1742774400000

Work done

After figuring out that torch-rnn library was I a precursor of PyTorch, I resolved to build my LSTM cell with PyTorch in this second Jupyter notebook

Discussions

I proposed to Bill to continue by doing a visualisation work on some indices of the hidden state vector of the LSTM cell, similar to what Karpathy did in his blog. Bill proposed to use seaborn library, and I eventually noticed that seaborn’s heatmap could indeed be used for this purpose. The visualisation would replace the testing of the generated words.

Next Goal

I will try to add a spellchecker in the words-generating function to ensure that no real words are retrieve by the LSTM cell and write a function to save the “pseudo-dictionary”.