tf.keras implementation

sayred1 / tf.keras_GCN

Translation of AGCN implemented with tf to an implementation of tf.keras.

1 stars 0 forks source link

tf.keras implementation #3

Open qzhu2017 opened 5 years ago

qzhu2017 commented 5 years ago

Todo

1, add functions for attention GCN and gated GCN
2, check if the normalization helps.
3, investigate how embedding works in tf.keras.

Done

1, analyze the evolution of weights/bias. (by adding histogram_freq in keras.callbacks.TensorBoard)

qzhu2017 commented 5 years ago

After a survey on the embedding techniques. I believe it is much easier to do it with scikit-learn, instead of Tensorflow. Let's forget about this idea for now.

qzhu2017 commented 5 years ago

@sayred1 I suggest you add the attention GCN and gated GCN now. I may spend some time on other things.

sayred1 commented 5 years ago

@qzhu2017 Will do, I just checked whether or not normalizing helps with results, it seems to not. Each run seemed to converge to nearly the same values. I uploaded two quick plots of MSE and MAE training comparisons for four runs (two normalized, two unnormalized), two layers, 1000 epochs, 100 batches, over 1000 molecules.

I'll go ahead and add the rest of the methods into your code.

sayred1 commented 5 years ago

Hey, @qzhu2017 If you don’t mind, I’d like to add GCN gate and attention tomorrow. I currently have two assignments due within the next 24 hours.

On Mar 17, 2019, at 14:10, Qiang Zhu notifications@github.com wrote:

@sayred1 https://github.com/sayred1 I suggest you add the attention GCN and gated GCN now. I may spend some time on other things.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/sayred1/tf.keras_GCN/issues/3#issuecomment-473715194, or mute the thread https://github.com/notifications/unsubscribe-auth/Ah5lj06Q9a8Xk4mAUsnXOKLv3mzPdKXRks5vXq9QgaJpZM4b4Ptr.

sayred1 commented 5 years ago

@qzhu2017 I've finished adding the other models into your code and I am currently running some tests before uploading to the repo.

Also, I forgot to mention that I will be going out of town in a few hours from now until Thursday. If I can't get the testing done soon, I will upload the script to the repo.