Pre-filling: np.<empty|zeros|ones|full>[_like]

• filled with garbage (not clearing memory, fastest)
  • np.empty(SHAPE, dtype=float64)
  • np.empty_like(ARRAY)
• filled with 0s
  • np.zeros(SHAPE, dtype=float64)
  • np.zeros_like(ARRAY)
• filled with 1s
  • np.ones(SHAPE, dtype=float64)
  • np.ones_like(ARRAY)
• filled with VALUE
  • np.full(SHAPE, VALUE, dtype=DTYPE)
  • np.full_like(ARRAY, VALUE)

import numpy as np
a = np.zeros((4, 4), dtype = np.int64)
a

array([[0, 0, 0, 0],
       [0, 0, 0, 0],
       [0, 0, 0, 0],
       [0, 0, 0, 0]])

• Example: The _like variants inherit shape and dtype
• Example: Set dtype to int to get CPU-native integers
• Example: np.<empty|zeros|ones> default to dtype=float64
• Example: np.full infers DTYPE from VALUE