Files, JSON and CSV

Programs forget everything when they stop. To keep data — your notes, a dataset, the answers a model gave — you write it to a file. AI work is full of files: CSV spreadsheets of training data, JSON configuration, JSON Lines datasets for fine-tuning. This page covers all three.

Reading and writing text files

from pathlib import Path

notes = Path("notes.txt")

# Write (creates the file, or replaces it)
with open(notes, "w", encoding="utf-8") as f:
    f.write("Python is fun\n")
    f.write("পাইথন শেখা সহজ\n")

# Append a line
with open(notes, "a", encoding="utf-8") as f:
    f.write("Files are easy\n")

# Read line by line
with open(notes, encoding="utf-8") as f:
    for number, line in enumerate(f, start=1):
        print(number, line.rstrip())

print(notes.exists(), notes.suffix, notes.stat().st_size, "bytes")

1 Python is fun
2 পাইথন শেখা সহজ
3 Files are easy
True .txt 68 bytes

• with open(...) as f: opens the file and always closes it at the end of the block, even if an error happens. Always use with.
• The mode: "r" read (the default), "w" write — replaces the whole file, "a" append to the end.
• encoding="utf-8": always write it. Without it, Windows may use an old encoding and Bangla text becomes garbage or crashes with a UnicodeDecodeError.
• pathlib.Path represents a file path and works the same on Windows, macOS and Linux. Note how the Bangla line takes more bytes than its 14 characters — UTF-8 uses 3 bytes for each Bangla character.

JSON: the language of APIs

JSON (JavaScript Object Notation) is a text format for structured data. It looks almost exactly like Python dictionaries and lists, and every AI API sends and receives it. The json module converts both ways:

import json

settings = {
    "model": "gpt-5.5",
    "temperature": 0.2,
    "languages": ["en", "bn"],
    "greeting": "স্বাগতম",
}

# Python → JSON text in a file
with open("settings.json", "w", encoding="utf-8") as f:
    json.dump(settings, f, ensure_ascii=False, indent=2)

# JSON text → Python
with open("settings.json", encoding="utf-8") as f:
    loaded = json.load(f)

print(loaded["languages"][1], loaded["greeting"])
print(json.dumps({"ok": True, "items": None}))   # to a string, not a file

bn স্বাগতম
{"ok": true, "items": null}

The file it wrote, settings.json:

{
  "model": "gpt-5.5",
  "temperature": 0.2,
  "languages": [
    "en",
    "bn"
  ],
  "greeting": "স্বাগতম"
}

Notice the small differences: JSON writes true, false and null where Python has True, False and None. ensure_ascii=False keeps Bangla readable in the file instead of turning it into \u09b8 codes, and indent=2 makes it pretty.

CSV: spreadsheets as text

A CSV (comma-separated values) file is a table in plain text — what you get when you export from Excel or Google Sheets. Here is reviews.csv:

product,stars,comment
Headphones,5,Great sound
Charger,2,Stopped working
Keyboard,4,Nice to type on

csv.DictReader turns each row into a dictionary, using the first line as the keys:

import csv

with open("reviews.csv", newline="", encoding="utf-8") as f:
    rows = list(csv.DictReader(f))

print(len(rows), "rows")
print(rows[0])

positive = [r for r in rows if int(r["stars"]) >= 4]
print([r["product"] for r in positive])

3 rows
{'product': 'Headphones', 'stars': '5', 'comment': 'Great sound'}
['Headphones', 'Keyboard']

Everything read from a CSV is text — that is why we wrote int(r["stars"]). For bigger tables you will use pandas (page 14), which does these conversions for you.

JSON Lines: one record per line

AI datasets — fine-tuning data, evaluation sets, batch API jobs — usually use JSON Lines (.jsonl): one complete JSON object on each line. You can add a line without rewriting the file, and read a huge file one line at a time:

import json

examples = [
    {"prompt": "2 + 2", "answer": "4"},
    {"prompt": "Capital of Bangladesh", "answer": "Dhaka"},
]

with open("dataset.jsonl", "w", encoding="utf-8") as f:
    for row in examples:
        f.write(json.dumps(row, ensure_ascii=False) + "\n")

with open("dataset.jsonl", encoding="utf-8") as f:
    for line in f:
        row = json.loads(line)
        print(row["prompt"], "->", row["answer"])

2 + 2 -> 4
Capital of Bangladesh -> Dhaka

Try it yourself

1. Write five of your favourite quotes to quotes.txt, one per line, then read them back and print the longest.
2. Save a dictionary of your study plan (subject → hours) as JSON, load it again and print the total hours.
3. From reviews.csv, write a new file bad_reviews.jsonl containing only reviews with fewer than 3 stars.