Hello everyone,
I’m working on the assignment for week 3 and observe a pretty weird behaviour. I train the model and observe the following graph for the development of the losses over time:
The accuracy graph looks like this, i.e. training and validation accuracy are 1 throughout all epochs!:
Besides the weirdness of the graphs: Since they seem not to change, no matter what the architecture of my model is, I suppose, something else must be going on.
Any hints, where to look?
Regards,
Thomas
I would look at the model itself maybe you are missing some connections between layers. Also I would check the input data are fed properly. You might be using a wrong activation at the output of your model that is possible too. So I would start with these.
I solved the problem - for those interested, it was a really embarassing mistake…
For those who want to know: In the function parse_data_from_file I wanted to append a 0 in case the label from the file was “0” and a 1 in any other case. Unfortunately I checked for equality with 0, not for equality with “0” and thus ended up generating only 1s as labels. This messed everything up 