deepcsv 0.4.0

Automatically walks through folders and subfolders, finds all CSV and XLSX files, detects and fixes data issues, and saves the results as Parquet files while keeping the exact same folder structure.

deepcsv

Stop losing your data types when working with CSV files.
deepcsv automatically cleans messy CSV/XLSX data and converts it into ML-ready Parquet format.

Installation

pip install deepcsv

Example

Before

# CSV column value
"['a', 'b', 'c']"

After

# Automatically converted
['a', 'b', 'c']

Usage

import deepcsv

df = deepcsv.ConvertListStrToList("path/to/file.csv")

---

What it does

• Walks through all folders and subfolders automatically
• Finds every CSV and XLSX file
• Detects columns that contain list strings like "['item1', 'item2']" and converts them into real Python arrays for faster performance
• Detects columns with mixed data types and tries to fix them automatically
• Warns you when a column has mixed types so you know what was changed
• Saves the results as Parquet files to preserve the converted data types

---

Why Parquet?

CSV files cannot store arrays or preserve data types.
Parquet solves this by keeping the exact types after conversion and is much faster for data processing workflows.

---

Why arrays instead of Python lists?

Arrays are significantly faster for numerical operations and machine learning workflows, especially when working with large datasets.

---

Functions

ConvertListStrToList(file_path)

Reads a CSV file, converts list strings to arrays, fixes mixed-type columns, and returns a clean DataFrame.

import deepcsv

df = deepcsv.ConvertListStrToList("path/to/file.csv")

---

ReadAllCSVData(path)

Walks through all folders and subfolders, applies ConvertListStrToList on every CSV and XLSX file, and saves the results as Parquet files in a new folder called All CSV Data is Converted Here.

import deepcsv

deepcsv.ReadAllCSVData("path/to/folder")

---

Notes

• Only files that contain list string columns are saved as Parquet
• Mixed-type columns are converted to float automatically when possible
• Skips NaN values without breaking
• Requires pyarrow for Parquet support

---

Requirements

• Python >= 3.7
• pandas
• pyarrow