Okay I'm not completely sure about this so take it with a grain of salt but I don't think you're hyperparameter/epoch tuning at is doing what you expect. With jupyter notebooks, it saves the models and each time you run an epoch, it continues tuning the previous weights. In order to really display epoch differences, you need to restart the runtime and repeat the process. If you notice, each time you run the code, the "first epoch accuracy" increases significantly. The first time you ran it, the accuracy was 83% after the first epoch. After the 10th, it was 90.6%. Then, for the next iteration (8 epochs), the accuracy was 91.2% after the first epoch. Then, when running on just a single epoch, it started at 93%. Likely this is because the model continued to train an additional 9 epochs. So, in fact, the single epoch data is ironically quite overfitting.(03:37:00 - 06:52:08)
TensorFlow 2.0 Complete Course - Python Neural Networks for Beginners Tutorial
freeCodeCamp.org
※本サイトに掲載されているチャンネル情報や動画情報はYouTube公式のAPIを使って取得・表示しています。
Timetable
動画タイムテーブル
動画数:1636件