Skip to content

🧑‍🍳 Recipes

FlowyML Notebook includes a powerful Recipe System — pre-configured, reusable code templates for common ML tasks. Think "snippets on steroids," with usage tracking, team sharing, and automatic surfacing.

Recipe Library

Recipes Panel

83 built-in recipes organized by 15 categories, with search, sort, and one-click insertion

Built-in Categories

Category Count Examples
⭐ Core 14 FlowyML Step, Pipeline, Conditional Branching, Scheduling
📦 Assets 11 Dataset loading, Model versioning, Experiment tracking
⏩ Parallel 5 Parallel map, Retry logic, Checkpointing, Smart caching
👁 Observability 7 GenAI tracing, OpenAI/LangChain observability, Drift detection
📈 Evals 6 Eval suites, Judge arena, Structured assessment
🛡 Production 6 Config management, FastAPI serving, Docker, CI/CD, Monitoring, A/B Testing
🧠 Deep Learning 5 PyTorch training, Transfer learning, Optuna tuning, SHAP explainability
🌐 NLP 5 Text classification, Sentiment analysis, Topic modeling, NER, Embeddings
🕐 Time Series 5 Seasonal decomposition, ARIMA, Prophet, Anomaly detection, Rolling features
🏗 Data Engineering 5 ETL pipeline, Data quality, DuckDB analytics, SQLAlchemy, Schema evolution
💾 Data Prep 7 Imputation, Feature selection, Outlier treatment, SMOTE
🤖 ML / Models 3 XGBoost/LightGBM baselines, Feature engineering
📊 Visualization 4 Plotly charts, SHAP/LIME, Dashboard templates
🦄 Ecosystem 4 KDP preprocessing, KerasFactory model, MLPotion training, E2E pipeline

Using a Recipe

  1. Open the Snips panel in the sidebar
  2. Browse or search for a recipe (search filters by name, category, and tags)
  3. Click + to insert into your notebook at the cursor position
  4. Customize the template for your use case

Most Used First

Recipes track usage counts automatically. Your most-used recipes surface first in the Popular tab — no manual favorites needed.

Creating Custom Recipes

Save any cell as a reusable recipe:

  1. Write your code in a cell
  2. Click the "Save as Recipe" icon in the cell toolbar
  3. Provide a Name, Category, Tags, and Description

Custom recipes are stored locally as .recipe.json files:

1
2
3
4
~/.flowyml/recipes/
├── custom-1711234567890.recipe.json
├── custom-1711234567891.recipe.json
└── usage.json                        # Usage tracking data

Recipe Format

Each recipe is a self-contained JSON file:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
{
  "id": "custom-1711234567890",
  "name": "XGBoost Quick Baseline",
  "category": "ML",
  "description": "Train an XGBoost classifier with cross-validation",
  "tags": ["xgboost", "classification", "baseline"],
  "source": "import xgboost as xgb\nfrom sklearn.model_selection import cross_val_score\n\nmodel = xgb.XGBClassifier(n_estimators=100)\nscores = cross_val_score(model, X, y, cv=5, scoring='f1')\nprint(f'F1: {scores.mean():.4f} ± {scores.std():.4f}')",
  "builtin": false,
  "created_at": "2026-03-23T15:00:00",
  "updated_at": "2026-03-23T15:00:00"
}

Managing Recipes

Via Python API

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
from flowyml_notebook.recipes_store import RecipeStore

store = RecipeStore()

# List all custom recipes (with usage counts)
recipes = store.list_recipes()

# Save a new recipe
store.save_recipe({
    "name": "My Custom Pipeline",
    "category": "Core",
    "tags": ["pipeline", "custom"],
    "source": "from flowyml import pipeline\n\n@pipeline\ndef my_pipeline():\n    ...",
})

# Track usage (called automatically when inserted)
store.track_usage("custom-1711234567890")

# Delete a recipe
store.delete_recipe("custom-1711234567890")

Import & Export

Move recipes between machines or share outside of GitHub:

1
2
3
4
5
# Export all recipes as JSON
all_recipes = store.export_all()

# Import recipes from another source
store.import_recipes(imported_list, overwrite=False)

Sharing Recipes via GitHub

When connected to a GitHub repository (see Collaboration), recipes can be pushed to the shared team hub:

Method How
Push to Hub Toggle "Share to Hub" when saving → recipe is committed to .flowyml-hub/recipes/
Pull from Hub Click Pull in Git panel → team recipes appear in the Shared tab
Manual Export Export as JSON and share directly

Shared Recipe Catalog

The hub maintains a catalog.json index for fast browsing:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
{
  "recipes": [
    {
      "id": "xgboost-baseline",
      "name": "XGBoost Quick Baseline",
      "category": "ML",
      "description": "Train an XGBoost classifier with CV",
      "tags": ["xgboost", "baseline"],
      "shared_by": "alice",
      "shared_at": "2026-03-23T15:00:00"
    }
  ],
  "updated_at": "2026-03-23T15:00:00"
}

When a teammate pushes a recipe, it appears in the Shared tab for everyone who pulls.

Best Practices

Recipe Tips

  • Name clearly — XGBoost Quick Baseline > my_model
  • Tag generously — Tags power search: ["xgboost", "classification", "baseline"]
  • Keep focused — One recipe = one task. Don't mix data loading with model training
  • Document parameters — Use comments to explain what to customize
  • Share to Hub — If it helped you, it'll help your team

🦄 UnicoLab Ecosystem Recipes

FlowyML Notebook includes 4 builtin ecosystem recipes that provide ready-to-use workflows for the UnicoLab ML stack:

Recipe Package What It Does
🧪 KDP Smart Preprocessing kdp Auto-configure Keras preprocessing layers — feature types, distribution-aware encoding, tabular attention
🏗 KerasFactory Quick Model kerasfactory Build tabular models with BaseFeedForwardModel or custom GRN/attention architectures
⚗ MLPotion Training Pipeline mlpotion Managed training with ModelTrainer and type-safe ModelTrainingConfig
🦄 UnicoLab End-to-End Pipeline All 3 Complete KDP → KerasFactory → MLPotion workflow in one recipe

Ecosystem recipes appear alongside your custom and shared recipes in the Snips panel. They are tagged with unicolab for easy search.

Installing Ecosystem Packages

1
pip install "flowyml-notebook[keras]"

See Ecosystem for the full integration guide.