๐ญ Model Observatory
37 ML models ยท ensemble voting ยท Darwin tracks accuracy ยท GPU retraining weekly
Total Models
37
ensemble running
Avg Accuracy
โ%
across all 37
Best Model
โ
loading...
Drift Alerts
0
all stable
Transformers7 models
TFT
loading
PatchTST
loading
iTransformer
loading
Autoformer
loading
Informer
loading
Crossformer
loading
TimesNet
loading
Recurrent (RNN)4 models
BiLSTM
loading
GRU
loading
Neural ODE
loading
Liquid NN
loading
CNN4 models
CNN-1D
loading
CNN-2D
loading
WaveNet
loading
TCN
loading
Gradient Boosting5 models
XGBoost
loading
LightGBM
loading
CatBoost
loading
Random Forest
loading
Extra Trees
loading
Reinforcement Learning4 models
PPO
loading
SAC
loading
TD3
loading
A2C
loading
Graph Neural Networks3 models
GCN
loading
GAT
loading
Temporal GNN
loading
Statistical4 models
Ridge
loading
Elastic Net
loading
NBEATS
loading
DLinear
loading
Anomaly Detection3 models
VAE
loading
LSTM-AE
loading
Isolation Forest
loading
NLP / Sentiment3 models
FinBERT
loading
RoBERTa
loading
VADER
loading
๐ Accuracy Leaderboard โ Top 10Updated by Darwin hourly
| # | Model | Accuracy | Bar | Predictions Today | Drift | Last Eval |
|---|---|---|---|---|---|---|
| Loading model data... | ||||||
๐ฌ Ensemble Architecture
Layer 1: 37 independent models predict direction (โ1 to +1) concurrently
Layer 2: Weighted meta-learner stacks outputs (weights updated weekly by Darwin)
Layer 3: Temperature scaling calibrates final probability
Output: Signal type + calibrated confidence [0โ1] + SHAP feature attribution
Training: Modal Labs A100 GPU ยท 5-fold time-series cross-validation ยท Optuna HPO
Data: 5 years OHLCV + 400+ features per symbol ยท Updated weekly
Layer 2: Weighted meta-learner stacks outputs (weights updated weekly by Darwin)
Layer 3: Temperature scaling calibrates final probability
Output: Signal type + calibrated confidence [0โ1] + SHAP feature attribution
Training: Modal Labs A100 GPU ยท 5-fold time-series cross-validation ยท Optuna HPO
Data: 5 years OHLCV + 400+ features per symbol ยท Updated weekly
๐ง SHAP Feature Attribution โ Oracle (Latest RELIANCE Signal)shapley values ยท additive explanation
Base value: 50%Output: 81.0% BUY confidence