File size: 2,559 Bytes
0fcb1cf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
---
license: cc0-1.0
language:
  - en
tags:
  - competition
  - baseline
  - machine-learning
  - kaggle
task_categories:
  - other
pretty_name: Competition Baseline Kit
size_categories: n<1K
---

# Competition Baseline Kit

**CPU-only baselines for entering Hugging Face competitions in minutes — free to run.**

One command gets you a working submission on most HF competitions (generic CSV-submission
type). Replace the model with something better as the deadline allows; the plumbing stays.

## How HF competitions work (two types)

1. **Generic** — you get a training CSV + a test CSV, you upload a `predictions.csv`.
   All test data is public. Baselines below cover this.
2. **Script** — you submit code that generates predictions; test data can be hidden.
   The kit's scripts adapt (write your training inside a `train()` and save the model).

## Quick start

```bash
# text classification/regression (e.g. movie-genre-prediction style)
python baseline_text.py --train train.csv --test test.csv --target label --text-col text

# tabular (e.g. wyze-rule-recommendation style)
python baseline_tabular.py --train train.csv --test test.csv --target label
```

Output: `submissions/predictions.csv` — exactly what the competition UI asks you to upload.

## Kit contents

| File | Use |
|---|---|
| `baseline_text.py` | TF-IDF (1-2 grams) + LogisticRegression — strong text baseline |
| `baseline_tabular.py` | HistGradientBoosting (classifier/regressor auto-detect) — handles NaN + categories |

## The winning workflow (this is the actual edge)

1. **Monitor daily** — new competitions get announced on the [HF forum (Community Calls)](https://discuss.huggingface.co/c/community-calls) and the [competitions org](https://huggingface.co/competitions). Being first means more submissions and more learning.
2. **Enter within 24h** of launch with a baseline — you now have a leaderboard position and can iterate publicly.
3. **Iterate free**: feature engineering, ensembling, and (if the task needs it) small fine-tunes on CPU/Colab-free resources.
4. **Publish your solution** on the Hub after the deadline — it becomes portfolio material (models, datasets, write-ups) that feeds the consulting side of the profile.

## Known active/live reference

- KDD Cup 2026 — Tencent UNI-REC ($885k pool) and HKUST Data Agents ($30k pool) announced March 2026: https://dataagent.top/
- ICML 2026 agent reproduction challenge (closed Aug 2, 2026) — pattern to watch for 2027

## License

CC0 — steal the baselines, they're meant to be forked.