Closed MLuchmann closed 6 months ago
The tests failed because the activation has to be applied even for depth = 1 (otherwise we have no universal approximation properties). I changed this and therefore was able to remove the changes in the tests. They now use smaller networks but as they still work, it's fine. Also added an assertion to make sure nobody uses the DeepResidualNetwork with depth = 0.
Bugfix: Add activation function after first layer
Description
No activation function was applied after the first layer in the residual network. Also, the actual width of a network was given by width+1. Now, it correctly correspond to the width parameter.
To make the test scripts consistent to previous tests, the widths had to be adjusted by +1. Without this fix, the tests failed
Checklist for Contributors
feature/title-slug
convention.Bugfix: Title
convention.Checklist for Reviewers: