Free tools Windows power users keep installed
One-click scans. No signup required.
Choose the file format based on what you need to do with the array: use np.save and np.load for a NumPy-native round trip, np.savetxt for readable numeric text, CSV for tabular exchange, or JSON for nested application data. The examples below show how to write and read each format, plus the main fidelity and safety trade-offs.
Which format should you choose?
| Format | Best fit | Main benefit | Important trade-off |
|---|---|---|---|
.npy |
One array that you will load back into NumPy | NumPy’s binary format is designed for saving and loading arrays | It is not intended to be human-readable text |
.npz |
Several named arrays in one file | Groups multiple arrays in one NumPy archive; an option is compressed | Reading it requires NumPy-compatible software |
| Text or delimited text | Inspection or simple numeric exchange | Readable, with delimiter and formatting controls | np.savetxt supports one- and two-dimensional arrays; text conversion choices matter |
| CSV | Tabular exchange with spreadsheets or other tools | Familiar rows and columns | CSV does not retain NumPy dtype or shape metadata, and other programs may infer types differently |
| JSON | Nested values used in application data or interchange | Common text structure for nested data | Convert the array to lists, and preserve or reapply dtype and shape explicitly if exact reconstruction matters |
For durable NumPy-specific storage, prefer .npy or .npz over raw binary methods such as ndarray.tofile and np.fromfile. NumPy cautions that raw file I/O loses endianness and precision information. See NumPy’s file I/O guidance.
Save one array in NumPy’s binary format
Use np.save to write an array to an .npy file, then np.load to read it:
import numpy as np
arr = np.array([[1, 2], [3, 4]])
np.save("array.npy", arr)
restored = np.load("array.npy", allow_pickle=False)If you pass a filename or path without the .npy extension, np.save appends it. The save API defaults to allow_pickle=True; set it to False when you do not need object arrays. See the NumPy save reference.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →#1 Best Overall
Pickle-enabled object arrays carry security and portability drawbacks. Do not load pickle-enabled files from untrusted sources. Keep the load setting compatible with the file you are opening, and use allow_pickle=False when object dtype is unnecessary; NumPy explains the risk in its I/O guidance.
Save several arrays in one archive
Use np.savez for an uncompressed archive or np.savez_compressed for its compressed counterpart. Supply names as keyword arguments so you can retrieve arrays by name:
np.savez("arrays.npz", first=arr, second=arr * 2)
with np.load("arrays.npz", allow_pickle=False) as data:
first = data["first"]
second = data["second"]
np.savez_compressed("arrays-compressed.npz", first=arr, second=arr * 2)The archive APIs are listed in NumPy’s input and output reference.
Write readable text or numeric CSV
For a simple numeric matrix, np.savetxt writes text and accepts a delimiter. Use a matching delimiter when reading it back:
np.savetxt("array.txt", arr)
np.savetxt("array.csv", arr, delimiter=",")
restored = np.loadtxt("array.csv", delimiter=",")np.savetxt is documented for one- or two-dimensional arrays. Its formatting and delimiter options let you control the text representation, but text is not the same as preserving all NumPy array metadata. For data with missing values or more involved parsing needs, consider np.genfromtxt and choose its missing-value policy deliberately. See the NumPy I/O API index and file I/O guidance.
Use Python’s CSV module for general rows
When values need CSV quoting, may contain delimiters, or are irregular text rather than a simple numeric matrix, Python’s csv module is often a better fit:
import csv
with open("rows.csv", "w", newline="", encoding="utf-8") as f:
writer = csv.writer(f)
writer.writerows(arr.tolist())When passed a file object, Python recommends opening it with newline="". The writer stringifies non-string values. On reading, csv.reader returns strings by default, so convert fields explicitly when you need numeric values. CSV dialects also differ between applications; confirm the delimiter, quoting, header, encoding, and line-ending expectations of the program that will read the file. See the Python CSV documentation.
Write an array as JSON
Python’s JSON encoder does not directly serialize a NumPy ndarray. Convert it to nested built-in Python lists with tolist(), then dump the result:
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →import json
import numpy as np
arr = np.array([[1, 2], [3, 4]])
with open("array.json", "w", encoding="utf-8") as f:
json.dump(arr.tolist(), f)
with open("array.json", encoding="utf-8") as f:
nested = json.load(f)
restored = np.array(nested)json.load returns ordinary Python containers, not an ndarray. Reconstruct one with np.array when needed. If exact dtype or shape matters—particularly for empty arrays, unusual dtypes, or application-specific values—include that metadata in a documented schema and restore it deliberately. NumPy describes the list conversion approach in its file I/O guide; supported JSON value mappings are in the Python JSON documentation.
Python’s JSON encoder allows NaN and infinities by default, even though they are outside strict JSON. Set allow_nan=False if you want the encoder to raise ValueError for those values, and define the intended policy for non-finite numbers. Also, JSON is not a framed protocol: repeated calls to json.dump on the same file do not create one valid JSON document.
Handle large arrays and choose deliberately
For large .npy files, NumPy supports memory mapping through np.load(..., mmap_mode=...), which can provide access without reading the entire file into memory at once. Memory mapping is not chunking or compression. For exact behavior and mode options, consult NumPy’s I/O guide.
Quick Recap
- Choose
.npyfor one array that will be reused in NumPy. - Choose
.npzto group named arrays. - Choose text or CSV when people or other tools need readable rows, and account for conversions and consumer-specific CSV rules.
- Choose JSON for nested application data, converting with
tolist()and documenting metadata or non-finite-number handling as needed.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors
