[Debugger Plugin] Fix a bug in handling string scalar #1131
Conversation
|
Please note that the Travis test failure is unrelated to this PR. |
|
Gentle ping. |
| self.assertEqual([['corge']], data) | ||
|
|
||
| def testArrayViewOnScalarString(self): | ||
| # Construt a numpy array that corresponds to a TensorFlow string tensor |
There was a problem hiding this comment.
Maybe s/array/scalar to make this comment more clear?
| self.assertEqual([['foo', 'bar'], ['baz', 'corge']], data) | ||
|
|
||
| def testArrayViewSlicingStringTensorToScalar(self): | ||
| # Construt a numpy array that corresponds to a TensorFlow string tensor |
There was a problem hiding this comment.
nit: s/Construt/Construct here and elsewhere.
| @@ -82,25 +82,42 @@ def array_view(array, slicing=None, mapping=None): | |||
| """ | |||
|
|
|||
| dtype = str(array.dtype) | |||
| # String-type TensorFlow Tensors are represented as object-type arrays in | |||
| # numpy. We map the type name back to 'string' for clarity. | |||
| if dtype == 'object': | |||
There was a problem hiding this comment.
Wow, is this the only way to identify a numpy array that represents a TensorFlow string tensor? Do the methods in this StackOverflow post work at all?
There was a problem hiding this comment.
For a tensorflow string-type tensor, the corresponding numpy array has the dtype "object", i.e., different from a regular string-type array in numpy. This is how tensorflow is and we need to live with that.
| return dtype, shape, sliced_array.tolist() | ||
|
|
||
| if np.isscalar(sliced_array) and str(dtype) == 'string': | ||
| # When a string Tensor (for which dtype is 'object') is sliced down to only |
There was a problem hiding this comment.
Does this logic exist because numpy.reshape(sliced_array, array.shape) could separate the string into several strings?
There was a problem hiding this comment.
The logic in this special case is because in the case of a string scalar, sliced array is just a Python string, i.e., not a numpy array. This is specific to the scalar case, i.e., if you have vector of strings with length > 1, you still get a numpy array.
| @@ -82,25 +82,42 @@ def array_view(array, slicing=None, mapping=None): | |||
| """ | |||
|
|
|||
| dtype = str(array.dtype) | |||
| # String-type TensorFlow Tensors are represented as object-type arrays in | |||
| # numpy. We map the type name back to 'string' for clarity. | |||
| if dtype == 'object': | |||
There was a problem hiding this comment.
For a tensorflow string-type tensor, the corresponding numpy array has the dtype "object", i.e., different from a regular string-type array in numpy. This is how tensorflow is and we need to live with that.
| return dtype, shape, sliced_array.tolist() | ||
|
|
||
| if np.isscalar(sliced_array) and str(dtype) == 'string': | ||
| # When a string Tensor (for which dtype is 'object') is sliced down to only |
There was a problem hiding this comment.
The logic in this special case is because in the case of a string scalar, sliced array is just a Python string, i.e., not a numpy array. This is specific to the scalar case, i.e., if you have vector of strings with length > 1, you still get a numpy array.
| self.assertEqual([['corge']], data) | ||
|
|
||
| def testArrayViewOnScalarString(self): | ||
| # Construt a numpy array that corresponds to a TensorFlow string tensor |
| self.assertEqual([['foo', 'bar'], ['baz', 'corge']], data) | ||
|
|
||
| def testArrayViewSlicingStringTensorToScalar(self): | ||
| # Construt a numpy array that corresponds to a TensorFlow string tensor |
Related to: #1115 After this change, visualization of string-type tensors is still not fully working, but at least the user can slice a string tensor to a single element (i.e., a string scalar) and see its value.
Related to: #1115
After this change, visualization of string-type tensors is
still not fully working, but at least the user can slice a
string tensor to a single element (i.e., a string scalar) and
see its value.