Bayaml

Bayaml is a lightweight ML & AI orchestration framework designed for structured, reproducible, and deployable machine learning workflows.

It features a deterministic execution engine and an intelligent Auto Mode (.auto()) that interprets structured instructions using rule-based NLP, without relying on external AI services. Your data remains completely local and secure.

Auto Mode automatically manages:

  • Data loading

  • Safe preprocessing

  • Task detection (classification or regression)

  • Structured pipeline construction

  • Model training

  • Evaluation

  • Deployment

Bayaml includes built-in AutoML capabilities such as model comparison, cross-validation, leaderboard tracking, and best-model selection — all executed within reproducible, hash-based execution plans.

It provides structured APIs across simple, fluent, and advanced orchestration layers, along with natural-language-driven ML pipelines and export options including REST deployment and ONNX.

Designed for engineers, researchers, startups, and ML teams.

Why Bayaml?

Modern ML pipelines are often:

  • Hard to reproduce

  • Poorly structured

  • Over-engineered

  • Difficult to deploy

Bayaml provides a clean orchestration layer on top of scikit-learn to make ML:

  • Reproducible

  • Structured

  • Deterministic

  • Deployable

Without sacrificing flexibility.

Natural Language ML (.auto())

Train models using structured instructions:

from bayaml import Project

p = Project()

result = p.auto(
    "use https://raw.githubusercontent.com/mwaskom/seaborn-data/master/tips.csv "
    "treat tip as target "
    "train regression model using linear regression"
)

print(result)

What Happens Automatically

  • Dataset loading (CSV, URL, DataFrame)

  • Target detection

  • Task detection (classification/regression)

  • Automatic categorical encoding

  • Safe pipeline ordering

  • Train-test split

  • Model training

  • Evaluation

  • Deterministic plan hashing

This makes Bayaml a lightweight AutoML orchestration engine.

Execution Plan Preview

Preview generated pipelines before execution:

plan = p.auto(
    "use data.csv treat price as target train regression model",
    preview=True
)

for step in plan.steps:
    print(step.name)

Bayaml builds a deterministic execution plan with hashing for reproducibility.

Core Capabilities

  • Structured ML Orchestration Layered architecture wrapping a deterministic execution engine.

  • Auto Encoding Categorical columns automatically encoded before training.

  • Auto Task Detection Classification vs regression inferred automatically.

  • AutoML Engine Model comparison, leaderboard tracking, cross-validation.

  • Model Deployment Export models for production use.

Deployment

REST Deployment

p.auto(
    "use data.csv treat target as target "
    "train classification model "
    "deploy as rest"
)

Generates a deployable REST bundle.

ONNX Export (Edge / C++ Ready)

p.auto(
    "use data.csv treat target as target "
    "train regression model "
    "deploy in c++"
)

Exports model as ONNX.

Installation

pip install bayaml

Dependencies:

  • pandas

  • numpy

  • scikit-learn

  • matplotlib

  • pyyaml

API Layers

Bayaml supports three levels of abstraction.

1. Simple API

from bayaml import quick_train

metrics = quick_train(
    data="data.csv",
    target="Target",
    model="linear_regression"
)

2. Fluent API

from bayaml import Bayaml

metrics = (
    Baya("data.csv", target="Target")
    .train("logistic_regression")
    .evaluate()
)

3. Advanced Orchestration API

from bayaml import Project

project = Project()
project.data.load("data.csv")
project.data.set_target("Target")
project.split.train_test()
project.model.create("linear_regression")
project.model.train()
project.evaluate.evaluate_regressor()

Full modular control.

AutoML

from bayaml import automl

result = automl(
    data="data.csv",
    target="Target"
)

print(result["best_model"])
print(result["best_score"])

Includes:

  • Cross-validation

  • Model comparison

  • Leaderboard generation

  • Best model selection

  • Run tracking

Export System

Export metrics, predictions, and reports:

  • CSV

  • JSON

  • Excel (XLSX)

  • PDF

  • DOCX

  • PNG / JPG

Output Modes

Bayaml supports multiple output rendering modes, allowing results to be displayed in different formats depending on workflow, integration needs, or presentation style.

These modes control how model results, metrics, and execution details are formatted — without affecting the underlying execution.

Available Output Modes

  1. pretty – Human-readable formatted output

  2. original – Raw native Bayaml execution result

  3. sklearn – scikit-learn compatible output format

  4. pandas – Pandas DataFrame structured output

  5. numpy – NumPy array structured output

  6. json – Machine-readable JSON output

  7. table – Clean tabular display format

  8. markdown – Markdown-rendered output

  9. latex – LaTeX formatted output

  10. diagnostic – Detailed execution diagnostics

Example Usage

result = p.auto(
    "use iris.csv treat species as target train classification model",
    mode="pretty"
)

result = p.auto(
    "use iris.csv treat species as target train classification model",
    mode="json"
)

Each mode adapts the output structure while keeping the execution deterministic and reproducible.

Why This Matters

These output modes allow Bayaml to integrate seamlessly with:

  • Research workflows (LaTeX, Markdown)

  • Production systems (JSON, NumPy)

  • Data pipelines (Pandas, sklearn)

  • Debugging and diagnostics

  • Human-readable reporting

This flexibility bridges experimentation, reporting, and deployment.

Architecture

Core Engine ↓ Execution Plan Builder ↓ Deterministic Plan Hashing ↓ Project API ↓ AutoML / CLI / Simple API

All APIs wrap the same deterministic engine.

No duplicated logic.

Reproducibility

Every .auto() run generates:

  • Dataset hash

  • Plan hash

  • Ordered execution steps

Ensuring reproducible ML workflows.

Developer Setup

git clone https://github.com/adityassarode/bayaml
cd bayaml
pip install -e .[dev]
pytest

Positioning

Bayaml is ideal for:

  • ML engineers building reproducible pipelines

  • Startups needing fast ML deployment

  • Data scientists wanting structured automation

  • Teams needing deterministic ML workflows

License

MIT License

Author

Bayaml is built and maintained by Aditya Sarode, focused on scalable AI systems, ML architecture, and production-ready engineering.

Documentation: