v1: configs + metrics + GBM + figures

.gitattributes

@@ -33,3 +33,11 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+catboost.cbm filter=lfs diff=lfs merge=lfs -text
+figures/fig_dl_02_lorenz_curves.png filter=lfs diff=lfs merge=lfs -text
+figures/fig_dl_03_training_curves.png filter=lfs diff=lfs merge=lfs -text
+figures/fig_dl_08_calibration_deciles.png filter=lfs diff=lfs merge=lfs -text
+figures/fig_dl_09_actual_vs_predicted.png filter=lfs diff=lfs merge=lfs -text
+figures/fig_dl_11_ensemble_variance.png filter=lfs diff=lfs merge=lfs -text
+figures/fig_dl_12_pdp_top6.png filter=lfs diff=lfs merge=lfs -text
+figures/fig_dl_17_combined_dashboard.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,274 @@
+---
+license: mit
+tags:
+  - tabular
+  - regression
+  - catboost
+  - xgboost
+  - cann
+  - ft-transformer
+  - tabm
+  - localglmnet
+  - drn
+  - ensemble
+  - allstate
+  - insurance
+  - claims-severity
+  - actuarial
+  - kaggle
+library_name: catboost
+pipeline_tag: tabular-regression
+metrics:
+  - gini
+  - mae
+  - rmse
+model-index:
+  - name: allstate-tabular-models
+    results:
+      - task:
+          type: tabular-regression
+          name: Tabular Regression (Gamma)
+        dataset:
+          type: kaggle/allstate-claims-severity
+          name: Allstate Claims Severity (Kaggle competition)
+        metrics:
+          - type: gini
+            value: 0.3473
+            name: Test Gini (CANN-GBM, best)
+          - type: mae
+            value: 1158
+            name: Test MAE (CANN-GBM, USD)
+---
+
+# Allstate Claims Severity - Tabular Models (8 architectures, Gamma)
+
+Pre-trained models for the
+[Allstate Claims Severity Kaggle competition](https://www.kaggle.com/c/allstate-claims-severity),
+covering all **eight** architectures from the
+[tabular-data-modelling-pipeline](https://github.com/timothy22000/tabular_data_modelling_pipeline)
+on the full 188,318-row training set.
+
+> **Dataset not redistributed here.** The Allstate competition rules
+> restrict redistribution to non-commercial use. To reproduce these
+> models, download the data from Kaggle yourself (see below) - this repo
+> ships the **config file**, **pre-trained weights**, and **baseline
+> metrics** only.
+
+## Results
+
+All 8 architectures + NNLS-stacked ensemble, default hyperparameters (no
+Optuna tuning), 3-seed ensembles per DL architecture, gamma family + log
+link, 80/20 random split (seed 42).
+
+| Rank | Model | Test Gini | Test MAE (USD) | Test RMSE | A/E ratio | n params | Training time |
+|---:|---|---:|---:|---:|---:|---:|---:|
+| 1 | **CANN-GBM** | **0.3473** | 1,158 | 1,839 | 1.010 | 300,519 | 14.8 min |
+| - | Stacked ensemble (NNLS) | 0.3472 | **1,144** | 1,864 | 1.063 | (9 weights) | - |
+| 2 | XGBoost | 0.3468 | 1,152 | 1,850 | 1.027 | 778 trees | 0.7 min |
+| 3 | CatBoost | 0.3461 | 1,165 | 1,856 | 1.015 | 946 trees | 2.3 min |
+| 4 | CANN | 0.3457 | 1,168 | 1,857 | 1.016 | 300,519 | 15.3 min |
+| 5 | DRN | 0.3454 | 1,175 | 1,861 | **1.004** | 300,714 | 14.4 min |
+| 6 | LocalGLMnet | 0.3428 | 1,197 | 1,903 | 1.008 | 193,134 | 58.4 min |
+| 7 | TabM | 0.3427 | 1,393 | 2,446 | 1.553 | 1,735,956 | 52.9 min |
+| 8 | FT-Transformer† | 0.0279 | 2,142 | 3,508 | 3.052 | 700,611 | 231 min |
+
+† **FT-Transformer underfit.** Despite 188k training rows, the
+transformer architecture failed to converge under default hyperparameters
+within the early-stopping window. Predictions are off by a calibration
+factor of ~3. Two of its three ensemble members hit a flat local minimum
+near the global mean; the third descended properly but couldn't recover
+the ensemble. We ship the weights for completeness but **do not
+recommend using FT-T predictions from this collection** - retrain with
+Optuna tuning if you need a competitive transformer baseline.
+
+- **Test set:** 37,664 rows (20% of 188,318)
+- **Target:** `loss` (claim severity, USD)
+- **Loss:** Gamma deviance via `reg:gamma` (XGBoost) / `Tweedie:variance_power=1.99` (CatBoost) / explicit gamma NLL (DL)
+- **Cap:** 99.5th percentile (= ~$15,200; ~940 rows winsorised)
+- **Random seed:** 42
+
+For reference, the **Kaggle competition leaderboard top scores hit MAE
+~1126** using extensive tuning + cross-validation. This pipeline's
+**stacked ensemble at MAE 1,144 lands within 1.6% of that** using only
+default hyperparameters - which speaks to the strength of the pipeline's
+default settings rather than anything novel about the modelling.
+
+## How to use this collection
+
+### Step 1: Get the data from Kaggle
+
+The Allstate competition data is not redistributed in this repo.
+
+```bash
+# Set up Kaggle API auth: https://github.com/Kaggle/kaggle-api#api-credentials
+# Accept competition rules at:
+#   https://www.kaggle.com/c/allstate-claims-severity/rules
+# Then:
+pip install kaggle
+kaggle competitions download -c allstate-claims-severity
+unzip allstate-claims-severity.zip
+# Resulting train.csv is what these models were trained on
+```
+
+Or use the pipeline's downloader (handles the above):
+
+```bash
+git clone https://github.com/timothy22000/tabular_data_modelling_pipeline
+cd tabular_data_modelling_pipeline
+pip install -e ".[all]"
+python scripts/download_data.py --dataset allstate --kaggle
+# Saves to data/allstate.csv (188318 rows × 132 cols)
+```
+
+### Step 2: Load any of the pre-trained models
+
+CatBoost:
+```python
+from huggingface_hub import hf_hub_download
+from catboost import CatBoostRegressor
+import pandas as pd
+
+path = hf_hub_download("t22000t/allstate-tabular-models", "catboost.cbm")
+model = CatBoostRegressor()
+model.load_model(path)
+
+df = pd.read_csv("data/allstate.csv")  # downloaded in Step 1
+features = [f"cat{i}" for i in range(1, 117)] + [f"cont{i}" for i in range(1, 15)]
+preds = model.predict(df[features])
+```
+
+XGBoost:
+```python
+from huggingface_hub import hf_hub_download
+import xgboost as xgb
+
+path = hf_hub_download("t22000t/allstate-tabular-models", "xgboost.json")
+booster = xgb.Booster()
+booster.load_model(path)
+```
+
+XGBoost requires the exact preprocessing path used at training time. The
+easiest way to reproduce inference is to clone the pipeline repo and run
+the prediction script — see the pipeline README.
+
+### Step 3 (alternative): Re-run the full training
+
+```bash
+git clone https://github.com/timothy22000/tabular_data_modelling_pipeline
+cd tabular_data_modelling_pipeline
+pip install -e ".[all]"
+python scripts/download_data.py --dataset allstate --kaggle
+
+OMP_NUM_THREADS=1 python train.py \
+    --config configs/example_allstate.py \
+    --input data/allstate.csv \
+    --skip-tuning --skip-interpretability \
+    --architectures catboost xgboost cann cann_gbm ft_transformer tabm localglmnet drn
+```
+
+Expected wall-clock: 6-7 hours on Apple M-series CPU/MPS. Most of the
+time is FT-Transformer (4 hours, the worst-performing architecture);
+the GBMs + GLM-based DL models finish in under 90 min combined.
+
+(`OMP_NUM_THREADS=1` is only needed on macOS arm64 to avoid an OpenMP
+conflict; Linux runs are unaffected.)
+
+## Files
+
+| File | What it is | Size |
+|---|---|---|
+| `example_allstate.py` | DatasetConfig (target=`loss`, gamma family, all 130 features) |
+| `catboost.cbm` | Trained CatBoost (Tweedie:variance_power=1.99) | ~3 MB |
+| `xgboost.json` | Trained XGBoost Booster (`reg:gamma`) | 5 MB |
+| `cann_member{0,1,2}.pt` | CANN 3-seed ensemble | ~1.5 MB each |
+| `cann_gbm_member{0,1,2}.pt` | CANN-GBM 3-seed ensemble | ~1.5 MB each |
+| `ft_transformer_member{0,1,2}.pt` | FT-Transformer 3-seed ensemble (underfit - see results note) | ~3 MB each |
+| `tabm_member{0,1,2}.pt` | TabM 3-seed ensemble | ~7 MB each |
+| `localglmnet_member{0,1,2}.pt` | LocalGLMnet 3-seed ensemble | ~1 MB each |
+| `drn_member{0,1,2}.pt` | DRN 3-seed ensemble | ~1.5 MB each |
+| `evaluation_summary.csv` | Per-model train/test Gini, MAE, RMSE, A/E ratio, gamma deviance | 660 B |
+| `ensemble_weights.json` | NNLS weights over the 8 base predictions |
+| `dashboard_dl_models.html` | Interactive Plotly dashboard |
+| `figures/fig_dl_*.png` | Standalone publication figures |
+| `model_summary.json` | Structured run record (config, metrics, timing) |
+
+Total collection size: ~31 MB.
+
+## Training configuration
+
+| Setting | Value |
+|---|---|
+| Pipeline | [tabular-data-modelling-pipeline](https://github.com/timothy22000/tabular_data_modelling_pipeline) v0.1.0 |
+| Architectures | All 8 (catboost, xgboost, cann, cann_gbm, ft_transformer, tabm, localglmnet, drn) |
+| Hyperparameters | Defaults - **no Optuna tuning** |
+| DL ensemble size | 3 seeds per architecture |
+| Family / link | Gamma / log |
+| XGBoost objective | `reg:gamma` |
+| CatBoost loss | `Tweedie:variance_power=1.99` |
+| Train/test split | Random 80/20, seed 42 |
+| Cap percentile | 99.5 (=$15,200; ~940 winsorised) |
+| Hardware | Apple M-series, MPS device for DL |
+| Total wall-clock | 6h 47m (407 min) |
+
+## Limitations
+
+- **Default hyperparameters only.** No Optuna tuning. Kaggle leaderboard
+  winners used extensive tuning + bagging - expect ~0.02-0.03 Gini lift
+  and another ~$10-20 MAE reduction with tuning.
+- **FT-Transformer underfit** - documented above. Don't use those weights
+  directly; retrain with tuning if you need a transformer baseline.
+- **No interpretability artefacts** (Captum attributions, partial
+  dependence plots) - skipped to keep wall-clock under control. Run
+  without `--skip-interpretability` to compute them on a re-run.
+- **All 130 features anonymised.** No domain interpretability is
+  possible directly - `cat1`...`cat116` and `cont1`...`cont14` carry no
+  semantic meaning, so monotonicity constraints, base levels, and GLM
+  factor choices were left empty.
+- **Random split, not stratified.** `loss` is heavy-tailed; a quantile-
+  stratified split would give a more representative test set.
+- **Trained on competition `train.csv` only** (test.csv is unlabelled).
+  Not directly comparable to the official leaderboard.
+
+## Intended use
+
+- **Baseline for actuarial / claims-severity research** on the canonical
+  Kaggle dataset.
+- **Comparing your new tabular architecture** against eight strong baselines
+  on real insurance data.
+- **Teaching gamma-family regression** at meaningful scale (188k rows).
+- **Sanity check** for reimplementations of CatBoost/XGBoost/CANN-GBM/etc.
+
+## Citation
+
+```bibtex
+@software{tabular_data_modelling_pipeline,
+  author = {Mun, Timothy},
+  title  = {tabular-data-modelling-pipeline},
+  url    = {https://github.com/timothy22000/tabular_data_modelling_pipeline},
+  year   = {2026}
+}
+
+@misc{allstate2016,
+  author       = {Allstate Insurance Company},
+  title        = {Allstate Claims Severity},
+  year         = {2016},
+  url          = {https://www.kaggle.com/c/allstate-claims-severity},
+  note         = {Kaggle Competition}
+}
+```
+
+Please also cite the individual architecture papers - see the
+[main repo README](https://github.com/timothy22000/tabular_data_modelling_pipeline#the-eight-architectures).
+
+## License
+
+MIT for the model code and pipeline. The underlying Allstate dataset is
+distributed under Kaggle competition terms (**non-commercial use only**);
+this repository does **not** redistribute the raw data.
+
+## Related
+
+- 📦 [Pipeline: tabular-data-modelling-pipeline](https://github.com/timothy22000/tabular_data_modelling_pipeline)
+- 🤖 Companion model collections (full datasets included):
+  - [`t22000t/house-prices-tabular-models`](https://huggingface.co/t22000t/house-prices-tabular-models) - gamma, 1.5k rows
+  - [`t22000t/bike-sharing-tabular-models`](https://huggingface.co/t22000t/bike-sharing-tabular-models) - poisson, 17k rows

catboost.cbm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34d77a256f1059a0d2e71e3c8027e9ee979683df7371d2a9843e47ef75c7ec94
+size 4980160

ensemble_weights.json

@@ -0,0 +1,11 @@
+{
+  "glm": 0.0,
+  "catboost": 0.0,
+  "xgboost": 0.900443769806392,
+  "cann": 0.0,
+  "cann_gbm": 0.0,
+  "ft_transformer": 0.0,
+  "tabm": 0.099556230193608,
+  "localglmnet": 0.0,
+  "drn": 0.0
+}

evaluation_summary.csv

@@ -0,0 +1,10 @@
+model,gini_train,gini_test,mae,rmse,cv_rmse,ae_ratio,gamma_deviance,n_params,training_time
+cann_gbm,0.3497,0.3473,1158.43,1839.36,0.605,1.0099,-10437.003248,300519,881.6
+stacked_ensemble,0.3585,0.3472,1144.24,1863.54,0.613,1.0626,-10548.74513,9,0.0
+xgboost,0.3579,0.3468,1152.26,1849.69,0.6084,1.0268,-10490.842521,778,38.2
+catboost,0.3455,0.3461,1165.41,1855.64,0.6104,1.0151,-10545.968021,946,135.1
+cann,0.3521,0.3457,1168.04,1857.14,0.6109,1.0159,-10608.506085,300519,918.3
+drn,0.3503,0.3454,1174.7,1861.03,0.6122,1.0043,-10640.272218,300714,863.2
+localglmnet,0.3485,0.3428,1196.51,1903.44,0.6261,1.008,-10907.573099,193134,3502.3
+tabm,0.3565,0.3427,1392.63,2445.72,0.8045,1.5535,-18795.918859,1735956,3171.0
+ft_transformer,0.0268,0.0279,2141.76,3507.74,1.1538,3.0518,-95665.216816,700611,13875.2

example_allstate.py

@@ -0,0 +1,44 @@
+"""Allstate Claims Severity config.
+
+Kaggle competition:
+    https://www.kaggle.com/c/allstate-claims-severity
+
+NOTE: Kaggle competition rules permit non-commercial use. You must
+accept the competition terms before downloading. Set up Kaggle API
+auth (~/.kaggle/kaggle.json) then run:
+
+    python scripts/download_data.py --dataset allstate --kaggle
+
+Predicts ``loss`` (claim severity, USD) from 130 anonymised features
+(116 categorical, 14 continuous, plus ``id``). Gamma + log link.
+"""
+import os
+import sys
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from dataset_config import DatasetConfig
+
+
+# 116 categorical features named cat1..cat116
+_CATEGORICAL = [f"cat{i}" for i in range(1, 117)]
+# 14 continuous features named cont1..cont14
+_CONTINUOUS = [f"cont{i}" for i in range(1, 15)]
+
+
+config = DatasetConfig(
+    target_col="loss",
+    weight_col=None,
+    split_col=None,
+    exclude_cols=["id"],
+    continuous_features=_CONTINUOUS,
+    categorical_features=_CATEGORICAL,
+    derived_features={},
+    glm_factors=_CONTINUOUS[:4] + _CATEGORICAL[:4],
+    base_levels={},  # All features are anonymised - use mode levels at runtime
+    monotone_constraints={},  # No domain knowledge for anonymised features
+    family="gamma",
+    link="log",
+    prediction_floor=1.0,
+    cap_percentile=99.5,
+)

model_summary.json

@@ -0,0 +1,417 @@
+{
+  "pipeline": "dl",
+  "timestamp": "2026-06-29T00:17:06.382614",
+  "config": {
+    "seed": 42,
+    "n_tuning_trials": 30,
+    "cv_folds": 5,
+    "quick": false,
+    "architectures": [
+      "catboost",
+      "xgboost",
+      "cann",
+      "cann_gbm",
+      "ft_transformer",
+      "tabm",
+      "localglmnet",
+      "drn"
+    ],
+    "epochs": 300,
+    "patience": 30,
+    "batch_size": 512,
+    "n_ensemble": 3,
+    "catboost_iterations": 2000,
+    "mono_lambda": 0.1
+  },
+  "best_model": "cann_gbm",
+  "double_lift_vs_glm": 1.702,
+  "glm_gini": 0.20404812297155872,
+  "models": {
+    "catboost": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.345528,
+        "mae": 1131.0156,
+        "rmse": 1698.0539,
+        "cv_rmse": 0.564025,
+        "ae_ratio": 1.006675,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 2990.6396,
+        "gamma_deviance": -41071.576052
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.346107,
+        "mae": 1165.4098,
+        "rmse": 1855.6354,
+        "cv_rmse": 0.610391,
+        "ae_ratio": 1.015143,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 2994.727,
+        "gamma_deviance": -10545.968021
+      },
+      "training_time": 135.13505816459656,
+      "best_params": {},
+      "n_params": 946
+    },
+    "xgboost": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.357934,
+        "mae": 1051.5452,
+        "rmse": 1574.9415,
+        "cv_rmse": 0.523132,
+        "ae_ratio": 1.015477,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 2964.716,
+        "gamma_deviance": -36639.871748
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.346848,
+        "mae": 1152.2617,
+        "rmse": 1849.6937,
+        "cv_rmse": 0.608437,
+        "ae_ratio": 1.026758,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 2960.8489,
+        "gamma_deviance": -10490.842521
+      },
+      "training_time": 38.16731905937195,
+      "best_params": {
+        "max_depth": 6,
+        "eta": 0.05,
+        "n_estimators": 1000,
+        "reg_alpha": 0.1,
+        "reg_lambda": 1.0,
+        "subsample": 0.85,
+        "colsample_bytree": 0.85,
+        "early_stopping_rounds": 50
+      },
+      "n_params": 778
+    },
+    "cann": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.352146,
+        "mae": 1099.7003,
+        "rmse": 1648.2283,
+        "cv_rmse": 0.547475,
+        "ae_ratio": 1.007827,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 2987.2207,
+        "gamma_deviance": -38722.049985
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.345682,
+        "mae": 1168.0435,
+        "rmse": 1857.1383,
+        "cv_rmse": 0.610886,
+        "ae_ratio": 1.015883,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 2992.5446,
+        "gamma_deviance": -10608.506085
+      },
+      "training_time": 918.2992675304413,
+      "best_params": {
+        "hidden_dims": [
+          128,
+          64
+        ],
+        "dropout": 0.2,
+        "lr": 0.001,
+        "weight_decay": 0.001
+      },
+      "n_params": 300519,
+      "n_ensemble_members": 3,
+      "best_epochs": [
+        11,
+        10,
+        16
+      ]
+    },
+    "cann_gbm": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.349731,
+        "mae": 1110.4079,
+        "rmse": 1662.8043,
+        "cv_rmse": 0.552316,
+        "ae_ratio": 1.001403,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 3006.3831,
+        "gamma_deviance": -39527.447422
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.347317,
+        "mae": 1158.4278,
+        "rmse": 1839.3574,
+        "cv_rmse": 0.605037,
+        "ae_ratio": 1.009918,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 3010.2214,
+        "gamma_deviance": -10437.003248
+      },
+      "training_time": 881.5784821510315,
+      "best_params": {
+        "hidden_dims": [
+          128,
+          64
+        ],
+        "dropout": 0.2,
+        "lr": 0.001,
+        "weight_decay": 0.001
+      },
+      "n_params": 300519,
+      "n_ensemble_members": 3,
+      "best_epochs": [
+        11,
+        10,
+        8
+      ]
+    },
+    "ft_transformer": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.026796,
+        "mae": 2113.1906,
+        "rmse": 3373.5147,
+        "cv_rmse": 1.120545,
+        "ae_ratio": 3.022223,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 996.155,
+        "gamma_deviance": -374262.364084
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.027911,
+        "mae": 2141.7642,
+        "rmse": 3507.7363,
+        "cv_rmse": 1.153832,
+        "ae_ratio": 3.051806,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 996.156,
+        "gamma_deviance": -95665.216816
+      },
+      "training_time": 13875.232380867004,
+      "best_params": {
+        "d_model": 64,
+        "n_heads": 4,
+        "n_layers": 3,
+        "dropout": 0.1,
+        "ffn_factor": 4,
+        "lr": 0.0005,
+        "weight_decay": 0.0001
+      },
+      "n_params": 700611,
+      "n_ensemble_members": 3,
+      "best_epochs": [
+        0,
+        0,
+        206
+      ]
+    },
+    "tabm": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.35646,
+        "mae": 1331.63,
+        "rmse": 2243.6843,
+        "cv_rmse": 0.745261,
+        "ae_ratio": 1.541768,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 1952.6945,
+        "gamma_deviance": -64699.206702
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.342687,
+        "mae": 1392.6337,
+        "rmse": 2445.7232,
+        "cv_rmse": 0.804494,
+        "ae_ratio": 1.553486,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 1956.9371,
+        "gamma_deviance": -18795.918859
+      },
+      "training_time": 3170.9919884204865,
+      "best_params": {
+        "n_members": 8,
+        "hidden_dims": [
+          128,
+          64
+        ],
+        "dropout": 0.2,
+        "lr": 0.001,
+        "weight_decay": 0.0001
+      },
+      "n_params": 1735956,
+      "n_ensemble_members": 3,
+      "best_epochs": [
+        99,
+        112,
+        0
+      ]
+    },
+    "localglmnet": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.348541,
+        "mae": 1134.7861,
+        "rmse": 1701.8198,
+        "cv_rmse": 0.565276,
+        "ae_ratio": 1.000483,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 3009.1502,
+        "gamma_deviance": -40236.055751
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.342759,
+        "mae": 1196.5108,
+        "rmse": 1903.4373,
+        "cv_rmse": 0.626115,
+        "ae_ratio": 1.00803,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 3015.8595,
+        "gamma_deviance": -10907.573099
+      },
+      "training_time": 3502.3410143852234,
+      "best_params": {
+        "hidden_dims": [
+          64,
+          32
+        ],
+        "dropout": 0.3,
+        "coeff_reg": 1.0,
+        "lr": 0.0001,
+        "weight_decay": 0.01
+      },
+      "n_params": 193134,
+      "n_ensemble_members": 3,
+      "best_epochs": [
+        125,
+        138,
+        147
+      ]
+    },
+    "drn": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.350309,
+        "mae": 1110.3369,
+        "rmse": 1659.1151,
+        "cv_rmse": 0.551091,
+        "ae_ratio": 0.99661,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 3020.8442,
+        "gamma_deviance": -39493.391953
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.345411,
+        "mae": 1174.6962,
+        "rmse": 1861.0343,
+        "cv_rmse": 0.612167,
+        "ae_ratio": 1.004295,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 3027.0735,
+        "gamma_deviance": -10640.272218
+      },
+      "training_time": 863.1595370769501,
+      "best_params": {
+        "hidden_dims": [
+          128,
+          64
+        ],
+        "dropout": 0.2,
+        "kl_alpha": 0.1,
+        "lr": 0.001,
+        "weight_decay": 0.0001
+      },
+      "n_params": 300714,
+      "n_ensemble_members": 3,
+      "best_epochs": [
+        10,
+        15,
+        9
+      ]
+    },
+    "stacked_ensemble": {
+      "metrics_train": {
+        "split": "train",
+        "n": 150568,
+        "n_params": 0,
+        "gini": 0.358506,
+        "mae": 1045.0175,
+        "rmse": 1588.6652,
+        "cv_rmse": 0.52769,
+        "ae_ratio": 1.051201,
+        "mean_actual": 3010.6021,
+        "mean_predicted": 2863.963,
+        "gamma_deviance": -36577.102398
+      },
+      "metrics_test": {
+        "split": "test",
+        "n": 37750,
+        "n_params": 0,
+        "gini": 0.347197,
+        "mae": 1144.2383,
+        "rmse": 1863.5389,
+        "cv_rmse": 0.612991,
+        "ae_ratio": 1.062628,
+        "mean_actual": 3040.0753,
+        "mean_predicted": 2860.9032,
+        "gamma_deviance": -10548.74513
+      },
+      "training_time": 0.034310340881347656,
+      "best_params": {},
+      "n_params": 9,
+      "base_weights": {
+        "glm": 0.0,
+        "catboost": 0.0,
+        "xgboost": 0.900443769806392,
+        "cann": 0.0,
+        "cann_gbm": 0.0,
+        "ft_transformer": 0.0,
+        "tabm": 0.099556230193608,
+        "localglmnet": 0.0,
+        "drn": 0.0
+      }
+    }
+  }
+}