
> allow_input_downcast=True,
> mode='FAST_RUN')
>
> training()
>
>
> Given the error output what is strange to me is the fact that at this point
> if i just run the code I should be getting the computation graph instead
> I'm getting complaints about the Y variable?
> I don't understand why since I'm not calling "train" or "predict" in order
> to specify arguments.
> At this point I would just expect theano to build the computation graph? Am
> I missing somehting?
>
> --
>
> ---
> You received this message because you are subscribed to the Google Groups
> "theano-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
--
Pascal
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
