search

fchollet / deep-learning-with-python-notebooks

Jupyter notebooks for the code samples of the book "Deep Learning with Python"
MIT License
18.17k stars 8.53k forks source link

The model from first conv2d example doesn't train well. #117

Open PPrasai opened 5 years ago

PPrasai commented 5 years ago

The model built in 5.1-introduction-to-convnets.ipynb does not train at all. The training accuracy stays at around 0.09

PPrasai commented 5 years ago

However adding a MaxPooling2D(2,2) after the last Conv2D layer trains the model well.

RorisangSitoboli commented 4 years ago

Hi. Even after adding the MaxPooling2D(2,2) i still get a model that achieves a max accuracy of 0.91 and then drops back to 0.73 within the first epoch and then its backwards to 0.09 when all 5 epochs have been completed. Funny that on Google Colab the code runs perfectly and the model achieves a final accuracy of 98.84%.

I have been stuck at this stage the past 3 days. Is there something else i can try?