Raffi Khatchadourian
Raffi Khatchadourian
Indeed, setting [`t2`](https://github.com/ponder-lab/ML/blob/870a2f509a5acc4a81f34979854d32108eb54323/com.ibm.wala.cast.python.test/data/tf2_test_reshape.py#L11) to dimensions, i.e., `[-1, 28, 28, 1]`, compatible with MNIST, i.e., `[28, 28]`, works.
Thus, this issue is a byproduct of two different problems: 1. Regardless of their actual initial dimensions, *all* tensors are initialized with the MNIST dataset dimensions. https://github.com/wala/ML/issues/267 2. Tensor tracking...
> This error looks to be a shape mismatch, but it's also affecting the tensor type inference. If there's a tensor shape mismatch, tensor types aren't propagated as they normally...
``` [Node: synthetic < PythonLoader, Ltensorflow/functions/ones, do()LRoot; > Context: CallStringContext: [ script tf2_test_add7.py.do()LRoot;@103 ], v2] --> [SITE_IN_NODE{:Llist in CallStringContext: [ com.ibm.wala.FakeRootClass.fakeRootMethod()V@2 ]}] [Node: synthetic < PythonLoader, Ltensorflow/functions/ones, do()LRoot; > Context:...
``` callees of node Lscript tf2_test_add7.py : [import, add, ones, ones] IR of node 2, context CallStringContext: [ com.ibm.wala.FakeRootClass.fakeRootMethod()V@2 ] ... 100 v251 = new @100 tf2_test_add7.py [14:16] -> [14:22]...
Currently, we are inferring initial tensor shapes from literal values that are being sent to "generator" APIs. The return values of those APIs are then stored into variables with inferred...
The current fundamental issue here is that we are using the pointer analysis to do shape inference. Perhaps there needs to be another pass that uses the tensor analysis in...
Sorry, I'll fix the errors now.
Ah, looks like I need to add more changes from our fork to get to this to work. I'll turn it into a draft PR for now.