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

3.5-classifying-movie-reviews Further experiments #131

Open hkhrais opened 4 years ago

hkhrais commented 4 years ago

In this example (IMDB) the author asks to use 1 or 3 hidden layers

We were using 2 hidden layers. Try to use 1 or 3 hidden layers and see how it affects validation and test accuracy.

These experiments will help convince you that the architecture choices we have made are all fairly reasonable, although they can still be improved!

when using 2 hidden layers as he did you get 0.884 ... when i used one hidden layer i get 0.887 which got me confused,, not sure why 2 layers sound like "optimal architecture"?

ref: https://github.com/fchollet/deep-learning-with-python-notebooks/blob/master/3.5-classifying-movie-reviews.ipynb

Phanttan commented 3 years ago

Dear @hkhrais ,

I think the meaning is not pursue that using less layer is better. Only get experience, in this data the more layer is maybe not linear than better. With me, I will not concentrate to more dense (I tried 3-layer and the result is the same to 2-layer model, loss more time). As book's recommendation we can try with another method to improve model. I did not use Relu, and constitute by tanh and got a little improvement. https://colab.research.google.com/drive/15kQxc7vqVoMGj2zgzVN9ss4iqukjHkrE?authuser=1 I hope it is useful to you!