topprismdata/cultivating-ml-agent / skills/examples/kaggle-discussion-search
kaggle-discussion-search
VerifiedUse when looking for top-solution tricks on a Kaggle competition, when needing CV-LB gap discussions, when validating a strategy against community experience, or when seeking hyperparameter advice from competition winners. Triggers on phrases like "what did top scorers do", "CV-LB gap in this competition", "winning trick for X", or when starting any Kaggle competition.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 4.6 KB |
SKILL.md
View on GitHub---
name: kaggle-discussion-search
description: |
Use when looking for top-solution tricks on a Kaggle competition, when needing CV-LB gap discussions, when validating a strategy against community experience, or when seeking hyperparameter advice from competition winners. Triggers on phrases like "what did top scorers do", "CV-LB gap in this competition", "winning trick for X", or when starting any Kaggle competition.
---
# Kaggle Discussion Search (Community Knowledge)
## Context
Kaggle discussions contain **non-official, practitioner-tested wisdom** that often beats academic papers in practical value. Top solution write-ups, hyperparameter choices, CV-LB gap analyses β all are public. This skill teaches the agent to mine this gold mine at decision time, especially when starting a new competition or validating an existing skill against community evidence.
The core insight: **paper says "5-10% improvement", Kaggle discussion says "actually 2% on LB, watch out for X"**. Different signal, complementary.
## Guidance
### Basic Usage
```python
from framework.src.knowledge import KaggleForum
kf = KaggleForum()
# Get top-voted discussions for a specific competition
top_discs = kf.get_top_solutions("spaceship-titanic", top_n=20)
for d in top_discs:
print(f"[{d.votes}β {d.comments_count}π¬] {d.title}")
print(f" by {d.author}: {d.url}")
```
### Search by Keyword
```python
# Find discussions mentioning a specific technique
discs = kf.search_competition(
competition_slug="store-sales-time-series-forecasting",
query="CatBoost ensemble",
sort_by="votes"
)
```
### Global Search (Cross-Competition)
```python
# Search across all Kaggle discussions
discs = kf.search_global("CV-LB gap", max_results=20)
```
### Decision Workflow
```
Starting new Kaggle competition?
β
1. kf.get_top_solutions(competition_slug, top_n=30)
β Read the top-voted discussions
β Note: hyperparameter choices, CV strategies, common pitfalls
β
2. kf.search_competition(slug, query="LB score CV correlation")
β Find CV-LB gap analysis specific to this competition
β
3. Cross-reference with skills:
β cv-lb-gap-acknowledgment (general principle)
β time-series-walk-forward-validation (if time series)
β
4. After submission, search for related discussions:
β kf.search_competition(slug, query="your technique here")
β Validate your approach
```
### Combine with Papers
```python
from framework.src.knowledge import KnowledgeAggregator
agg = KnowledgeAggregator()
# One-shot: papers + Kaggle + vault combined
report = agg.search_all("feature engineering saturation", max_per_source=10)
# report.scholar_papers, report.arxiv_papers, report.kaggle_discs
```
## Why This Matters
| Source | Strength | Weakness |
|---|---|---|
| arxiv paper | Novel, rigorous, citable | Often idealistic, not battle-tested |
| **Kaggle discussion** | **Real-world validated, hyperparameter-specific** | **Ancedotal, scattered** |
| Skills (static) | Curated, distilled | Cutoff knowledge |
Real example: The `feature-engineering-saturation-detection` skill was created after seeing **multiple Kaggle discussions** warning about feature engineering saturation in tabular competitions β combining community signal with formal theory.
## When to Apply
### When to Use
- Starting any new Kaggle competition (read top solutions first)
- Looking for hyperparameter advice from competition winners
- Validating whether your CV score will translate to LB
- Finding "non-obvious" tricks (post-processing, ensemble weights, etc.)
- Comparing your approach to community consensus
### When NOT to Use
- Pure research / academic projects (no Kaggle context)
- When network unavailable
- For deep technical theory (use arxiv instead)
- Privacy-sensitive contexts
## Notes
- **Sort by votes** first β top-voted discussions are usually most informative
- **Read comments, not just post** β often the real insight is in comments
- **Cross-check across competitions** β same author posting on multiple competitions is a strong signal
- **Combine with paper search** β best insights usually combine theory (arxiv) + practice (Kaggle)
- **Cache results** in vault (importance=0.7+) for future reference
- See also: `arxiv-paper-search`, `cv-lb-gap-acknowledgment`, `memory-hierarchy-management`
## References
- Implementation: `framework/src/knowledge/kaggle_forum.py`
- Aggregator: `framework/src/knowledge/aggregator.py`
- Kaggle discussions: https://www.kaggle.com/discussions
- Real skills enabled by Kaggle mining: `feature-engineering-saturation-detection`, `cv-lb-gap-acknowledgment`, `ml-sweet-spot`
- Related skill: `arxiv-paper-search` (complementary data source)Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |