CODEWITHSUNDEEP
pythonc++python-3.xtensorflowboost
jm Park
jm Park

Reputation: 11

Python -> Boost Python + C++ error

I am replacing the tensor flow Python code with boost Python C++ code.

The code that runs in Python does not run in Boost Python.

I am creating the mnist example, but the softmax_cross_entropy function is strange.

In Python:

loss_op = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=label, logits=logit))

In Boost Python:

py::object loss_op = py::eval("tf.nn.softmax_cross_entropy_with_logits_v2(labels='label_', logits='logit')", main_namespace);

The variables being used, label_ and logit, are of type float32.

Tensor("Placeholder_1:0", shape=(?, 10), dtype=float32)
Tensor("add_3:0", shape=(?, 10), dtype=float32)

Boost Python shows this error:

  File "<string>", line 1, in <module>
  File "C:\ProgramData\Anaconda3\envs\tf35-gpu\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 1872, in softmax_cross_entropy_with_logits_v2
    precise_logits = _flatten_outer_dims(precise_logits)
  File "C:\ProgramData\Anaconda3\envs\tf35-gpu\lib\site-packages\tensorflow\python\ops\nn_ops.py", line 1628, in _flatten_outer_dims
    output_shape = [product, shape[-1]]
IndexError: list index out of range

How can I fix this?

Upvotes: 0

Views: 209

Answers (1)

Theorem
Theorem

Reputation: 91

I don't think it is an error of boostpython, it is more like python index issue. Maybe you should check 

output_shape = [product, shape[-1]]

to see if the -1 should be out side the bracket?

Upvotes: 1

Related Questions