I’ve been following part 1 of the DL course. I see that the accuracy I’m getting in my runs seem to be worse than the ones in Jeremy’s videos (first two videos). I had started the course 3 months back, and back then I had accuracies similar to the ones showcased in the video. I took a break in between due to work, and restarted the course a couple weeks back with the standard
git pull and
conda env update, and that’s when I see that my accuracy values have decreased, including the number of incorrect classifications in the confusion matrix.
Below is the screenshot from Jeremy’s video:
And below is the screenshot from my notebook:
Both use the lesson1-rxt50.ipynb notebook, and I believe that this is a stark difference in accuracy (the ones that matter in Kaggle competitions at least). Can someone throw some light into as to why it could be happening? @jeremy - apologies in advance if seems to be a stupid question.