Fixes #595, support operator InvertPermutation #1117
Conversation
Signed-off-by: xavier dupré <[email protected]>
Signed-off-by: xavier dupré <[email protected]>
tf2onnx/onnx_opset/math.py
Outdated
|
|
||
| @tf_op("InvertPermutation") | ||
| class InvertPermutationOp: | ||
| supported_dtypes = [ |
There was a problem hiding this comment.
This seems unused. What is it for?
There was a problem hiding this comment.
The operator only supports these types. I was that on other operators. I assumed some verification would automatically happen. Based on your remark, I assume not. I'll do that then.
There was a problem hiding this comment.
Just searched the repo for supported_dtypes. It is usually declared in a function and then iterated through for casting purposes. I think you might have seen it and mistakenly thought it should be assigned in a class. You defined it in RFFT, ComplexAbs, and Atan2 but I don't think it should be included. Sorry I didn't catch that sooner.
tf2onnx/onnx_opset/math.py
Outdated
| outputs = [topk_unused, utils.port_name(topk_indices, 1)] | ||
| topk_node = ctx.make_node( | ||
| "TopK", inputs=[neg_node.output[0], shape_node.output[0]], | ||
| name=utils.make_name(node.name + '_topk'), outputs=outputs) |
There was a problem hiding this comment.
The output names should be node_name:0 and node_name:1 which I think make_node will assign automatically. make_name is for making node names, not output names, I think.
Signed-off-by: xavier dupré <[email protected]>
No description provided.