TFLSTM-18

┏━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━┓
┃   ┃ Name                     ┃ Type                 ┃ Params ┃
┡━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━┩
│ 0 │ criterion                │ WeightedMSELoss      │      0 │
│ 1 │ model                    │ TransformerLSTMModel │  4.3 M │
│ 2 │ model.encoder_grn        │ GatedResidualNetwork │  199 K │
│ 3 │ model.decoder_grn        │ GatedResidualNetwork │  199 K │
│ 4 │ model.encoder            │ LSTM                 │  1.1 M │
│ 5 │ model.decoder            │ LSTM                 │  1.1 M │
│ 6 │ model.transformer_blocks │ ModuleList           │  1.8 M │
│ 7 │ model.output_head        │ Linear               │    257 │
└───┴──────────────────────────┴──────────────────────┴────────┘
Trainable params: 4.3 M
Non-trainable params: 0
Total params: 4.3 M
Total estimated model params size (MB): 17

Same as TFLSTM-16, but without Bdot in the past, and adding back measured field to the past as normal. We also make the encoder/decoder GRN hidden dim the model dimension instead of the input dimension.o

Trained on GPUs 1-3 on ml004

We evaluate it on the dataset from Dedicated MD 2025-07-23 and find that it can predict the flat bottom field within $3\times 10^{-5}T$ on drift-corrected data. I.e. the model learns real field.