This is an implementation of a Long Short-Term Memory (LSTM) network for sequence modeling, built using TensorFlow.
LSTM is a type of Recurrent Neural Network (RNN) that can model temporal dependencies in data by maintaining a hidden state and selectively forgetting and remembering information over time. This makes it well-suited for tasks such as speech recognition, language translation, and image captioning.
Python 3.5 or higher
Pytorch
NumPy
Pandas
Clone this repository.
Install the dependencies listed above.
Prepare your training data in a suitable format (e.g. a NumPy array with shape (num_samples, sequence_length, num_features)).
Instantiate an instance of the LSTM class with the desired hyperparameters.
Train the model
Test the model
hidden_size: the number of neurons in each LSTM layer
num_layers: the number of LSTM layers in the network
dropout: the dropout rate applied between LSTM layers
learning_rate: the learning rate used for optimization
batch_size: the batch size used for training
epochs: the number of training epochs
This implementation was created by Tanmay Agrawal. It is based on the TensorFlow documentation and the following resources:
https://towardsdatascience.com/a-comprehensive-guide-to-neural-machine-translation�using-seq2sequence-modelling-using-pytorch-41c9b84ba350
https://machinelearningmastery.com/define-encoder-decoder-sequence-sequence-model�neural-machine-translation-keras/