Klino — Fast Data Cleaning CLI for ML, NLP, CV & Time Series

One command. Clean data.
Klino is a zero-config command-line tool that prepares raw datasets for machine learning — handling CSV, Excel, text, HTML, images, and time series with a single command.

---

Why Klino?

Data cleaning is the most time-consuming step in any ML project. Klino automates the full pipeline:

• Scan first, clean second — inspect every problem before touching a single row
• Auto-detects file type — no flags, no config files
• Industry-standard algorithms — IQR + Z-score hybrid outlier removal, ADF stationarity test, WordNet lemmatization, letterbox pad-resize
• Two engines — pandas (free, < 50 MB) and polars ⚡ (Pro, any size) with automatic fallback
• Non-destructive — original file is never modified; output is always saved as clean_<filename> in the same folder

---

Installation

pip install klino

Requires Python ≥ 3.11.

---

Quick Start

# 1. Scan your data — see every problem before cleaning
klino scan data.csv

# 2. Clean it — one command does everything
klino ml data.csv

# 3. Verify — scan the output to confirm all issues are gone
klino scan clean_data.csv

Output is saved as clean_data.csv in the same folder as the original file.

---

Commands

klino scan — Inspect Data (Read-Only)

Scans your dataset and reports every issue in a rich table. Nothing is modified.
Run it before cleaning to see how much work needs to be done, and again after cleaning to verify the output is clean.

klino scan data.csv           # ML scan  — CSV / Excel
klino scan corpus.txt         # NLP scan — text / HTML (auto-detected)
klino scan ./images/          # CV scan  — image folder (auto-detected)
klino scan-ts sales.csv       # TS scan  — time series (explicit)
klino scan-nlp reviews.html   # NLP scan — explicit
klino scan-cv ./train/        # CV scan  — explicit

ML scan checks: missing values (per column), duplicate rows, constant columns, outlier rows (IQR + Z-score), category inconsistencies, whitespace issues.

NLP scan checks: HTML tags, URLs, @mentions, #hashtags, emojis, contractions, extra whitespace, duplicate sentences.

CV scan checks: corrupt / unreadable images, duplicate images (MD5 hash), images below 32×32 px, too-dark images (brightness < 20), overexposed images (brightness > 240), inconsistent sizes.

TS scan checks: missing values, duplicate timestamps, future data leakage, irregular frequency, spike / outlier rows (rolling Z-score), non-numeric columns.

---

klino ml — Clean CSV / Excel for ML

klino ml data.csv                        # auto engine
klino ml data.csv --engine pandas        # force pandas
klino ml data.csv --engine polars        # force polars ⚡
klino ml data.csv -o /path/to/output.csv # custom output path
klino ml                                 # scans current folder for CSV/Excel

Pipeline (in order):

1. Auto-fix data types (numeric, datetime coercion)
2. Remove duplicate rows
3. Drop constant columns (zero variance)
4. Fill missing values — numeric → median, categorical → mode
5. Remove outliers — IQR (Tukey's fence) ∩ Z-score (3σ) intersection
6. Normalise category values — strip whitespace, title-case
7. Clean text columns — collapse whitespace

---

klino nlp — Clean Text / HTML for NLP

klino nlp corpus.txt
klino nlp page.html --lang english
klino nlp tweets.csv -o clean_tweets.txt

Pipeline (in order):

1. Strip HTML tags and decode entities
2. Expand contractions (don't → do not)
3. Remove emojis
4. Remove URLs, @mentions, #hashtags
5. Lowercase
6. Remove punctuation
7. Remove stopwords
8. Lemmatize (WordNet — verbs + nouns)
9. Remove duplicate sentences

---

klino cv — Clean Image Folders for CV / Deep Learning

klino cv ./images/
klino cv ./images/ --size 224x224       # custom target size (default: 224×224)
klino cv ./images/ -o ./cleaned/        # custom output folder

Pipeline (in order):

1. Skip corrupt / unreadable images
2. Skip images below minimum dimension (32×32 px)
3. Deduplicate by MD5 hash of raw bytes
4. Filter by brightness — skip too-dark (< 20) and overexposed (> 240)
5. Aspect-ratio–preserving letterbox pad-resize to target size
6. Normalise pixel values to float32 [0, 1]
7. Save as .npy files (compatible with NumPy, PyTorch, TensorFlow)

---

klino ts — Clean Time Series Data

klino ts sensor.csv
klino ts logs.xlsx --engine polars
klino ts data.csv -o clean_data.csv

Pipeline (in order):

1. Detect and parse datetime column, convert to UTC
2. Remove duplicate timestamps
3. Remove future rows (data leakage prevention)
4. Auto-resample to regular frequency (inferred from median gap)
5. Fill missing values via time-aware interpolation → ffill → bfill
6. Remove spikes — rolling Z-score per column (window=5, σ > 3.0)
7. Noise smoothing — rolling mean (window=3)
8. Stationarity check — ADF test + first-order differencing per column

---

License Commands

klino activate YOUR-LICENSE-KEY     # activate Klino Pro (1 machine per key)
klino deactivate                    # release key to move to a new machine
klino status                        # show tier, expiry, days remaining
klino --version                     # show installed version

---

Free vs Pro

Feature	Free	Pro
File size limit	< 50 MB	Unlimited
Engine	pandas	polars ⚡ (CPU-parallel)
All cleaning steps	✅	✅
Files 50–200 MB	❌	✅ Full pipeline
Files ≥ 200 MB	❌	✅ Fast mode (safe steps)
Offline grace period	❌	✅ 7 days
Machines per key	—	1 (transferable)

Pro license: https://vikasawasti.gumroad.com/l/klinopro
Pay → key emailed instantly → klino activate YOUR-KEY

---

Output Files

All cleaned files are saved in the same folder as the input file, prefixed with clean_:

Input	Output
data.csv	clean_data.csv
corpus.txt	clean_corpus.txt
page.html	clean_page.txt
sensor.xlsx	clean_sensor.xlsx
./images/	./images_clean/ (sibling folder)

---

Engines

Engine	Speed	Memory	When
pandas	Baseline	Standard	Files < 50 MB (free tier)
polars ⚡	3–8× faster	Lower	Files ≥ 50 MB (Pro)

Engine is selected automatically based on file size. Override with --engine pandas or --engine polars.

---

Requirements

• Python ≥ 3.11
• Dependencies are installed automatically: pandas, polars, rich, click, nltk, beautifulsoup4, opencv-python-headless, pillow, openpyxl, lxml, requests
• No GPU required — all processing is CPU-only

---

License

MIT — see LICENSE