🔥 PipelineScript - Human-Readable ML Pipeline Language

Transform machine learning pipelines from code into conversation.

---

🚀 What is PipelineScript?

PipelineScript is a revolutionary Domain-Specific Language (DSL) that makes machine learning pipelines readable, debuggable, and accessible to everyone. No more nested code, complex APIs, or cryptic configurations.

Before PipelineScript:

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from xgboost import XGBClassifier
from sklearn.metrics import accuracy_score

# Load data
data = pd.read_csv('data.csv')

# Clean
data = data.dropna()

# Encode categoricals
from sklearn.preprocessing import LabelEncoder
for col in data.select_dtypes(['object']).columns:
    data[col] = LabelEncoder().fit_transform(data[col])

# Split
X = data.drop('target', axis=1)
y = data['target']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

# Scale
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

# Train
model = XGBClassifier()
model.fit(X_train, y_train)

# Evaluate
y_pred = model.predict(X_test)
print(f"Accuracy: {accuracy_score(y_test, y_pred)}")

# Export
import pickle
with open('model.pkl', 'wb') as f:
    pickle.dump(model, f)

With PipelineScript:

load data.csv
clean missing
encode
split 80/20 --target target
scale
train xgboost
evaluate
export model.pkl

That's it. Same functionality, 90% less code, infinitely more readable.

---

✨ Key Features

1. 🗣️ Human-Readable Syntax

Write ML pipelines like you'd describe them to a colleague:

load sales.csv
filter revenue > 1000
clean outliers
split 75/25 --target revenue
train xgboost
evaluate

2. 🐛 Interactive Debugging

Step through your pipeline like a regular program:

from pipelinescript import debug

debug("""
    load data.csv
    clean missing
    train xgboost
""")

Debugger commands:

• step - Execute next step
• break 3 - Set breakpoint at step 3
• context - Show current data and model
• inspect model - Inspect specific variable
• continue - Run until completion

3. 📊 Built-in Visualization

Automatically visualize your pipeline structure:

from pipelinescript import run

run(script, visualize=True)

Generates ASCII or graphical pipeline diagrams showing data flow.

4. 🔗 Method Chaining API

Prefer Python? Use the fluent API:

from pipelinescript import Pipeline

result = (Pipeline()
    .load("data.csv")
    .clean_missing()
    .encode()
    .split(0.8, target="label")
    .train("xgboost")
    .evaluate()
    .export("model.pkl")
    .run())

5. ⚡ Quick Builders

Pre-built pipelines for common tasks:

from pipelinescript.pipeline import quick_classification

# One line for complete classification pipeline
result = quick_classification("data.csv", "label", "xgboost")

---

📦 Installation

pip install pipelinescript

Optional dependencies:

# For XGBoost models
pip install xgboost

# For visualization
pip install matplotlib

# For all features
pip install pipelinescript[full]

---

🎯 Quick Start

1. Create a Pipeline File (.psl)

my_pipeline.psl:

load iris.csv
clean missing
encode
split 80/20 --target species
train random_forest
evaluate
export iris_model.pkl

2. Run It

Command Line:

pipelinescript run my_pipeline.psl

Python:

from pipelinescript import run

result = run("my_pipeline.psl")

if result.success:
    print(f"✅ Accuracy: {result.context.metrics['accuracy']:.4f}")

That's it! Your model is trained, evaluated, and exported.

---

📖 Language Reference

Commands

Data Loading

load <filepath>              # Load data from file

Supported formats: CSV, Excel, JSON, Parquet

Data Cleaning

clean missing                # Remove rows with missing values
clean duplicates             # Remove duplicate rows
clean outliers               # Remove statistical outliers (IQR method)

Data Transformation

encode                       # Encode categorical variables
scale                        # Scale numeric features (StandardScaler)
filter <condition>           # Filter rows (e.g., "age > 18")
select <col1> <col2> ...     # Select specific columns

Train/Test Split

split 80/20                  # Split data 80% train, 20% test
split 0.8 --target label     # Split with specific target column
split 75/25 --target price   # Custom ratio with target

Model Training

train xgboost                # XGBoost (requires xgboost package)
train random_forest          # Random Forest
train logistic               # Logistic Regression
train linear                 # Linear Regression
train auto                   # Auto-select based on task

Evaluation

predict                      # Make predictions on test set
evaluate                     # Compute evaluation metrics

Model Export/Import

export model.pkl             # Save model to file
save model.pkl               # Alias for export
import model.pkl             # Load model from file

Options

Options use --flag or -f syntax:

split 80/20 --target revenue
train xgboost --n_estimators 100

Comments

Use # for comments:

# Load and prepare data
load data.csv
clean missing  # Remove nulls

# Train model
train xgboost

---

🔥 Examples

Example 1: Basic Classification

load titanic.csv
clean missing
encode
split 80/20 --target survived
train random_forest
evaluate
export titanic_model.pkl

Example 2: Regression with Preprocessing

load housing.csv
clean outliers
select bedrooms bathrooms sqft price
scale
split 75/25 --target price
train linear
evaluate

Example 3: XGBoost with Feature Selection

load sales.csv
filter revenue > 1000
select date product revenue region
clean missing
encode
split 80/20 --target revenue
train xgboost
evaluate
export sales_model.pkl

Example 4: Interactive Debugging

from pipelinescript import debug

script = """
load data.csv
clean missing
split 80/20 --target label
train xgboost
evaluate
"""

result = debug(script)

# In debugger:
# (pdb) step           # Execute next step
# (pdb) context        # Show current state
# (pdb) inspect model  # Look at model
# (pdb) continue       # Run to completion

Example 5: Python API

from pipelinescript import Pipeline

# Method chaining
pipeline = (Pipeline()
    .load("data.csv")
    .clean_missing()
    .clean_outliers()
    .encode()
    .scale()
    .split(0.8, target="label")
    .train_xgboost()
    .evaluate()
    .export("model.pkl")
)

# Execute
result = pipeline.run()

# Show results
if result.success:
    print(f"Duration: {result.duration:.2f}s")
    print(f"Metrics: {result.context.metrics}")

Example 6: Quick Builders

from pipelinescript.pipeline import (
    quick_classification,
    quick_regression,
    quick_train
)

# Classification in one line
result = quick_classification("iris.csv", "species", "xgboost")

# Regression in one line
result = quick_regression("housing.csv", "price", "random_forest")

# Train and export in one line
result = quick_train("data.csv", "target", "model.pkl")

---

🎨 Visualization

ASCII Pipeline Diagram

from pipelinescript import run

run(script, visualize=True)

Output:

════════════════════════════════════════════════
    📊 PIPELINE VISUALIZATION
════════════════════════════════════════════════

    START
      │
      ▼
    ┌─────────────┐
    │ LOAD data.csv │
    └─────────────┘
      │
      ▼
    ┌──────────────┐
    │ CLEAN missing │
    └──────────────┘
      │
      ▼
    ┌──────────────┐
    │ TRAIN xgboost │
    └──────────────┘
      │
      ▼
    END

Graphical Pipeline (with matplotlib)

from pipelinescript import parse
from pipelinescript.visualizer import PipelineVisualizer

ast = parse(script)
visualizer = PipelineVisualizer()
visualizer.visualize_pipeline(ast, save_path="pipeline.png")

Generates a beautiful flowchart visualization.

---

🐛 Interactive Debugging

PipelineScript includes a powerful interactive debugger inspired by Python's pdb:

from pipelinescript import debug

debug("""
    load data.csv
    clean missing
    split 80/20 --target label
    train xgboost
    evaluate
""")

Debugger Commands

Command	Alias	Description
run	r	Run until completion/breakpoint
step	s, next, n	Execute next step
continue	c, cont	Continue execution
break <n>	b	Set breakpoint at step n
clear <n>		Clear breakpoint
list	l, ls	List all steps
context	ctx, vars	Show execution context
inspect <var>	i, p	Inspect variable
restart		Restart from beginning
quit	q, exit	Quit debugger

Example Debugging Session

(pdb) list
Pipeline Steps:
══════════════════════════════════════════════
   → 1. load
     2. clean
     3. split
     4. train
     5. evaluate
══════════════════════════════════════════════

(pdb) break 4
🔴 Breakpoint set at step 4

(pdb) run
▶️  Step 1: load
   Loaded 150 rows from iris.csv

▶️  Step 2: clean
   Removed 0 rows with missing values

▶️  Step 3: split
   Split data: 120 train, 30 test (80/20)

🔴 Breakpoint at step 4

(pdb) context
📊 Execution Context:
══════════════════════════════════════════════
  data: DataFrame (150, 5)
    columns: ['sepal_length', 'sepal_width', 'petal_length', 'petal_width', 'species']
  X_train: (120, 4)
  X_test: (30, 4)

  Recent log entries:
    • Loaded 150 rows from iris.csv
    • Removed 0 rows with missing values
    • Split data: 120 train, 30 test (80/20)
══════════════════════════════════════════════

(pdb) step
▶️  Step 4: train
   Trained XGBClassifier

(pdb) inspect model
model: XGBClassifier
  Value: XGBClassifier(...)

(pdb) continue
▶️  Step 5: evaluate
   Accuracy: 0.9667

✅ Pipeline execution completed!

---

🏗️ Architecture

PipelineScript consists of five core components:

┌─────────────────────────────────────────────┐
│          PipelineScript Engine              │
├─────────────────────────────────────────────┤
│                                             │
│  1. Parser     →  Lexical analysis & AST   │
│  2. Compiler   →  AST to executable steps  │
│  3. Executor   →  Step execution engine    │
│  4. Debugger   →  Interactive debugging    │
│  5. Visualizer →  Pipeline visualization   │
│                                             │
└─────────────────────────────────────────────┘

1. Parser (parser.py)

• Lexical analysis (tokenization)
• Syntax parsing
• AST generation

2. Compiler (compiler.py)

• Compiles AST into executable steps
• Integrates with sklearn, xgboost
• Handles data transformations

3. Executor (executor.py)

• Executes compiled steps
• Manages execution context
• Handles errors and logging

4. Debugger (debugger.py)

• Interactive step-through execution
• Breakpoints and inspection
• Context visualization

5. Visualizer (visualizer.py)

• ASCII pipeline diagrams
• Graphical visualizations
• DAG export

---

🎯 Use Cases

1. Rapid Prototyping

Test different models and preprocessing strategies in minutes:

load data.csv
clean missing
split 80/20 --target label
train xgboost
evaluate

2. Teaching & Learning

Perfect for teaching ML concepts without drowning in code:

# Clear, readable steps students can understand
load iris.csv
split 70/30 --target species
train random_forest
evaluate

3. Reproducible Research

Pipeline scripts are version-controllable and self-documenting:

# research_pipeline.psl
load experiment_data.csv
clean outliers
split 80/20 --target outcome
train xgboost
evaluate

4. Automated ML

Easily generate and test multiple pipelines programmatically:

models = ['xgboost', 'random_forest', 'logistic']

for model in models:
    pipeline = Pipeline().load("data.csv").clean_missing()
    pipeline.split(0.8, target="label").train(model).evaluate()
    result = pipeline.run()
    print(f"{model}: {result.context.metrics['accuracy']}")

5. Production Pipelines

Export trained pipelines as standalone Python scripts or containers.

---

🔬 Advanced Usage

Custom Preprocessing

from pipelinescript import Pipeline

pipeline = Pipeline()
pipeline.load("data.csv")

# Custom filtering
pipeline.filter("age > 18 and income < 100000")

# Select features
pipeline.select("age", "income", "education")

# Continue pipeline
pipeline.clean_missing().encode().scale()
pipeline.split(0.8, target="default").train("xgboost")

result = pipeline.run()

Accessing Context

result = pipeline.run()

if result.success:
    # Access data
    print(result.context.data.head())
    
    # Access model
    model = result.context.model
    
    # Access metrics
    print(result.context.metrics)
    
    # Access predictions
    predictions = result.context.predictions
    
    # Access log
    for entry in result.context.log:
        print(entry)

Extending PipelineScript

Add custom commands by extending the compiler:

from pipelinescript.compiler import PipelineCompiler
from pipelinescript.parser import ASTNode

class CustomCompiler(PipelineCompiler):
    def __init__(self):
        super().__init__()
        self.commands['my_command'] = self._compile_my_command
    
    def _compile_my_command(self, node: ASTNode):
        def custom_step(context):
            # Your custom logic
            return context
        
        return CompiledStep('my_command', custom_step, [], {}, node.line)

---

🚧 Roadmap

• v0.2.0: GPU support (RAPIDS, cuML)
• v0.3.0: Deep learning models (PyTorch, TensorFlow)
• v0.4.0: AutoML integration
• v0.5.0: Distributed training (Ray, Dask)
• v0.6.0: Model serving integration
• v0.7.0: Pipeline scheduling and monitoring
• v1.0.0: Production-ready feature complete

---

🤝 Contributing

Contributions welcome! Areas needing help:

1. Additional model types (SVM, KNN, etc.)
2. More preprocessing options
3. Better visualizations
4. Documentation improvements
5. Test coverage

See CONTRIBUTING.md for guidelines.

---

📄 License

MIT License - see LICENSE file.

---

🙏 Acknowledgments

PipelineScript was inspired by:

• SQL's declarative simplicity
• UNIX pipes' composability
• scikit-learn's consistent API
• The need for ML democratization

---

📊 Comparison

Feature	PipelineScript	Sklearn	Keras	MLflow
Human-readable syntax	✅	❌	❌	❌
Interactive debugging	✅	❌	❌	❌
Built-in visualization	✅	❌	✅	✅
One-line pipelines	✅	❌	❌	❌
No code required	✅	❌	❌	❌
Production ready	🚧	✅	✅	✅

---

🎓 Examples & Tutorials

See the examples/ directory for:

• simple_classification.psl - Basic classification
• xgboost_pipeline.psl - XGBoost example
• regression.psl - Regression pipeline
• python_examples.py - Python API examples
• iris.csv - Sample dataset

---

📞 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: idrissbadoolivier@gmail.com

---

🌟 Star History

If you find PipelineScript useful, please star the repo! ⭐

---