prepstack 0.1.0

Modular data analysis and ML workflow toolkit

prepstack

prepstack is a modular data analysis and machine learning workflow toolkit designed to reduce repetitive preprocessing, validation, and experimentation overhead while preserving full transparency and user control.

It provides explicit, composable building blocks for common prepstackl and ML tasks, with optional guidance for interpretability and correctness.

#Design Principles

⦁ Explicit over implicit No hidden transformations or silent assumptions.

⦁ Modular by default Each step is independently callable and testable.

⦁ Guardrails, not automation prepstack highlights risks and inconsistencies without enforcing opinionated pipelines.

⦁ Learning-aware execution Optional guidance explains what is happening and why, without changing results.

⦁ Notebook and production friendly Identical APIs for exploration and deployment.

---

#Installation

pip install prepstack

---

#Guidance Mode

All functions support a guidance parameter:

guidance="on" # descriptive output, warnings, suggestions guidance="off" # silent execution

This allows prepstack to be used both as a learning aid and as a production utility.

---

#Module Overview

⦁ prepstack.cleaning - Basic data hygiene and structural cleanup

⦁ prepstack.transform - Feature transformation, encoding, scaling, joins, and aggregation

⦁ prepstack.explore - Structured exploratory analysis

⦁ prepstack.model_prep - Validation, feature selection, splitting, and class balancing

⦁ prepstack.model - Model fitting and prediction wrappers

---

#Example Workflow

import pandas as pd

df = pd.DataFrame({ "user_id": range(1, 51), "age": [22,25,30,35,40]*10, "country": ["US","IN","US","UK","IN"]*10, "usage": [5,10,20,30,40]*10, "plan": ["free","free","pro","pro","free"]*10, "churn": [0,0,1,1,0]*10 })

from prepstack.cleaning import clean_basic from prepstack.transform.columns import encoding, scaling, string_ops from prepstack.model_prep.feature_selection import combined_feature_selection from prepstack.model_prep.splits import stratified_split from prepstack.model_prep.class_balance import smote_balance from prepstack.model_prep.validation.core import validate_full from prepstack.model import fit_model, predict from prepstack.model_prep.evaluation import classification_metrics, evaluation_summary

1️⃣ CLEAN

df1 = clean_basic(df, guidance="on")

2️⃣ TRANSFORM

df2 = string_ops.clean_strings(df1, columns=["country", "plan"], guidance="on") df3 = df3, encoding_info = encoding.encode_category( df2, columns=["country", "plan"], method="onehot", guidance="on")

df4 = scaling.scale_numeric( df3, cols=["age", "usage"], method="standard", guidance="on")

3️⃣ VALIDATE

validate_full( df4, schema={"age": "float64", "usage": "float64", "churn": "int64"}, guidance="on" )

4️⃣ FEATURE SELECTION

selected_features = combined_feature_selection(df4, target="churn", guidance="on")

X = df4[selected_features] y = df4["churn"]

5️⃣ SPLIT

X_train, X_test, y_train, y_test = stratified_split( df4[selected_features + ["churn"]], target="churn", guidance="on" )

6️⃣ BALANCE

Xb, yb = smote_balance(X_train, y_train, guidance="on")

7️⃣ MODEL

model = fit_model(Xb, yb, model_type="logistic", guidance="on")

8️⃣ EVALUATE

preds = predict(model, X_test)

metrics = classification_metrics(y_test, preds, guidance="on")

evaluation_summary(metrics, task="classification", guidance="on")

---

#Intended Use Cases

⦁ Reusable analytics pipelines

⦁ SaaS / churn / behavioural datasets

⦁ ML experimentation with safety checks

⦁ Analysts transitioning into modeling

⦁ Teams that want consistency without abstraction loss

---

#Scope & Non-Goals

prepstack does not aim to replace:

⦁ pandas

⦁ scikit-learn

⦁ specialized AutoML frameworks

⦁ Instead, it standardizes the glue logic between them.