4. Data Types (dtypes)

NumPy dtypes control memory and precision.

a = np.array([1, 2, 3], dtype=np.int8)
a[0] = 200            # overflow!
print(a)              # [-56   2   3]   (200 wraps around int8)

⚠️ Common Mistake: Integer overflow is silent. int8 holds only -128…127. Use a wider dtype for large values.

Changing dtypes & memory optimization#

big = np.arange(1_000_000)                 # int64 -> 8 MB
small = big.astype(np.int32)               # -> 4 MB
print(big.nbytes, small.nbytes)            # 8000000 4000000

🚀 Best Practice: Downcast to the smallest dtype that safely fits your data (int32, float32) to halve memory on large datasets — crucial before feeding data to ML models.

⭐ Interview Question: astype vs view for dtype change? astype converts values into a new buffer (safe). Reinterpreting bytes via .view(dtype) reuses the same bytes with a different interpretation (dangerous, rarely what you want).