How to stack NumPy arrays

Stacking keeps several equal-shaped NumPy arrays as separate layers inside one result. It fits repeated measurements, channel data, batches, and other cases where each input has the same positions and the new layer dimension should remain visible.

np.stack() joins arrays along a new axis, so the result has one more dimension than each input. axis=0 puts the input arrays on the first dimension, while axis=-1 places the new dimension at the end.

Use stacking only when every input already has the same shape. If the arrays should extend rows, columns, or another dimension that already exists, np.concatenate() is the better operation.

1. Create a script that stacks two equal-shaped readings along the first axis and the last axis.

   array-stack.py

   import numpy as np
    
   morning = np.array([18, 21, 24])
   evening = np.array([15, 19, 23])
    
   by_reading = np.stack((morning, evening), axis=0)
   by_city = np.stack((morning, evening), axis=-1)
    
   print("morning shape:", morning.shape)
   print("evening shape:", evening.shape)
   print("by reading:")
   print(by_reading)
   print("by reading shape:", by_reading.shape)
   print("by city:")
   print(by_city)
   print("by city shape:", by_city.shape)

   Both inputs have shape (3,). axis=0 creates one row per reading, and axis=-1 pairs the corresponding positions.

1. Run the script and verify the stacked shapes.

   $ python3 array-stack.py
   morning shape: (3,)
   evening shape: (3,)
   by reading:
   [[18 21 24]
    [15 19 23]]
   by reading shape: (2, 3)
   by city:
   [[18 15]
    [21 19]
    [24 23]]
   by city shape: (3, 2)

   The stacked arrays have one more dimension than the inputs. The same two (3,) arrays become (2, 3) with axis=0 and (3, 2) with axis=-1.

1. Test mismatched input shapes before stacking new data.

   $ python3 - <<'PY'
   import numpy as np
   
   a = np.array([1, 2, 3])
   b = np.array([[4, 5, 6]])
   
   try:
       np.stack((a, b))
   except ValueError as error:
       print(f"{type(error).__name__}: {error}")
   PY
   ValueError: all input arrays must have the same shape

   Do not reshape around this error until the new axis meaning is clear. A forced reshape can make unrelated rows or positions look comparable.