urancon/deepSTRF-statenet-gru-ns1

StateNet (deepSTRF)

Pretrained checkpoint produced with deepSTRF.

Usage

from deepSTRF.models.audio import StateNet  # adjust import path if needed

model = StateNet.from_pretrained("<owner>/<repo>")
model.eval()

Config

{
  "connectivity": "LC",
  "hidden_channels": 7,
  "kernel_size": 7,
  "n_frequency_bands": 34,
  "out_neurons": 119,
  "output_activation": null,
  "prefiltering": null,
  "rnn_type": "GRU",
  "stride": 3,
  "temporal_window_size": 1
}

Metadata

{
  "dataset": "NS1 (Harper et al. 2016, Rahman et al. 2020)",
  "library": "deepSTRF",
  "library_commit": "c07e561f7835454fdacbb35f38a8957668b8d1fd",
  "model_kwargs": {
    "connectivity": "LC",
    "hidden_channels": 7,
    "kernel_size": 7,
    "n_frequency_bands": 34,
    "rnn_type": "GRU",
    "stride": 3
  },
  "out_neurons": 119,
  "split": {
    "test_stim_idx": [
      17,
      18,
      19
    ],
    "train_stim_idx": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7,
      8,
      9,
      10,
      11,
      12,
      13
    ],
    "val_stim_idx": [
      14,
      15,
      16
    ]
  },
  "task": "audio neural response prediction",
  "test_metrics": {
    "cc_mean": 0.659480631351471,
    "cc_median": 0.681122899055481,
    "cc_norm_mean": 0.7703116536140442,
    "cc_norm_median": 0.7958264350891113,
    "loss": 0.015971951186656952
  },
  "training": {
    "device": "cuda",
    "epochs_run": 80,
    "fitter_kwargs": {
      "max_epochs": 80,
      "mode": "max",
      "monitor": "val_cc_norm",
      "patience": 15
    },
    "loss": "mse_loss vs PSTH",
    "optimizer": "AdamW",
    "optimizer_kwargs": {
      "lr": 0.001,
      "weight_decay": 0.0
    },
    "seed": 0,
    "train_time_min": 5.85
  }
}