Saving and Loading NumPy Data

Why save NumPy arrays?

Saving arrays lets you:

• reuse cleaned/processed features
• speed up workflows (avoid re-parsing CSV)
• share data between scripts/notebooks

NumPy’s own binary format is .npy (one array) / .npz (multiple arrays) — uncompressed by default, and much faster to reload than re-parsing text.

Save and load single array: .npy

Save

save

import numpy as np
 
arr = np.arange(10)
np.save("my_array.npy", arr)

Load

load

import numpy as np
 
arr2 = np.load("my_array.npy")
print(arr2)

diagram Choosing a NumPy save format mermaid

  flowchart TD
  A["Need to persist array data"] --> B{"One array or several?"}
  B -- "One" --> C["np.save('file.npy', arr)"]
  B -- "Several" --> D["np.savez('file.npz', a=arr1, b=arr2)"]
  A --> E{"Need it human-readable
or shared as plain text?"}
  E -- "Yes" --> F["np.savetxt / np.loadtxt"]
  D --> G{"Compresses well?"}
  G -- "Yes" --> H["np.savez_compressed"]

Pick .npy/.npz for fast, exact round-trips of array data; use text formats only for simple, human-readable numeric matrices.

Save and load multiple arrays: .npz

If the file path doesn’t end in .npy/.npz, NumPy appends the extension automatically.

npz

import numpy as np
 
x = np.arange(5)
y = np.arange(5) ** 2
 
np.savez("data.npz", x=x, y=y)
 
loaded = np.load("data.npz")
print(loaded.files)
print(loaded["x"], loaded["y"])

If the saved data compresses well, np.savez_compressed produces a smaller file at the cost of a bit more CPU time:

compressed

import numpy as np
 
arr = np.arange(10)
np.savez_compressed("arrays_compressed.npz", a=arr, b=arr)

Text formats (CSV-like)

Save to text

savetxt

import numpy as np
 
mat = np.array([[1.1, 2.2], [3.3, 4.4]])
np.savetxt("matrix.csv", mat, delimiter=",", fmt="%.2f")

Load from text

loadtxt

import numpy as np
 
mat2 = np.loadtxt("matrix.csv", delimiter=",")
print(mat2)

Note

For real-world analytics, Pandas is usually better for CSV/Excel. NumPy text files are great for simple numeric matrices.

From the book

McKinney is upfront that this section covers only NumPy’s built-in binary format — for loading tabular or text data (like CSVs with headers and mixed types), pandas gives a much friendlier and more forgiving experience. Reach for np.save/np.load when you’re persisting pure numeric arrays between steps of a NumPy-only pipeline.

Tips

• For best speed and exact dtype preservation: use .npy / .npz.
• Use .npz when you want one file containing multiple arrays.

Next

Continue to: Conditional Logic, Sorting & Set Logic to express if-style logic as array operations and use sort, unique, and set functions.

🧪 Try It Yourself

Exercise 1 – Save a Single Array

pch.exerciseTag pch.exerciseTitle

Exercise 2 – Save Multiple Arrays with savez

pch.exerciseTag pch.exerciseTitle

Exercise 3 – Round-trip Through a Text File

pch.exerciseTag pch.exerciseTitle