Error Rate = 1 for all epochs while trying Lesson 1


I tried to implement the lesson 1 is it a bird or forest. I am getting error rate as 1 for epochs. But when i use it the model predicts the bird or forest correctly. This is my first time doing anything in DL so i dont know what to make out of this.

PFA the snapshot of my jupyter runtime

can you try another model than resnet18 and see what result you will get?

Hi @Kamui,

Tried with resnet34, same display. I also tried to plot confusion matrix. I am guessing that it might have something to do with error rate but there was no confusion as per the plot.


I have installed fastai, pytorch using miniconda, I dont know may be the distribution here is not proper.

Will give it a try using mamba as suggested in the course or colab

the models are overfitting to your dataset which is small by the way, try to have a bigger dataset try for example hundreds of bird images and hundreds of forest images

I am having this issue as well. I am getting an error_rate of 1.0000
I tried to change the model to look for butterfly and mountain (instead of bird or forest) respectively, and i when training, I still get an error rate of 1.0000

It correctly predicts a butterfly with probability of 1.0000
however, interestingly, when i pass in the original bird.jpg, it also predicts that it is a butterfly with a probability of 0.9739

for reference, here is my ss:

interestingly, this is what I got in colab when i executed the datablock. Bottom left is not a butterfly…