Skip to content

Experimentns

Jump to bottom
Pritam Sukumar edited this page Jan 2, 2020 · 1 revision

Todo

Add mathematical things to GLOVE?

Features with no cleanup

Bench mark to beat is

Now that we're using NNs and NLP techniques, does all that cleanup make sense?

  • Trying on MTH_NN_GPU_colab_acc82 to see what happens. Still keeping HTML tags going.

Validation accuracy is low at only 65%. Training accuracy is very high at 83%. The network is probably overfitting to the training set because of all the symbols etc.

Clone this wiki locally