Export bug conv2D when not naming biases

What commit version of aidge do you use

• aidge_core: 82dba706e8ea2adf6558560ef4a0e67e42176dfe
• aidge_export_cpp: dev 52348ca9
• On both main and dev

Problem description

Conv2D biases are missing when not naming them with aidge_export_cpp.export_utils.set_nodes_names(s)

Reproducible example code

1. Create a conv2D graph
2. Set the input / weights / biases value.
3. Schedule / Forward
4. Export the graph to export_conv2d_0
5. Use set_nodes_names on the scheduler, and export it to export_conv2d_1.

Result : export_conv2d_0 output is wrong because biases are missing from the parametersfolder. export_conv2d_1 works as expected.

Straightforward code :

test_conv2d_export.py

Another minimal test based on test_export.py methodology :

test_conv2d_export_kernel.py

When the node has no name, no biases are exported.

@louislerbourg a workaround is possible and you're not blocked by this problem?

@axelfarr @cmoineau :What behavior do you suggest when the node has no name?

changed milestone to %aidge v0.8.0

added Bug 🐛 label

changed milestone to %aidge v0.7.0

Indeed, nodes cannot be exported without a name. How to best name the nodes, even when they are not initially named is still a old, long and hot debate!

To resolve this issue, I propose to add a check during export to raise an error if a name is missing, in MR aidge_core!448 (merged).

mentioned in merge request aidge_core!448 (merged)

closed

Thanks very much, indeed i'm not stuck with this, the check should handle this wrong silent behaviour.