Skip to content

Niteeshaballa/Learn-Alphabets-LSTM

Repository files navigation

Learn-Alphabets-LSTM

LSTMs learning alphabets in different ways

  1. one_char to one_char -- Naive LSTM learning one char to one char with no dependencies and achieved an accuracy of 100%.

  2. one_char to one_char state -- An LSTM learning alphabets sequence with state maintained within a batch and achieved an accuracy of 100%.

  3. feature window to one_char -- An LSTM learning alphabets with a feature size and 1 time step and achieved an accuracy of 86.96%.

  4. time steps window to one_char -- An LSTM learning alphabets with some time steps and 1 feature step and achieved an accuracy of 100%.

  5. Var length input to one char -- An LSTM learning alphabet sequences with variable length and outputs one character and achieved an accuracy of 97.50%. It takes a long time to train on CPU. The model weights ("model500epochs.h5") and the model structure("model.json") are also uploaded.

About

LSTMs learning alphabets in different ways

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published