Skip to content

adding_a_model.md

Latest commit

 

History

History
88 lines (53 loc) · 2.31 KB

adding_a_model.md

File metadata and controls

88 lines (53 loc) · 2.31 KB
Raw

Adding a New Model

This guide explains how to add a new SNN or ANN forecasting model to SpikingTSF.

Step 1: Add the Model File

Create models/<YourModel>.py. The model class should:

  • Accept configuration arguments (via the args namespace passed from run_long.py)
  • Implement a forward(x) method taking (batch, seq_len, n_vars) and returning (batch, pred_len, n_vars)
  • Follow the interface of existing models (e.g., models/SpikF.py, models/TSGRU.py)

Include a comment at the top citing the original paper.

Step 2: Register in exp/exp_ETT.py

This is where models are actually imported and instantiated. You need to make three additions:

a) Add the import at the top:

from models.YourModel import YourModel

b) Add it to the _MODEL_REGISTRY dict (or the equivalent mapping used to build the model):

'YourModel': YourModel,

c) Classify it by adding its name to the appropriate set:

  • _ANN_MODELS — if it is an ANN (no spikes)
  • _CD_MODELS — if it uses spikingjelly.clock_driven and needs cd_functional.reset_net()
  • _AB_MODELS — if it uses spikingjelly.activation_based
  • _SELF_RESET_MODELS — if the model resets itself internally during forward()

Check which SpikingJelly backend your model uses and pick accordingly.

Step 3: Register in run_long.py

Add the model name to the choices list for the --model argument:

parser.add_argument('--model', ..., choices=[
    ...
    'YourModel',
])

Step 4: Add a Config File

Create configs/<YourModel>/ETTh1.yaml with default hyperparameters. Use existing YAML configs as templates.

Step 5: Add Scripts

Create scripts/<YourModel>/run_ETTh1.sh with the standard 3-seed pattern. Add scripts for other datasets as available.

Step 6: Update MODEL_ZOO.md

Add a row with model name, type, Spike?, source paper, source repo, file path, and status.

Step 7: Add Citation

In MODEL_ZOO.md and in the model file, provide the BibTeX citation for the original paper.

Step 8: Add Results (if available)

If you have results following the protocol in PROTOCOL.md, add them to RESULTS.md.

Step 9: Open a Pull Request

Follow the checklist in CONTRIBUTING.md and the PR template.