PatchTST baselines (channel-independent, standard protocol)
Trained on a free Colab T4 with the exact hyperparameters of the PatchTST paper (d_model=512, e_layers=3, patch_len=16, stride 8, dropout 0.3, lr 1e-4, patience 3, lookback 512, horizon 48). Channel-independent: each channel is a separate sample.
Contains {ds}.pt + {ds}.json for ETTh1, ETTh2, ETTm1, exchange_rate, electricity,
traffic, plus standard_benchmark.json (MASE/MAE/MSE/sMAPE under the protocol in
benchmark_standard.py of the NanoForecast repo).
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support