How to run FOCAL and other models on our own dataset?

[shahaamirbader]

Hi,

I was wondering how can I run FOCAL and the other models you have provided on my own time series dataset that contain signals related to ECG, EEG of participants?

[tomoyoshki]

Hi there,

Running FOCAL on Other Datasets

• We store our data in .pt files and use index files (e.g., train_index.txt, val_index.txt, test_index.txt) as pointers to these .pt files. You can refer to input_utils/multi_modal_dataset.py to see how the index files and .pt files are loaded.
• To use your own dataset with our code, please preprocess your data into a similar format.
• For time-series signals, we sample the sequences using the partition_subsequences function in input_utils/multi_modal_dataset.py, based on the suffix of the .pt files (e.g., x_{sequence number}.pt).
• Each .pt file should contain a sample in a dictionary format similar to the following:

  {
      "data": {
          "location1": {
              "mod1": Tensor,
              "mod2": Tensor
          } 
      },
      "label": LABEL
  }

• For reference, please review the processed MOD .pt files.

Once your data is prepared, you can:

1. Update the task_default_task in params/params_util.py.
2. Create a YAML file similar to data/MOD.yaml. This YAML file should include all the dataset and model configurations.
   • Key fields to update:
     • location_names (currently, we only support one location)
     • modality_names
     • num_segments
     • loc_mod_in_freq_channels
     • loc_mod_in_time_channels
     • loc_mod_in_spectrum_len
     • seq_len
   • You can also check out our paper for more details on these configurations.
3. Ensure the model configuration in your YAML file matches the shape of your data.

Then, simply specify the dataset you would like to use in your command line: python train.py -model=[MODEL] -dataset=TestDataset -learn_framework=FOCAL.

I hope this helps! Please let me know if you have any other questions.