Training ML Models with Scikit-learn & TensorFlow – Build & Save Your Models Like a Pro

Scikit‑learn Overview

Open‑source Python library with consistent APIs for classification, regression, clustering, model selection, preprocessing, and more. Ideal for prototyping and small‑to‑medium projects.

Core Concepts

• Estimators (fit), Predictors (predict), Transformers (transform)
• Pipelines, model selection, cross‑validation, metrics
• Model persistence using joblib or pickle

Project: Student Pass/Fail Prediction (Scikit‑learn)

Build a pipeline to predict student outcomes and save the trained model.

# requirements
pip install pandas scikit-learn joblib fastapi uvicorn

# data_generation.py
import pandas as pd
import random

random.seed(42)
schools = ['Greenwood High', 'Sunrise Public', 'Hillview School']
data = []
for _ in range(100):
    school = random.choice(schools)
    study_hours = round(random.uniform(1, 10), 1)
    absences = random.randint(0, 10)
    grade = round(random.uniform(40, 100), 1)
    passed = 1 if grade >= 50 else 0
    data.append([school, study_hours, absences, grade, passed])

df = pd.DataFrame(data, columns=['school', 'study_hours', 'absences', 'grade', 'passed'])
df.to_csv("students.csv", index=False)

# train_model.py
import pandas as pd
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
import joblib

df = pd.read_csv("students.csv")
X = df.drop("passed", axis=1)
y = df["passed"]

preprocessor = ColumnTransformer([
    ("school_encoder", OneHotEncoder(), ["school"])
], remainder="passthrough")

pipeline = Pipeline([
    ("preprocess", preprocessor),
    ("classifier", RandomForestClassifier(random_state=42))
])

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
pipeline.fit(X_train, y_train)

joblib.dump(pipeline, "student_model.pkl")

# predict.py
import joblib
import pandas as pd

model = joblib.load("student_model.pkl")

sample = pd.DataFrame([{
    "school": "Greenwood High",
    "study_hours": 6.5,
    "absences": 2,
    "grade": 78.0
}])

pred = model.predict(sample)
print(f"Predicted: {'Pass' if pred[0] == 1 else 'Fail'}")

# serve_model.py
from fastapi import FastAPI
from pydantic import BaseModel
import pandas as pd
import joblib

app = FastAPI()
model = joblib.load("student_model.pkl")

class StudentData(BaseModel):
    school: str
    study_hours: float
    absences: int
    grade: float

@app.post("/predict")
def predict(data: StudentData):
    df = pd.DataFrame([data.dict()])
    prediction = model.predict(df)[0]
    return {"prediction": "Pass" if prediction == 1 else "Fail"}

# run
# uvicorn serve_model:app --reload

TensorFlow Overview

Open‑source framework for ML/DL from Google. Supports CPUs/GPUs/TPUs, Keras API, TensorBoard, and production serving.

• Tensors, Keras layers, optimizers, losses, metrics
• tf.function, SavedModel format, TensorFlow Serving

Modular Training Pipelines

# pipeline.py
from data_loader import load_data
from preprocessing import preprocess_data
from model import build_model
from train import train_model
from evaluate import evaluate_model
from save_load import save_model

raw_data = load_data("data/data.csv")
X_train, X_test, y_train, y_test = preprocess_data(raw_data)
model = build_model()
trained_model = train_model(model, X_train, y_train)
evaluate_model(trained_model, X_test, y_test)
save_model(trained_model, "models/model.pkl")

• Maintainable, testable, reusable components.
• Scales to multiple datasets/models.

Challenges

• Feature engineer a CSV with Pandas.
• Train and save a Scikit‑learn model on Iris or Titanic.
• Train a TensorFlow model on MNIST or Fashion‑MNIST and save.
• Evaluate and log metrics (accuracy, loss) to JSON/text.
• Modularize training into data.py, train.py, evaluate.py, save.py.
• Load saved models and run inference on new samples.
• Document the workflow in README.md.