GitHub - 0xdeadf1sh/T1DMAI: A model zoo of blood glucose forecasters trained & tested on my own data
A model zoo of blood glucose forecasters trained & tested on my own data - 0xdeadf1sh/T1DMAI
A model zoo of blood glucose forecasters trained & tested on my own data - 0xdeadf1sh/T1DMAI
Note
Most of the code in this repo was written by AI. If you don't like that, feel free to leave.
This repo is a zoo. Here I train & test many blood glucose forecasting models (including my own) on both my own data and metabonet (public).
My model is a small transformer that forecasts blood glucose 2 hours ahead. I've made two variants: D16 DILATE and D16 MSE which use DILATE and MSE loss for training, respectively. D16 refers to the width of the model (16). They are both trained on synthetic traces, and then a LoRA adapter is added for light fine-tuning, but I also test them without the adapter.
Feast thine eyes on some of their forecasts (these have been trained on synthetic traces, then fine-tuned with a LoRA adapter on my training split):
Caution
Research and educational use only - not a medical device. T1DMAI is a research artifact. It is not clinically validated, has no regulatory clearance, and its output is a forecast of research signals, not medical advice. It must not be used to make medical, diagnostic, or treatment decisions, to calculate or adjust insulin doses, or to manage diabetes in any way. For medical guidance consult a qualified healthcare professional. The software is provided "as is", without warranty of any kind (see LICENSE).
I have type 1 diabetes and I wanna know if I can train a BG forecasting model to help me with my daily management. To this end I've made some auxiliary tools:
For evaluation, I use the recommendations by Jacobs et al., IEEE Reviews in Biomedical Engineering, 2024 [1]. It's a nice paper that you should probably read.
All models suck. On my data XGBoost is the best one but my own model comes close. On MetaboNet fine-tuned Timestral wins. More details below:
| Model | Family | Description |
|---|---|---|
| Persistence | Baseline | Last reading carried forward. |
| Linear extrapolation | Baseline | Straight line through the last few readings. |
| AR (ridge) | Classical | Linear regression on recent glucose. |
| ARX (ridge) | Classical | Linear regression on recent glucose, insulin and carbs. |
| ARIMA | Classical | Glucose only. |
| SARIMAX | Classical | With insulin and carbs. |
| Kalman filter | Classical | Damped local linear trend. |
| SVR | Classical | Glucose, insulin and carbs. |
| Sparse GP | Classical | 1000 inducing points. |
| XGBoost | Classical | Glucose, insulin and carbs [18]. |
| LightGBM | Classical | Same features, depth, rate and rounds as XGBoost [19]. |
| MiniRocket | Classical | 9996 random-kernel features over 6 hours of glucose, insulin and carbs, plus the last reading and binned doses, into ridge regression per step [28]. |
| LSTM | Deep | 2 hours of history [5]. |
| LSTM (15 h) | Deep | Two layers of 128 over 15 hours, as in MetaboNet-Bench [3]. |
| GluNet | Deep | Dilated causal convolutions over 4 hours [6]. |
| PatchTST | Deep | [13] |
| N-HiTS | Deep | [14] |
| Chronos-Bolt tiny and mini | Foundation | 9M and 21M parameters, glucose only, zero-shot and fine-tuned [15]. |
| Chronos-2 | Foundation | 120M parameters, zero-shot. Past insulin and carbs; on my data also the future ones [16]. |
| TimesFM 2.5 | Foundation | 200M parameters, zero-shot. Glucose only on MetaboNet; on my data future insulin and carbs through in-context regression [17]. |
| Timestral 5M, 28M, 123M and 209M | Foundation | Zero-shot and fine-tuned. Past insulin and carbs; on my data also the future ones [27]. |
| T1DMAI D16 DILATE and D16 MSE | T1DMAI | 31k parameters. Glucose, carbs and insulin. |
Numbers in brackets point to References. Everything except the baselines and
the foundation models was trained from scratch on each dataset's training split. T1DMAI comes in two versions: sim only is the checkpoint in models/, and
sim + LoRA adds a frozen-base rank 1 adapter [23] (3771 parameters) trained on the real training
split.
| Architecture | encoder-only transformer, 30 min patches, one attention head, RoPE [26], SwiGLU |
| Size | 16 layers of width 16, 31k parameters |
| Loss | 7 quantile pinball loss plus a median term (DILATE [25] on soft-DTW [24], or MSE) |
| Inputs | CGM in a risk-scaled space, carb absorption curve, insulin action curve, mask bit |
| Output | 7 quantiles (5% to 95%) for every 5 minute step of the next 2 hours |
| History | up to 7 days |
| Simulator training | 1M simulated people, 7 days each, 10000 steps of batch 512 |
| Runtime | ExecuTorch or LiteRT, on the phone in T1DMDROID |
More in docs/USAGE.md and ARCHITECTURE.md.
My data. 115 days from T1DMDROID: days 1 to 80 for training, 81 to 115 for testing. Note that the model sees future inputs like insulin & carbs. I should probably also train the adapters while withholding those inputs.
| Source | Subjects | Person-days | Forecasts | CGM coverage % | Mean mg/dL | CV % | < 54 % | < 70 % | 70-180 % | > 180 % | > 250 % |
|---|---|---|---|---|---|---|---|---|---|---|---|
| t1dmdroid | 1 | 35 | 9764 | 97 | 130 | 28 | 0.4 | 3.0 | 89 | 8 | 0 |
MetaboNet [2]. Official split, 279 test people from 9 studies. Blind, because most of them use closed-loop pumps whose future insulin reacts to glucose the model hasn't seen.
| Source | Subjects | Person-days | Forecasts | CGM coverage % | Mean mg/dL | CV % | < 54 % | < 70 % | 70-180 % | > 180 % | > 250 % |
|---|---|---|---|---|---|---|---|---|---|---|---|
| AZT1D | 4 | 131 | 775 | 99 | 141 | 30 | 0.3 | 1.3 | 82 | 17 | 2 |
| BrisT1D | 2 | 312 | 1953 | 77 | 144 | 33 | 0.2 | 1.8 | 78 | 20 | 3 |
| CTR3 | 5 | 1157 | 1052 | 51 | 141 | 38 | 0.7 | 3.5 | 78 | 19 | 4 |
| HUPA-UCM | 4 | 315 | 896 | 100 | 132 | 36 | 0.9 | 6.8 | 78 | 16 | 2 |
| IOBP2 | 62 | 4737 | 14653 | 84 | 159 | 37 | 0.4 | 2.0 | 68 | 30 | 8 |
| Loop | 159 | 47438 | 203538 | 93 | 139 | 40 | 0.7 | 3.9 | 77 | 19 | 5 |
| PEDAP | 6 | 2070 | 6321 | 85 | 158 | 38 | 0.4 | 2.4 | 68 | 29 | 9 |
| ReplaceBG | 35 | 7967 | 35683 | 87 | 158 | 41 | 1.0 | 4.2 | 64 | 32 | 9 |
| ShanghaiT1DM | 2 | 19 | 28 | 33 | 159 | 39 | 1.0 | 3.8 | 63 | 33 | 7 |
| All | 279 | 64146 | 264899 | 90 | 143 | 40 | 0.7 | 3.7 | 75 | 22 | 6 |
| My data | MetaboNet | |
|---|---|---|
| Forecasts | every 5 min of the test days | every 10th row of the leaderboard template |
| Horizons | 30, 60, 90, 120 min | 30, 60, 90, 120 min |
| Training windows, other models | 50000 | 338000 (135000 for fine-tuned Chronos and Timestral) |
| LoRA steps | 2000 | 6000 |
| Model | Parameter count | Microseconds per forecast | History used |
|---|---|---|---|
| Persistence | 0 | 0.0042 | 5 min |
| Linear extrapolation | 0 | 0.033 | 30 min |
| AR (ridge) | 648 | 0.13 | 2 h |
| SARIMAX | 8,088 | 0.21 | 8 h |
| ARIMA | 4,632 | 0.27 | 4 h |
| ARX (ridge) | 1,116 | 0.36 | 6 h |
| Kalman filter | 864 | 0.48 | 3 h |
| SVR | 69,772 | 4.2 | 6 h |
| Sparse GP | 66,134 | 6.1 | 6 h |
| N-HiTS | 417,772 | 6.5 | 6 h |
| LSTM | 103,192 | 33 | 2 h |
| GluNet | 65,368 | 71 | 4 h |
| PatchTST | 182,266 | 74 | 6 h |
| XGBoost | 2,692,860 | 506 | 6 h |
| Chronos-Bolt tiny (fine-tuned) | 8,669,901 | 1,346 | 24 h |
| Chronos-Bolt tiny (zero-shot) | 8,669,901 | 1,361 | 24 h |
| Chronos-Bolt mini (zero-shot) | 21,256,141 | 3,871 | 24 h |
| Chronos-Bolt mini (fine-tuned) | 21,256,141 | 3,889 | 24 h |
| T1DMAI D16 DILATE | 31,253 | 5,679 | 168 h |
| T1DMAI D16 MSE | 31,253 | 7,342 | 168 h |
Timestral [27] on two 48 h spans of my test days, zero-shot and fine-tuned on my training split. Bands are the raw 50% and 90% quantiles of the model.
MiniRocket [28] on the same two 48 h spans, trained on my training split. Bands are residual quantiles on the last 5 days of the training split.
| Term | Meaning |
|---|---|
| Non-blind | The model gets the carbs and insulin logged after the forecast moment. |
| Blind | It doesn't. |
| Time gain | Horizon minus the lag of peak correlation between forecast and CGM. |
| Example forecasts | Median case of: falls into a low, rises over 40 mg/dL after a meal, calm nights. |
| After a meal | At least 5 g of carbs logged in the 2 hours before. |
| Parameter count | Stored weights and constants in the exported model. |
A * marks the lowest RMSE and the models not significantly worse than it.
My data| Model | Family | RMSE 30 | RMSE 60 | RMSE 90 | RMSE 120 | gRMSE 60 | MARD 30 | MARD 60 | MARD 120 | ME 60 | MRE 60 % | TG 30 | TG 60 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| XGBoost | Classical | 14.6 * | 22.4 * | 27.5 * | 31.6 * | 24.3 | 8.5 | 13.4 | 20.1 | +1.1 | +3.3 | 15 | 30 |
| LightGBM | Classical | 14.6 * | 22.4 * | 27.6 * | 31.5 * | 24.4 | 8.5 | 13.4 | 20.2 | +1.1 | +3.3 | 15 | 30 |
| T1DMAI D16 DILATE (sim + LoRA) | T1DMAI | 15.2 * | 22.8 * | 27.4 * | 31.7 * | 24.7 | 8.8 | 13.5 | 20.3 | -1.6 | +1.2 | 15 | 40 |
| LSTM (15 h) | Deep | 15.1 | 23.2 | 28.6 * | 32.9 | 24.9 | 8.9 | 14.4 | 21.9 | +2.5 | +4.4 | 10 | 20 |
| PatchTST | Deep | 15.1 | 23.3 | 28.8 * | 33.0 | 25.0 | 8.9 | 14.3 | 21.8 | +1.5 | +3.4 | 10 | 20 |
| SVR | Classical | 15.2 * | 23.4 * | 28.4 * | 31.8 * | 25.3 | 8.6 | 13.7 | 20.0 | -0.9 | +1.6 | 10 | 20 |
| T1DMAI D16 MSE (sim + LoRA) | T1DMAI | 15.1 * | 23.5 * | 28.4 * | 32.4 * | 25.4 | 8.7 | 13.9 | 20.3 | -0.2 | +2.0 | 20 | 50 |
| Sparse GP | Classical | 15.4 | 23.7 | 28.9 * | 32.8 * | 25.8 | 9.0 | 14.4 | 21.2 | +0.7 | +3.1 | 10 | 20 |
| GluNet | Deep | 15.4 | 24.0 | 29.8 | 34.5 | 25.6 | 8.9 | 14.4 | 22.3 | +1.0 | +2.8 | 10 | 20 |
| MiniRocket | Classical | 15.7 | 24.0 | 28.7 * | 31.9 * | 26.3 | 9.3 | 14.9 | 21.5 | +0.4 | +3.8 | 10 | 20 |
| Timestral 28M (fine-tuned) | Foundation | 15.4 * | 24.2 * | 29.8 * | 33.6 * | 26.5 | 8.7 | 14.1 | 21.0 | -1.6 | +1.2 | 10 | 25 |
| Timestral 123M (fine-tuned) | Foundation | 15.3 * | 24.2 | 30.5 | 34.7 | 26.3 | 8.7 | 14.3 | 21.8 | -0.6 | +1.8 | 15 | 30 |
| LSTM | Deep | 15.3 | 24.2 | 30.6 | 35.7 | 25.7 | 8.8 | 14.5 | 22.8 | +0.4 | +2.0 | 10 | 20 |
| ARX (ridge) | Classical | 15.8 | 24.3 | 28.9 * | 31.8 * | 26.5 | 9.4 | 15.2 | 21.3 | +0.9 | +3.9 | 5 | 15 |
| N-HiTS | Deep | 15.5 | 24.7 | 30.6 | 34.6 | 26.2 | 9.1 | 15.0 | 22.1 | +2.7 | +3.9 | 10 | 20 |
| Timestral 209M (fine-tuned) | Foundation | 15.3 * | 24.7 | 31.2 | 35.4 | 26.8 | 8.7 | 14.6 | 22.4 | -0.1 | +2.2 | 15 | 30 |
| Timestral 5M (fine-tuned) | Foundation | 16.2 | 25.4 | 30.8 | 34.4 | 28.1 | 9.1 | 14.9 | 22.3 | -1.6 | +1.6 | 10 | 20 |
| T1DMAI D16 DILATE (sim only) | T1DMAI | 16.2 | 26.1 | 33.2 | 39.9 | 28.6 | 9.6 | 16.0 | 25.2 | +0.6 | +3.9 | 10 | 15 |
| T1DMAI D16 MSE (sim only) | T1DMAI | 16.8 | 27.1 | 33.9 | 37.5 | 29.8 | 9.9 | 16.8 | 24.4 | +1.2 | +4.8 | 10 | 15 |
| AR (ridge) | Classical | 17.2 | 27.5 | 32.5 | 34.7 | 30.3 | 10.3 | 17.2 | 23.1 | -0.0 | +4.1 | 5 | 5 |
| Chronos-Bolt tiny (fine-tuned) | Foundation | 17.5 | 27.5 | 33.1 | 36.1 | 30.1 | 9.8 | 16.4 | 23.5 | -2.2 | +1.6 | 5 | 10 |
| Chronos-Bolt mini (fine-tuned) | Foundation | 17.5 | 27.6 | 33.1 | 36.1 | 30.1 | 9.8 | 16.5 | 23.7 | -1.8 | +1.9 | 5 | 10 |
| Timestral 123M (zero-shot) | Foundation | 16.9 | 27.8 | 34.4 | 38.2 | 30.1 | 9.6 | 16.5 | 24.1 | -2.8 | +0.7 | 5 | 10 |
| Timestral 5M (zero-shot) | Foundation | 17.1 | 28.0 | 34.4 | 37.7 | 30.4 | 9.8 | 16.7 | 24.1 | -2.7 | +1.0 | 5 | 10 |
| Timestral 209M (zero-shot) | Foundation | 17.1 | 28.4 | 35.2 | 39.2 | 30.5 | 9.6 | 16.7 | 24.6 | -2.4 | +0.9 | 5 | 10 |
| Timestral 28M (zero-shot) | Foundation | 17.2 | 28.4 | 35.3 | 39.2 | 30.6 | 9.8 | 16.8 | 24.8 | -2.8 | +0.7 | 5 | 10 |
| TimesFM 2.5 (zero-shot) | Foundation | 18.4 | 29.1 | 35.8 | 39.4 | 31.5 | 10.5 | 17.4 | 25.1 | -1.5 | +2.0 | 5 | 5 |
| Chronos-2 (zero-shot) | Foundation | 17.8 | 29.4 | 36.4 | 40.0 | 31.8 | 10.2 | 17.7 | 25.5 | -1.0 | +2.5 | 5 | 10 |
| SARIMAX | Classical | 17.7 | 29.8 | 38.7 | 45.5 | 31.6 | 10.2 | 17.9 | 28.8 | -3.8 | -0.9 | 5 | 5 |
| Chronos-Bolt tiny (zero-shot) | Foundation | 18.7 | 30.8 | 38.3 | 42.0 | 32.9 | 10.8 | 18.5 | 27.1 | +0.8 | +3.5 | 5 | 5 |
| Chronos-Bolt mini (zero-shot) | Foundation | 18.7 | 30.9 | 38.3 | 42.4 | 32.9 | 10.7 | 18.5 | 27.2 | +1.0 | +3.6 | 5 | 10 |
| Kalman filter | Classical | 18.5 | 31.3 | 39.6 | 45.0 | 33.2 | 10.6 | 18.7 | 28.6 | +0.2 | +2.7 | 0 | 5 |
| ARIMA | Classical | 18.7 | 31.5 | 39.8 | 45.1 | 33.3 | 10.7 | 18.8 | 28.7 | +0.2 | +2.7 | 0 | 5 |
| Persistence | Baseline | 20.0 | 32.1 | 39.7 | 44.7 | 34.2 | 11.7 | 19.4 | 28.5 | +0.2 | +3.3 | 0 | 0 |
| Linear extrapolation | Baseline | 24.3 | 48.9 | 71.0 | 89.2 | 50.1 | 13.3 | 27.9 | 56.4 | +0.1 | -0.3 | 10 | 15 |
| Model | DTS A 30 | DTS A 60 | DTS A 120 | DTS A+B 120 | Clarke A 60 | Clarke A+B 120 | CG-EGA hypo AP 30 | hypo EP 60 | eu EP 60 | hyper EP 60 |
|---|---|---|---|---|---|---|---|---|---|---|
| XGBoost | 91.3 | 77.9 | 62.7 | 96.9 | 78.3 | 96.6 | 78.4 | 51.3 | 0.0 | 0.1 |
| LightGBM | 91.4 | 78.0 | 62.1 | 96.8 | 78.3 | 96.6 | 77.7 | 57.7 | 0.0 | 0.1 |
| T1DMAI D16 DILATE (sim + LoRA) | 90.7 | 78.0 | 60.7 | 96.8 | 78.3 | 97.0 | 83.2 | 38.4 | 0.0 | 0.1 |
| LSTM (15 h) | 90.9 | 75.0 | 58.1 | 96.6 | 75.3 | 96.7 | 81.9 | 49.4 | 0.0 | 0.1 |
| PatchTST | 90.8 | 75.8 | 57.8 | 96.7 | 76.1 | 96.8 | 82.9 | 46.8 | 0.0 | 0.7 |
| SVR | 91.2 | 77.4 | 61.2 | 97.1 | 77.8 | 97.0 | 81.6 | 45.8 | 0.0 | 0.5 |
| T1DMAI D16 MSE (sim + LoRA) | 91.2 | 76.4 | 60.0 | 96.7 | 76.7 | 97.0 | 88.1 | 35.5 | 0.0 | 0.3 |
| Sparse GP | 90.2 | 75.2 | 58.4 | 96.8 | 75.6 | 96.8 | 76.8 | 57.4 | 0.0 | 0.3 |
| GluNet | 90.5 | 75.4 | 55.2 | 96.3 | 75.6 | 96.8 | 85.8 | 41.3 | 0.0 | 0.3 |
| MiniRocket | 89.6 | 73.5 | 59.9 | 96.5 | 73.7 | 96.6 | 68.4 | 78.1 | 0.0 | 0.8 |
| Timestral 28M (fine-tuned) | 90.9 | 76.1 | 59.0 | 96.6 | 76.3 | 96.8 | 81.6 | 43.5 | 0.0 | 0.7 |
| Timestral 123M (fine-tuned) | 91.1 | 76.0 | 57.7 | 96.1 | 76.4 | 96.8 | 83.2 | 37.4 | 0.0 | 0.4 |
| LSTM | 91.1 | 75.4 | 55.4 | 95.4 | 75.8 | 97.1 | 89.7 | 31.9 | 0.0 | 0.1 |
| ARX (ridge) | 89.9 | 73.8 | 59.1 | 96.7 | 74.1 | 96.7 | 72.9 | 69.7 | 0.0 | 0.1 |
| N-HiTS | 90.7 | 73.4 | 56.6 | 96.4 | 73.8 | 97.0 | 85.2 | 35.8 | 0.0 | 0.1 |
| Timestral 209M (fine-tuned) | 91.4 | 74.9 | 56.4 | 96.2 | 75.1 | 96.9 | 83.5 | 42.9 | 0.0 | 0.7 |
| Timestral 5M (fine-tuned) | 90.0 | 74.1 | 57.5 | 95.8 | 74.5 | 96.5 | 79.4 | 51.6 | 0.0 | 1.4 |
| T1DMAI D16 DILATE (sim only) | 89.0 | 72.1 | 50.8 | 93.0 | 72.4 | 95.8 | 73.5 | 62.6 | 0.1 | 0.4 |
| T1DMAI D16 MSE (sim only) | 88.2 | 69.7 | 54.0 | 93.9 | 70.1 | 96.2 | 69.4 | 71.9 | 0.1 | 0.5 |
| AR (ridge) | 87.1 | 67.6 | 55.4 | 95.9 | 67.7 | 96.5 | 55.5 | 96.8 | 0.0 | 0.3 |
| Chronos-Bolt tiny (fine-tuned) | 88.2 | 69.6 | 54.0 | 95.7 | 69.9 | 96.4 | 75.5 | 69.4 | 0.0 | 0.3 |
| Chronos-Bolt mini (fine-tuned) | 88.1 | 69.4 | 54.1 | 95.4 | 69.7 | 96.4 | 81.3 | 71.0 | 0.0 | 0.4 |
| Timestral 123M (zero-shot) | 88.9 | 69.2 | 52.9 | 94.9 | 69.6 | 96.3 | 82.9 | 55.5 | 0.0 | 0.1 |
| Timestral 5M (zero-shot) | 88.5 | 69.0 | 52.4 | 94.9 | 69.3 | 96.4 | 78.4 | 60.0 | 0.0 | 0.4 |
| Timestral 209M (zero-shot) | 88.7 | 68.8 | 52.1 | 94.5 | 69.2 | 96.3 | 85.8 | 51.6 | 0.0 | 0.1 |
| Timestral 28M (zero-shot) | 88.4 | 68.1 | 52.0 | 94.6 | 68.5 | 96.4 | 81.3 | 55.2 | 0.0 | 0.4 |
| TimesFM 2.5 (zero-shot) | 86.1 | 67.3 | 51.8 | 93.9 | 67.6 | 96.5 | 78.4 | 61.0 | 0.0 | 0.3 |
| Chronos-2 (zero-shot) | 87.1 | 66.6 | 51.6 | 93.8 | 66.9 | 95.8 | 82.9 | 63.5 | 0.0 | 0.4 |
| SARIMAX | 87.5 | 64.8 | 43.5 | 89.6 | 65.3 | 97.4 | 91.3 | 26.5 | 0.0 | 0.3 |
| Chronos-Bolt tiny (zero-shot) | 86.0 | 64.4 | 49.5 | 92.4 | 64.7 | 95.8 | 79.4 | 53.9 | 0.0 | 0.3 |
| Chronos-Bolt mini (zero-shot) | 86.2 | 64.6 | 49.9 | 92.3 | 64.9 | 95.6 | 82.6 | 54.5 | 0.0 | 0.3 |
| Kalman filter | 86.1 | 64.0 | 47.1 | 90.8 | 64.3 | 96.1 | 85.2 | 46.8 | 0.1 | 0.3 |
| ARIMA | 85.9 | 63.9 | 47.0 | 90.8 | 64.3 | 96.0 | 84.5 | 46.5 | 0.1 | 0.3 |
| Persistence | 82.9 | 62.5 | 47.5 | 91.0 | 63.1 | 96.1 | 79.7 | 51.3 | 0.0 | 0.3 |
| Linear extrapolation | 78.7 | 50.5 | 25.9 | 65.9 | 51.3 | 83.2 | 91.9 | 23.2 | 2.0 | 2.6 |
| Model | Sens 30 | Sens 60 | Spec 60 | PPV 60 | MCC 30 | MCC 60 | RMSE<70 30 | ME<70 30 | ME<70 60 |
|---|---|---|---|---|---|---|---|---|---|
| XGBoost | 0.49 | 0.28 | 0.996 | 0.71 | 0.63 | 0.43 | 12.2 | +8.3 | +17.1 |
| LightGBM | 0.52 | 0.24 | 0.996 | 0.64 | 0.64 | 0.38 | 12.3 | +8.7 | +17.8 |
| T1DMAI D16 DILATE (sim + LoRA) | 0.63 | 0.40 | 0.993 | 0.65 | 0.69 | 0.50 | 11.3 | +6.2 | +13.0 |
| LSTM (15 h) | 0.56 | 0.35 | 0.992 | 0.58 | 0.63 | 0.44 | 10.9 | +6.2 | +13.9 |
| PatchTST | 0.59 | 0.40 | 0.990 | 0.54 | 0.64 | 0.45 | 11.2 | +4.7 | +11.5 |
| SVR | 0.59 | 0.34 | 0.992 | 0.55 | 0.64 | 0.42 | 11.3 | +6.6 | +14.1 |
| T1DMAI D16 MSE (sim + LoRA) | 0.66 | 0.47 | 0.992 | 0.64 | 0.70 | 0.54 | 10.3 | +4.7 | +10.9 |
| Sparse GP | 0.53 | 0.24 | 0.995 | 0.61 | 0.61 | 0.37 | 12.6 | +8.8 | +17.9 |
| GluNet | 0.66 | 0.50 | 0.987 | 0.55 | 0.67 | 0.51 | 11.0 | +3.6 | +9.0 |
| MiniRocket | 0.38 | 0.16 | 0.997 | 0.61 | 0.51 | 0.30 | 14.3 | +11.0 | +23.7 |
| Timestral 28M (fine-tuned) | 0.63 | 0.42 | 0.990 | 0.58 | 0.65 | 0.48 | 10.9 | +5.7 | +13.4 |
| Timestral 123M (fine-tuned) | 0.65 | 0.50 | 0.991 | 0.62 | 0.68 | 0.54 | 11.4 | +5.6 | +13.4 |
| LSTM | 0.70 | 0.53 | 0.983 | 0.49 | 0.66 | 0.49 | 10.1 | +1.4 | +5.0 |
| ARX (ridge) | 0.41 | 0.20 | 0.995 | 0.53 | 0.51 | 0.31 | 12.9 | +9.5 | +20.7 |
| N-HiTS | 0.65 | 0.51 | 0.987 | 0.54 | 0.66 | 0.51 | 11.6 | +3.3 | +8.6 |
| Timestral 209M (fine-tuned) | 0.63 | 0.48 | 0.993 | 0.67 | 0.66 | 0.55 | 10.4 | +5.4 | +12.8 |
| Timestral 5M (fine-tuned) | 0.53 | 0.36 | 0.992 | 0.57 | 0.58 | 0.44 | 12.6 | +7.4 | +17.8 |
| T1DMAI D16 DILATE (sim only) | 0.51 | 0.21 | 0.997 | 0.69 | 0.64 | 0.37 | 13.2 | +9.2 | +20.2 |
| T1DMAI D16 MSE (sim only) | 0.42 | 0.14 | 0.997 | 0.62 | 0.58 | 0.28 | 14.1 | +10.5 | +22.3 |
| AR (ridge) | 0.23 | 0.00 | 0.999 | 0.00 | 0.33 | -0.01 | 15.8 | +13.9 | +30.3 |
| Chronos-Bolt tiny (fine-tuned) | 0.49 | 0.12 | 0.991 | 0.30 | 0.51 | 0.18 | 12.5 | +8.6 | +21.8 |
| Chronos-Bolt mini (fine-tuned) | 0.53 | 0.10 | 0.991 | 0.25 | 0.54 | 0.14 | 11.9 | +7.9 | +22.1 |
| Timestral 123M (zero-shot) | 0.58 | 0.31 | 0.983 | 0.36 | 0.55 | 0.31 | 11.6 | +6.0 | +17.4 |
| Timestral 5M (zero-shot) | 0.53 | 0.27 | 0.987 | 0.39 | 0.54 | 0.31 | 12.4 | +7.6 | +18.9 |
| Timestral 209M (zero-shot) | 0.61 | 0.37 | 0.980 | 0.37 | 0.56 | 0.35 | 11.1 | +5.3 | +15.9 |
| Timestral 28M (zero-shot) | 0.59 | 0.33 | 0.982 | 0.36 | 0.54 | 0.33 | 11.6 | +6.2 | +17.1 |
| TimesFM 2.5 (zero-shot) | 0.54 | 0.27 | 0.987 | 0.39 | 0.55 | 0.31 | 13.3 | +7.0 | +19.1 |
| Chronos-2 (zero-shot) | 0.59 | 0.24 | 0.983 | 0.31 | 0.53 | 0.25 | 11.7 | +5.7 | +21.5 |
| SARIMAX | 0.72 | 0.58 | 0.963 | 0.33 | 0.60 | 0.41 | 9.1 | +2.0 | +6.7 |
| Chronos-Bolt tiny (zero-shot) | 0.55 | 0.34 | 0.980 | 0.34 | 0.51 | 0.32 | 12.3 | +6.4 | +17.8 |
| Chronos-Bolt mini (zero-shot) | 0.60 | 0.29 | 0.982 | 0.34 | 0.54 | 0.30 | 11.9 | +5.8 | +18.5 |
| Kalman filter | 0.63 | 0.40 | 0.974 | 0.33 | 0.56 | 0.34 | 10.7 | +5.6 | +15.0 |
| ARIMA | 0.62 | 0.40 | 0.974 | 0.32 | 0.56 | 0.34 | 10.8 | +5.8 | +14.9 |
| Persistence | 0.55 | 0.36 | 0.981 | 0.36 | 0.54 | 0.34 | 12.8 | +8.1 | +18.1 |
| Linear extrapolation | 0.79 | 0.72 | 0.872 | 0.15 | 0.47 | 0.29 | 11.1 | -1.8 | -3.5 |
9661 forecasts at 60 min. * means lowest RMSE, or not significantly worse than it (Wilcoxon signed-rank test on the mean absolute-error difference per day, alpha 0.05/34).
MetaboNet| Model | Family | RMSE 30 | RMSE 60 | RMSE 90 | RMSE 120 | gRMSE 60 | MARD 30 | MARD 60 | MARD 120 | ME 60 | MRE 60 % |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Timestral 209M (fine-tuned) | Foundation | 19.4 * | 32.6 * | 40.3 * | 44.7 * | 39.0 | 10.0 | 17.2 | 24.0 | -2.4 | +2.9 |
| Timestral 123M (fine-tuned) | Foundation | 19.5 | 32.9 | 40.5 | 45.0 | 39.3 | 10.0 | 17.2 | 24.1 | -3.0 | +2.4 |
| XGBoost | Classical | 19.8 | 33.2 | 41.3 | 46.0 | 40.3 | 10.6 | 18.6 | 27.1 | +0.5 | +6.3 |
| LightGBM | Classical | 19.8 | 33.2 | 41.3 | 46.0 | 40.3 | 10.6 | 18.7 | 27.1 | +0.6 | +6.4 |
| Timestral 28M (fine-tuned) | Foundation | 19.8 | 33.3 | 41.0 | 45.5 | 39.9 | 10.2 | 17.4 | 24.3 | -3.1 | +2.3 |
| Sparse GP | Classical | 20.4 | 33.7 | 41.6 | 46.2 | 40.8 | 10.9 | 18.9 | 27.2 | +0.4 | +6.2 |
| Timestral 5M (fine-tuned) | Foundation | 20.0 | 33.8 | 41.6 | 46.0 | 40.4 | 10.3 | 17.7 | 24.6 | -3.2 | +2.3 |
| LSTM (15 h) | Deep | 20.6 | 33.9 | 41.8 | 46.5 | 40.7 | 10.9 | 18.7 | 27.0 | -0.1 | +5.3 |
| SVR | Classical | 20.4 | 34.0 | 42.0 | 46.8 | 41.0 | 10.6 | 18.0 | 25.2 | -3.3 | +2.8 |
| N-HiTS | Deep | 20.6 | 34.3 | 42.7 | 47.8 | 40.8 | 11.0 | 18.9 | 27.5 | +0.5 | +5.4 |
| PatchTST | Deep | 20.4 | 34.5 | 42.9 | 48.0 | 40.7 | 10.7 | 19.0 | 27.7 | +0.8 | +5.6 |
| Chronos-Bolt mini (fine-tuned) | Foundation | 21.0 | 34.5 | 42.0 | 46.2 | 40.8 | 10.8 | 18.1 | 25.0 | -2.1 | +3.0 |
| MiniRocket | Classical | 21.5 | 34.6 | 42.1 | 46.4 | 41.8 | 11.6 | 19.4 | 27.2 | +0.3 | +6.3 |
| T1DMAI D16 DILATE (sim + LoRA) | T1DMAI | 21.5 | 34.6 | 41.0 | 50.4 | 41.2 | 10.9 | 18.0 | 25.2 | -3.5 | +2.0 |
| Chronos-Bolt tiny (fine-tuned) | Foundation | 21.1 | 34.6 | 42.1 | 46.2 | 41.0 | 10.9 | 18.2 | 25.0 | -2.3 | +2.9 |
| GluNet | Deep | 20.5 | 34.8 | 43.6 | 49.0 | 41.1 | 10.8 | 18.9 | 27.8 | +0.4 | +5.0 |
| Timestral 5M (zero-shot) | Foundation | 20.5 | 35.0 | 43.5 | 48.5 | 41.4 | 10.5 | 18.2 | 25.8 | -2.8 | +2.2 |
| TimesFM 2.5 (zero-shot) | Foundation | 20.7 | 35.0 | 43.3 | 47.9 | 41.4 | 10.7 | 18.6 | 26.1 | -1.7 | +3.2 |
| T1DMAI D16 MSE (sim + LoRA) | T1DMAI | 21.8 | 35.1 | 41.0 | 46.8 | 41.3 | 11.2 | 18.1 | 25.1 | -3.8 | +1.3 |
| Timestral 123M (zero-shot) | Foundation | 20.5 | 35.1 | 43.6 | 48.5 | 41.4 | 10.5 | 18.2 | 25.7 | -3.2 | +1.8 |
| Timestral 209M (zero-shot) | Foundation | 20.4 | 35.1 | 43.7 | 48.7 | 41.2 | 10.5 | 18.3 | 26.1 | -2.3 | +2.4 |
| Timestral 28M (zero-shot) | Foundation | 20.5 | 35.2 | 43.8 | 48.9 | 41.4 | 10.5 | 18.2 | 25.9 | -2.9 | +1.9 |
| Chronos-2 (zero-shot) | Foundation | 21.0 | 35.3 | 43.4 | 48.0 | 41.6 | 10.8 | 18.7 | 26.2 | -1.3 | +3.6 |
| ARX (ridge) | Classical | 21.3 | 35.4 | 43.4 | 47.9 | 42.3 | 11.6 | 20.4 | 29.4 | +2.5 | +8.0 |
| AR (ridge) | Classical | 21.3 | 35.5 | 43.6 | 48.0 | 42.4 | 11.7 | 20.6 | 29.9 | +3.1 | +8.4 |
| LSTM | Deep | 21.2 | 35.9 | 45.2 | 51.0 | 41.9 | 11.1 | 19.4 | 28.6 | +0.5 | +4.7 |
| T1DMAI D16 DILATE (sim only) | T1DMAI | 23.4 | 36.6 | 44.0 | 48.0 | 44.0 | 12.2 | 20.4 | 28.8 | +0.0 | +6.0 |
| T1DMAI D16 MSE (sim only) | T1DMAI | 23.4 | 36.7 | 44.1 | 48.2 | 43.8 | 13.1 | 21.7 | 29.7 | +4.4 | +9.7 |
| SARIMAX | Classical | 21.8 | 36.9 | 46.4 | 52.3 | 42.9 | 11.5 | 20.4 | 30.3 | +1.5 | +5.7 |
| Chronos-Bolt tiny (zero-shot) | Foundation | 22.8 | 37.7 | 46.4 | 51.3 | 43.3 | 11.9 | 20.1 | 28.3 | +0.7 | +4.4 |
| Chronos-Bolt mini (zero-shot) | Foundation | 22.9 | 37.7 | 46.3 | 51.4 | 43.4 | 11.8 | 19.9 | 28.2 | +0.1 | +3.8 |
| Kalman filter | Classical | 22.2 | 38.4 | 48.9 | 55.6 | 43.9 | 11.4 | 20.3 | 30.3 | +0.2 | +3.5 |
| ARIMA | Classical | 22.5 | 39.3 | 50.2 | 57.4 | 44.7 | 11.6 | 20.9 | 31.5 | +0.2 | +3.4 |
| Persistence | Baseline | 25.2 | 40.0 | 49.4 | 55.5 | 46.2 | 13.1 | 21.3 | 30.3 | +0.3 | +4.4 |
| Linear extrapolation | Baseline | 30.7 | 59.2 | 81.8 | 98.8 | 64.0 | 15.6 | 31.4 | 56.8 | +0.1 | +0.7 |
| Model | DTS A 30 | DTS A 60 | DTS A 120 | DTS A+B 120 | Clarke A 60 | Clarke A+B 120 | CG-EGA hypo AP 30 | hypo EP 60 | eu EP 60 | hyper EP 60 |
|---|---|---|---|---|---|---|---|---|---|---|
| Timestral 209M (fine-tuned) | 87.2 | 69.7 | 54.8 | 94.4 | 69.9 | 93.5 | 65.1 | 78.6 | 0.1 | 2.9 |
| Timestral 123M (fine-tuned) | 87.2 | 69.6 | 54.6 | 94.3 | 69.8 | 93.5 | 67.5 | 76.8 | 0.1 | 2.9 |
| XGBoost | 85.9 | 67.0 | 49.8 | 93.0 | 67.1 | 93.1 | 33.1 | 99.6 | 0.1 | 3.1 |
| LightGBM | 85.9 | 66.9 | 49.8 | 93.0 | 67.0 | 93.1 | 33.2 | 99.5 | 0.1 | 3.1 |
| Timestral 28M (fine-tuned) | 86.8 | 69.0 | 54.0 | 94.2 | 69.1 | 93.4 | 67.4 | 76.5 | 0.1 | 3.1 |
| Sparse GP | 85.3 | 66.4 | 49.8 | 92.9 | 66.5 | 93.1 | 43.3 | 98.1 | 0.1 | 3.1 |
| Timestral 5M (fine-tuned) | 86.6 | 68.3 | 53.2 | 93.9 | 68.6 | 93.3 | 68.8 | 76.1 | 0.1 | 3.3 |
| LSTM (15 h) | 85.2 | 66.6 | 49.8 | 93.0 | 66.8 | 93.3 | 53.9 | 86.0 | 0.1 | 2.9 |
| SVR | 86.0 | 68.0 | 52.1 | 93.6 | 68.2 | 92.9 | 62.4 | 89.4 | 0.1 | 3.3 |
| N-HiTS | 85.1 | 66.3 | 49.3 | 92.4 | 66.5 | 93.3 | 57.4 | 80.0 | 0.1 | 2.9 |
| PatchTST | 85.7 | 65.9 | 48.8 | 92.3 | 66.1 | 93.3 | 63.0 | 80.8 | 0.1 | 2.6 |
| Chronos-Bolt mini (fine-tuned) | 85.3 | 67.7 | 53.1 | 93.7 | 67.9 | 93.3 | 64.1 | 78.7 | 0.1 | 3.0 |
| MiniRocket | 83.6 | 65.0 | 49.6 | 92.9 | 65.1 | 93.2 | 36.2 | 96.2 | 0.1 | 3.3 |
| T1DMAI D16 DILATE (sim + LoRA) | 84.9 | 67.8 | 49.5 | 93.1 | 68.0 | 91.8 | 71.7 | 79.8 | 0.1 | 3.2 |
| Chronos-Bolt tiny (fine-tuned) | 85.1 | 67.5 | 53.0 | 93.7 | 67.7 | 93.3 | 59.8 | 78.6 | 0.1 | 3.1 |
| GluNet | 85.5 | 66.2 | 48.5 | 92.0 | 66.4 | 93.1 | 65.0 | 76.2 | 0.1 | 2.8 |
| Timestral 5M (zero-shot) | 86.0 | 67.1 | 51.5 | 92.6 | 67.4 | 93.3 | 69.0 | 71.6 | 0.1 | 3.4 |
| TimesFM 2.5 (zero-shot) | 85.6 | 66.6 | 51.3 | 92.7 | 66.8 | 93.3 | 67.6 | 73.8 | 0.1 | 3.1 |
| T1DMAI D16 MSE (sim + LoRA) | 84.4 | 67.7 | 51.5 | 94.0 | 67.9 | 92.4 | 68.5 | 74.8 | 0.1 | 2.9 |
| Timestral 123M (zero-shot) | 86.1 | 67.2 | 51.6 | 92.7 | 67.4 | 93.2 | 71.4 | 69.1 | 0.1 | 3.4 |
| Timestral 209M (zero-shot) | 86.1 | 67.1 | 51.2 | 92.4 | 67.3 | 93.2 | 71.2 | 69.2 | 0.1 | 3.2 |
| Timestral 28M (zero-shot) | 86.0 | 67.1 | 51.3 | 92.5 | 67.3 | 93.2 | 71.4 | 68.6 | 0.1 | 3.3 |
| Chronos-2 (zero-shot) | 85.3 | 66.3 | 51.0 | 92.7 | 66.6 | 93.2 | 68.9 | 74.6 | 0.1 | 3.3 |
| ARX (ridge) | 83.9 | 62.2 | 45.5 | 92.0 | 62.4 | 93.2 | 52.0 | 93.3 | 0.1 | 2.9 |
| AR (ridge) | 83.7 | 61.8 | 45.1 | 91.7 | 62.0 | 93.2 | 51.7 | 93.6 | 0.1 | 2.8 |
| LSTM | 84.8 | 65.0 | 47.7 | 91.1 | 65.3 | 92.8 | 67.5 | 71.3 | 0.2 | 2.8 |
| T1DMAI D16 DILATE (sim only) | 82.3 | 62.4 | 45.3 | 92.4 | 62.6 | 92.7 | 56.0 | 86.3 | 0.1 | 3.3 |
| T1DMAI D16 MSE (sim only) | 80.1 | 59.2 | 45.0 | 91.8 | 59.3 | 93.3 | 34.6 | 93.9 | 0.1 | 3.1 |
| SARIMAX | 83.9 | 63.6 | 46.9 | 89.9 | 63.8 | 92.7 | 62.7 | 75.6 | 0.2 | 2.6 |
| Chronos-Bolt tiny (zero-shot) | 82.8 | 63.9 | 49.2 | 90.9 | 64.1 | 92.8 | 66.9 | 67.9 | 0.3 | 3.0 |
| Chronos-Bolt mini (zero-shot) | 83.1 | 64.2 | 49.4 | 91.0 | 64.4 | 92.7 | 70.7 | 66.9 | 0.3 | 3.1 |
| Kalman filter | 83.9 | 63.3 | 46.3 | 88.5 | 63.6 | 92.2 | 74.7 | 61.0 | 0.3 | 2.9 |
| ARIMA | 83.5 | 61.8 | 44.4 | 87.6 | 62.1 | 91.8 | 75.2 | 59.9 | 0.3 | 3.0 |
| Persistence | 79.6 | 61.0 | 46.1 | 88.7 | 61.5 | 92.1 | 57.3 | 68.3 | 0.2 | 3.4 |
| Linear extrapolation | 74.9 | 48.6 | 26.8 | 67.0 | 49.4 | 78.0 | 83.5 | 34.3 | 3.6 | 7.8 |
| Model | Sens 30 | Sens 60 | Spec 60 | PPV 60 | MCC 30 | MCC 60 | RMSE<70 30 | ME<70 30 | ME<70 60 |
|---|---|---|---|---|---|---|---|---|---|
| Timestral 209M (fine-tuned) | 0.38 | 0.10 | 0.995 | 0.40 | 0.43 | 0.19 | 21.5 | +14.3 | +33.8 |
| Timestral 123M (fine-tuned) | 0.41 | 0.11 | 0.994 | 0.37 | 0.44 | 0.19 | 20.9 | +13.6 | +32.6 |
| XGBoost | 0.06 | 0.00 | 1.000 | 0.71 | 0.20 | 0.03 | 26.4 | +22.3 | +45.7 |
| LightGBM | 0.07 | 0.00 | 1.000 | 0.65 | 0.21 | 0.03 | 26.5 | +22.4 | +45.8 |
| Timestral 28M (fine-tuned) | 0.41 | 0.11 | 0.993 | 0.35 | 0.44 | 0.18 | 21.0 | +13.5 | +32.5 |
| Sparse GP | 0.18 | 0.00 | 1.000 | 0.30 | 0.29 | 0.03 | 25.8 | +20.4 | +44.4 |
| Timestral 5M (fine-tuned) | 0.43 | 0.12 | 0.992 | 0.32 | 0.43 | 0.18 | 20.7 | +12.9 | +32.2 |
| LSTM (15 h) | 0.31 | 0.05 | 0.997 | 0.32 | 0.37 | 0.12 | 24.9 | +17.7 | +39.0 |
| SVR | 0.34 | 0.03 | 0.998 | 0.34 | 0.39 | 0.09 | 22.6 | +15.6 | +37.6 |
| N-HiTS | 0.34 | 0.10 | 0.992 | 0.28 | 0.39 | 0.15 | 24.4 | +16.3 | +37.0 |
| PatchTST | 0.38 | 0.09 | 0.992 | 0.27 | 0.41 | 0.14 | 23.0 | +14.7 | +36.3 |
| Chronos-Bolt mini (fine-tuned) | 0.38 | 0.09 | 0.994 | 0.34 | 0.40 | 0.16 | 22.2 | +14.6 | +34.1 |
| MiniRocket | 0.16 | 0.01 | 0.999 | 0.28 | 0.25 | 0.05 | 27.5 | +22.0 | +44.7 |
| T1DMAI D16 DILATE (sim + LoRA) | 0.42 | 0.06 | 0.997 | 0.38 | 0.41 | 0.15 | 20.6 | +13.3 | +33.4 |
| Chronos-Bolt tiny (fine-tuned) | 0.35 | 0.09 | 0.993 | 0.31 | 0.39 | 0.15 | 22.8 | +15.7 | +34.1 |
| GluNet | 0.41 | 0.13 | 0.988 | 0.26 | 0.42 | 0.16 | 22.8 | +13.9 | +34.9 |
| Timestral 5M (zero-shot) | 0.44 | 0.16 | 0.985 | 0.26 | 0.43 | 0.18 | 20.8 | +12.4 | +30.7 |
| TimesFM 2.5 (zero-shot) | 0.44 | 0.14 | 0.987 | 0.26 | 0.43 | 0.17 | 21.4 | +12.8 | +32.6 |
| T1DMAI D16 MSE (sim + LoRA) | 0.40 | 0.08 | 0.996 | 0.37 | 0.40 | 0.16 | 21.7 | +14.2 | +32.1 |
| Timestral 123M (zero-shot) | 0.47 | 0.18 | 0.983 | 0.25 | 0.44 | 0.19 | 20.4 | +11.4 | +29.7 |
| Timestral 209M (zero-shot) | 0.47 | 0.18 | 0.983 | 0.25 | 0.44 | 0.19 | 20.5 | +11.5 | +30.0 |
| Timestral 28M (zero-shot) | 0.47 | 0.18 | 0.982 | 0.25 | 0.44 | 0.19 | 20.3 | +11.3 | +29.5 |
| Chronos-2 (zero-shot) | 0.46 | 0.14 | 0.986 | 0.26 | 0.43 | 0.17 | 21.6 | +12.0 | +33.0 |
| ARX (ridge) | 0.28 | 0.03 | 0.996 | 0.19 | 0.33 | 0.07 | 24.3 | +17.4 | +42.6 |
| AR (ridge) | 0.28 | 0.03 | 0.996 | 0.19 | 0.33 | 0.06 | 24.0 | +17.4 | +42.9 |
| LSTM | 0.45 | 0.17 | 0.981 | 0.22 | 0.42 | 0.17 | 22.8 | +13.1 | +33.0 |
| T1DMAI D16 DILATE (sim only) | 0.32 | 0.07 | 0.991 | 0.20 | 0.34 | 0.10 | 24.9 | +16.9 | +39.3 |
| T1DMAI D16 MSE (sim only) | 0.18 | 0.03 | 0.997 | 0.22 | 0.27 | 0.07 | 28.6 | +22.5 | +45.1 |
| SARIMAX | 0.40 | 0.14 | 0.984 | 0.22 | 0.39 | 0.15 | 23.3 | +14.3 | +36.5 |
| Chronos-Bolt tiny (zero-shot) | 0.46 | 0.20 | 0.976 | 0.21 | 0.41 | 0.18 | 23.2 | +12.7 | +31.9 |
| Chronos-Bolt mini (zero-shot) | 0.49 | 0.20 | 0.976 | 0.21 | 0.41 | 0.18 | 22.4 | +11.5 | +31.0 |
| Kalman filter | 0.52 | 0.26 | 0.963 | 0.19 | 0.43 | 0.19 | 20.7 | +9.8 | +28.3 |
| ARIMA | 0.54 | 0.28 | 0.959 | 0.18 | 0.43 | 0.20 | 20.7 | +9.3 | +28.0 |
| Persistence | 0.37 | 0.19 | 0.975 | 0.19 | 0.35 | 0.16 | 26.1 | +16.3 | +33.6 |
| Linear extrapolation | 0.72 | 0.58 | 0.861 | 0.12 | 0.41 | 0.21 | 27.2 | -1.4 | +12.3 |
264893 forecasts at 60 min. * means lowest RMSE, or not significantly worse than it (Wilcoxon signed-rank test on the mean absolute-error difference per subject, alpha 0.05/34).
Raw numbers are in results/.
python -m zoo.run MODEL --cache CACHE --template template.parquet --truth truth.parquet [--causal] python finetune.py --checkpoint CKPT --cache CACHE --lora-rank 1 --val-days 10 --select-on rmse \ [--blind | --no-blind] python export_submission.py --checkpoint CKPT --template template.parquet --cache CACHE \ --max-context 336 --fill ladder --quantiles [--future-doses zero] python leaderboard.py MANIFEST.json --out results
Chronos-2, TimesFM 2.5 and Timestral run under .venv-zoo (requirements-zoo.txt).
RMSE (mg/dL) that each model's paper reports, beside the same model on my data and on MetaboNet. The papers use other people, splits, inputs and history lengths, so a gap between the two sides is not a like-for-like comparison. My data is non-blind, MetaboNet blind (Terms).
| Model | Paper | Paper data | Paper inputs | Paper 30 | Paper 60 | My data 30 | My data 60 | MetaboNet 30 | MetaboNet 60 |
|---|---|---|---|---|---|---|---|---|---|
| XGBoost | 7 | OhioT1DM 2018 (6) | CGM, doses, meals, activity | 20.4 | - | 14.6 | 22.4 | 19.8 | 33.2 |
| LightGBM | 3 | MetaboNet | CGM, carbs, insulin, time | 20.8 | 34.5 | 14.6 | 22.4 | 19.8 | 33.2 |
| T1DMAI D16 DILATE (sim + LoRA) | - | - | - | - | - | 15.2 | 22.8 | 21.5 | 34.6 |
| LSTM (15 h) | 3 | MetaboNet | CGM, carbs, insulin | 20.1 | 33.4 | 15.1 | 23.2 | 20.6 | 33.9 |
| PatchTST | 9 | DCLP3 train, OhioT1DM test (12) | CGM, carbs, insulin | 16.7 | 24.6 | 15.1 | 23.3 | 20.4 | 34.5 |
| SVR | 6 | OhioT1DM 2018 (6) | CGM, carbs, insulin | 21.8 | 34.3 | 15.2 | 23.4 | 20.4 | 34.0 |
| T1DMAI D16 MSE (sim + LoRA) | - | - | - | - | - | 15.1 | 23.5 | 21.8 | 35.1 |
| GluNet | 6 | OhioT1DM 2018 (6) | CGM, carbs, insulin | 19.3 | 31.8 | 15.4 | 24.0 | 20.5 | 34.8 |
| LSTM | 5 | OhioT1DM 2018 (6) | CGM | 18.9 | 31.4 | 15.3 | 24.2 | 21.2 | 35.9 |
| ARX | 6 | OhioT1DM 2018 (6) | CGM, carbs, insulin | 21.5 | 39.5 | 15.8 | 24.3 | 21.3 | 35.4 |
| T1DMAI D16 DILATE (sim only) | - | - | - | - | - | 16.2 | 26.1 | 23.4 | 36.6 |
| T1DMAI D16 MSE (sim only) | - | - | - | - | - | 16.8 | 27.1 | 23.4 | 36.7 |
| TimesFM 2.5 (zero-shot) | 4 | 12 datasets, mixed | CGM | 18.8 | 31.8 | 18.4 | 29.1 | 20.7 | 35.0 |
| Chronos-2 (zero-shot) | 4 | 12 datasets, mixed | CGM | 19.2 | 32.0 | 17.8 | 29.4 | 21.0 | 35.3 |
| ARIMA | 8 | OhioT1DM 2020 (6) | CGM, doses, meals, exercise | 23.7 | 36.4 | 18.7 | 31.5 | 22.5 | 39.3 |
| Persistence | 3 | MetaboNet | CGM | 26.3 | 41.5 | 20.0 | 32.1 | 25.2 | 40.0 |
RMSE in mg/dL at the single step 30 or 60 min ahead, as printed in each paper. T1DMAI has no published score.
Chronos-Bolt (Zhang et al. 2026) and N-HiTS (Potosnak et al. 2025; Sergazinov et al. 2024) are published only as RMSE averaged over every step up to the horizon, which is not the same quantity, so they have no row.
Copyright 2026 0xdeadf1sh. MIT. See also T1DMSIM, T1DMDROID and T1DMCOMMON.
Texto de github.com tal como lo publicó, leído de su página y mostrado aquí con el formato de El Autómata. El original, en github.com · Volver a la noticia de El Autómata