How to save and load NumPy arrays with NPZ

Related arrays often need to cross a process boundary without being split into separate files or flattened into text. An NPZ archive keeps those arrays in one NumPy-native container while preserving each member's shape and dtype.

The np.savez_compressed() function stores each keyword argument under that keyword, so descriptive names such as features and labels replace positional names such as arr_0. Compression reduces archive size at the cost of more work while writing; np.savez() provides the same multi-array layout without ZIP compression.

Loading an NPZ archive returns a dictionary-like NpzFile whose members are read by key. A context manager closes the underlying file handle, and allow_pickle=False keeps this numeric example from accepting pickled object arrays.

1. Create array-save-load-npz.py containing the archive path plus two source arrays.

   array-save-load-npz.py

   import numpy as np
   from pathlib import Path
    
   path = Path("model-batch.npz")
   features = np.arange(12, dtype=np.float32).reshape(3, 4)
   labels = np.array([0, 1, 1], dtype=np.int64)

2. Add the compressed archive write below the source arrays.

   np.savez_compressed(path, features=features, labels=labels)

   The keyword arguments become the features and labels member names inside the archive.

3. Add the archive round-trip block below the save call.

   with np.load(path, allow_pickle=False) as archive:
       print("keys:", archive.files)
       features_loaded = archive["features"]
       labels_loaded = archive["labels"]
       print("features shape:", features_loaded.shape)
       print("features dtype:", features_loaded.dtype)
       print("labels dtype:", labels_loaded.dtype)
       print("features match:", np.array_equal(features, features_loaded))
       print("labels match:", np.array_equal(labels, labels_loaded))

   The context manager closes the NpzFile after both named arrays have been read.

4. Run the completed script in the target archive directory to verify the round trip.

   $ python3 array-save-load-npz.py
   keys: ['features', 'labels']
   features shape: (3, 4)
   features dtype: float32
   labels dtype: int64
   features match: True
   labels match: True