How nicodjimenez/lstm works
This repository is an educational, minimal LSTM implementation rather than a general-purpose training product. Its documented workflow is to run the example, which trains a sequence predictor using Euclidean loss; internally, parameters and recurrent state support per-step computation, sequence-level backpropagation, and parameter updates. The example source was not excerpted, so its internal wiring is shown only where the README documents it; the implementation details below come from lstm.py.
Main entry points: lstm.py Sequence network · test.py Training example
Components 7
Learning example
- Training example
test.py - Euclidean loss
test.py
LSTM runtime
- LSTM parameters
lstm.py - Recurrent timestep
lstm.py - Sequence network
lstm.py - Cell state
lstm.py
Other components
- User
Connections 8
- User → Training example: runs
README.md - Training example → Sequence network: trains sequence
README.md - Sequence network → Euclidean loss: evaluates targets
README.md - Sequence network → Recurrent timestep: runs timesteps
lstm.py - Recurrent timestep → LSTM parameters: uses weights
lstm.py - Recurrent timestep → Cell state: updates activations
lstm.py - Sequence network → LSTM parameters: accumulates gradients
lstm.py - Sequence network → LSTM parameters: applies updates
lstm.py