I used various structerees for my network but none of it converged better than 40% so I decided to check if it converges to a constant, it does.

Then I checked if it converges in a “do nothing” case, the net takes 27 inputs and returns a single float, I set all input values to the same as the expected output value and even in this case nothing converges.

a view tested structures :

```
self.conv = nn.Sequential(
nn.Conv1d(1, 1, kernel_size=3, padding=1),
nn.AvgPool1d(3, padding=1),
nn.Linear(9, 9),
nn.Conv1d(1, 1, kernel_size=3, padding=1),
nn.AvgPool1d(3, padding=1),
nn.Linear(3, 3),
nn.Conv1d(1, 1, kernel_size=3, padding=1),
nn.AvgPool1d(3, padding=1)
)
```

```
self.dumb = nn.Sequential(
nn.Linear(27, 27),
nn.Linear(27, 1),
)
```

```
self.dumb = nn.Sequential(
nn.Linear(27, 1),
)
```

I used optim.Adam with learning rates from 0.1 to 0.001 and a batch size of 16 and 64 sets in total to cause overfitting

What could solve this issue ?