---
license: mit
size_categories:
- 1M<n<10M
task_categories:
- tabular-regression
tags:
- causal-discovery
- markov-boundary
- causal-inference
dataset_info:
  features:
  - name: task_id
    dtype: string
  - name: X
    list:
      list: float32
  - name: y
    list: float32
  - name: n_features
    dtype: int32
  - name: mb_mask
    list: bool
  - name: mb_ratio
    dtype: float32
  - name: density
    dtype: float64
  - name: dag
    struct:
    - name: dag_id
      dtype: string
    - name: num_nodes
      dtype: int32
    - name: edge_list
      list:
        list: int32
  - name: scm
    struct:
    - name: scm_type
      dtype: string
    - name: noise_model
      dtype: string
    - name: pnl
      dtype: bool
    - name: coeff_range
      dtype: float32
    - name: noise_std
      dtype: float32
  - name: target
    struct:
    - name: target_node
      dtype: int32
    - name: parent_mask
      list: bool
    - name: child_mask
      list: bool
    - name: spouse_mask
      list: bool
  - name: meta
    struct:
    - name: n_samples
      dtype: int32
    - name: seed
      dtype: int64
  splits:
  - name: f100
    num_bytes: 127915188
    num_examples: 300
  - name: f200
    num_bytes: 366215022
    num_examples: 450
  - name: f400
    num_bytes: 733959702
    num_examples: 450
  - name: f600
    num_bytes: 1106625222
    num_examples: 450
  - name: f800
    num_bytes: 1484273862
    num_examples: 450
  - name: f1000
    num_bytes: 1866885102
    num_examples: 450
  - name: f40
    num_bytes: 51338868
    num_examples: 300
  - name: f60
    num_bytes: 76379748
    num_examples: 300
  - name: f80
    num_bytes: 101951628
    num_examples: 300
  download_size: 5804489137
  dataset_size: 5915544342
configs:
- config_name: default
  data_files:
  - split: f100
    path: data/f100-*
  - split: f200
    path: data/f200-*
  - split: f400
    path: data/f400-*
  - split: f600
    path: data/f600-*
  - split: f800
    path: data/f800-*
  - split: f1000
    path: data/f1000-*
  - split: f40
    path: data/f40-*
  - split: f60
    path: data/f60-*
  - split: f80
    path: data/f80-*
---

## SCM3K

Benchmark dataset for the paper:

> **[The Good, the Bad, and the Ugly of Markov Boundary for Tabular Prediction](https://huggingface.co/papers/2605.29411)**
> [Shu Wan](https://huggingface.co/Shuwan), Abhinav Gorantla, Huan Liu, K. Selçuk Candan

3,450 tabular prediction tasks sampled from random structural causal
models (SCMs), totalling 3.45M records (1,000 samples per task).
Each task ships with the ground-truth Markov boundary of the target
node, so you can evaluate feature selection and prediction under
known causal structure. Nine feature-count levels from 40 to 1,000.

### Splits

One HF split per feature count `F`. No predefined train/test
partition — use HF slice syntax (e.g. `split="f200[:80%]"`).

| Split | `F`  | `num_nodes` | DAG density            | MB-ratio band     | Tasks |
|-------|------|-------------|------------------------|-------------------|-------|
| f40   | 40   | 41          | ER `[0.2, 0.4]`         | `[0.10, 0.90]`    | 300   |
| f60   | 60   | 61          | ER `[0.2, 0.4]`         | `[0.10, 0.90]`    | 300   |
| f80   | 80   | 81          | ER `[0.2, 0.4]`         | `[0.10, 0.90]`    | 300   |
| f100  | 100  | 101         | ER `[0.2, 0.4]`         | `[0.10, 0.90]`    | 300   |
| f200  | 200  | 201         | ER `[0.01, 0.02, 0.04]` | `[0.05, 0.95]`    | 450   |
| f400  | 400  | 401         | ER `[0.01, 0.02, 0.04]` | `[0.05, 0.95]`    | 450   |
| f600  | 600  | 601         | ER `[0.01, 0.02, 0.04]` | `[0.05, 0.95]`    | 450   |
| f800  | 800  | 801         | ER `[0.01, 0.02, 0.04]` | `[0.05, 0.95]`    | 450   |
| f1000 | 1000 | 1001        | ER `[0.01, 0.02, 0.04]` | `[0.05, 0.95]`    | 450   |
| | | | | **Total** | **3,450** |

### Row schema

Each row is one prediction task.

| Field | Type | What it stores |
|-------|------|----------------|
| `task_id` | string | unique task identifier |
| `X` | list\<list\<f32\>\> | feature matrix, 1,000 x F |
| `y` | list\<f32\> | target vector, length 1,000 |
| `n_features` | int | number of features (= F) |
| `mb_mask` | list\<bool\> | true Markov boundary mask over features |
| `mb_ratio` | float | fraction of features in the Markov boundary |
| `density` | float | edge density of the generating DAG |
| `dag` | struct | `dag_id`, `num_nodes`, `edge_list` |
| `scm` | struct | `scm_type`, `noise_model`, `pnl`, `coeff_range`, `noise_std` |
| `target` | struct | `target_node`, `parent_mask`, `child_mask`, `spouse_mask` |
| `meta` | struct | `n_samples`, `seed` |

### How the data was generated

DAGs: Erdos-Renyi, 5 graphs per (num_nodes, density) pair, seed 42.

SCMs: six families — `LINEAR_GAUSSIAN`, `LINEAR_NONGAUSSIAN`,
`NL_ANM_GAUSSIAN`, `NL_ANM_NONGAUSSIAN`, `PNL`, `HETEROSKEDASTIC`.
Each DAG gets 5 SCM instantiations with `n_samples=1000`,
`coeff_range=1.0`, `noise_std=0.5`.

### Quick start

```python
from datasets import load_dataset

ds = load_dataset("CSE472-blanket-challenge/SCM3K", split="f200")
task = ds[0]

X  = task["X"]          # 1000 x 200
y  = task["y"]          # 1000
mb = task["mb_mask"]    # ground-truth Markov boundary
```

### Citation

```bibtex
@article{wan2026gbu,
  title  = {The Good, the Bad, and the Ugly of Markov Boundary
            for Tabular Prediction},
  author = {Wan, Shu and Gorantla, Abhinav and Liu, Huan
            and Candan, K. Sel{\c{c}}uk},
  year   = {2026},
}
```