---

🚀 What’s New in v2.0

ezyml 2.0 is a major architectural upgrade. It is no longer just a trainer — it is a machine‑learning compiler.

🆕 Major Additions

• 🧠 ezyml compile – one command to generate models, metrics, APIs, demos, and infra
• 🧩 Pipeline‑Driven Execution – YAML‑based pipelines with visual DAGs
• 🎛 User‑Controlled Artifacts – generate only what you ask for
• 📊 Auto‑EDA + Evaluator – dataset profiling, metrics, plots
• 🧪 Production‑Ready Demos – high‑quality Streamlit UI generation
• 📦 Deployment Tooling – FastAPI, Docker, Kubernetes YAML
• 🔍 Dataset Fingerprinting – reproducibility by design

---

🌟 Why ezyml?

ezyml removes boilerplate across the entire ML lifecycle:

dataset → training → evaluation → deployment → demo

All without forcing you into a framework lock‑in.

Core Philosophy

• Explicit over magic – nothing is generated unless you ask
• Beginner‑friendly, expert‑capable
• Composable, inspectable, debuggable

---

📦 Installation

pip install ezyml==2.0.0

---

🚀 CLI Quickstart

Below are the most common ways users interact with ezyml — via the CLI or Python API.

---

🧠 Train a Model (CLI – v1 compatible)

ezyml train \
  --data data.csv \
  --target label \
  --model random_forest

This trains a model and prints evaluation metrics.

---

🧩 Compile an End-to-End ML System (CLI – v2.0)

Minimal (no pipeline, no extras):

ezyml compile \
  --data heart.csv \
  --target target

Generates:

build/
├── model.pkl
└── metrics.json

---

🎛 Compile With Explicit Outputs

ezyml compile \
  --data heart.csv \
  --target target \
  --api \
  --demo \
  --docker \
  --k8s

Each flag enables a specific artifact:

• --api → FastAPI inference app
• --demo → Interactive Streamlit demo
• --docker → Dockerfile
• --k8s → Kubernetes manifests

---

🧩 Compile Using a YAML Pipeline (Advanced)

ezyml compile \
  --pipeline pipeline.yaml \
  --data heart.csv \
  --target target \
  --all

---

🧩 Compile a Full ML System (v2.0)

ezyml compile \
  --pipeline pipeline.yaml \
  --data data.csv \
  --target label

Default output (minimal):

build/
├── model.pkl
└── metrics.json

---

🎛 User‑Controlled Outputs

ezyml compile \
  --pipeline pipeline.yaml \
  --data data.csv \
  --target label \
  --api \
  --demo \
  --docker \
  --k8s

---

🧪 Pipeline Example (YAML)

steps:
  trainer:
    type: EZTrainer
    params:
      model: random_forest
      target: label

---

🧠 Python API (Still Supported)

You can use ezyml programmatically without the CLI.

Basic Training

from ezyml.core import EZTrainer

trainer = EZTrainer(
    data="heart.csv",
    target="target",
    model="random_forest"
)

trainer.train()
trainer.save_model("model.pkl")
trainer.save_report("metrics.json")

---

Predictions in Python

import pandas as pd
from ezyml.core import EZTrainer

trainer = EZTrainer(data="heart.csv", target="target")
trainer.train()

X_new = pd.read_csv("new_samples.csv")
preds = trainer.predict(X_new)
print(preds)

---

Using ezyml as a Library Component

from ezyml.compiler.compile import compile_project
from ezyml.core import EZTrainer

trainer = EZTrainer(data="heart.csv", target="target")
trainer.train()

compile_project(
    trainer=trainer,
    schema={"age": "number", "chol": "number"},
    api=True,
    demo=True
)

---

📊 Evaluation & Analytics

• Accuracy, F1, ROC‑AUC (classification)
• MAE, RMSE, R² (regression)
• Confusion matrix, ROC & PR curves
• Drift‑ready metric storage

---

📦 Deployment Targets

Layer	Supported
API	FastAPI
Demo	Streamlit
Container	Docker
Orchestration	Kubernetes

---

🧰 Supported Models

Task	Models
Classification	logistic_regression, random_forest, xgboost, svm, naive_bayes, gradient_boosting, extra_trees, knn
Regression	linear_regression, ridge, lasso, elasticnet, random_forest, xgboost, svr, gradient_boosting
Clustering	kmeans, dbscan, agglo
Dim Reduction	pca, tsne

---

🔮 Roadmap

• Learner Mode (explain decisions)
• SHAP‑based explainability
• Model comparison dashboards
• Presets (--preset production)
• CI/CD & MLOps integrations

---

📜 License

MIT License – View License

---

👨‍💻 Author

Built with ❤️ by Raktim Kalita GitHub: https://github.com/Rktim