quantbeckman Golden Python rules

 quantbeckman Golden Python rules with examples

1. No Python loops in the critical path

Instead of using for, while, or comprehensions, rely on vectorized operations.

Bad:

import numpy as np

x = np.random.rand(1000000)
y = np.zeros_like(x)

for i in range(len(x)):
    y[i] = x[i] ** 2 + 3

Good:

y = x**2 + 3  # fully vectorized

✅ Python never iterates manually; the work is done in compiled NumPy code.

---

2. No Python objects in the critical path

Use typed arrays (numpy.ndarray, array.array, numba typed memory) instead of lists/dicts/tuples.

Bad:

lst = [i*i for i in range(1000000)]

Good:

import numpy as np
arr = np.arange(1000000, dtype=np.float64)
arr = arr**2

Here, the array stores typed floats, no Python objects.

---

3. Avoid repeated calls

Function calls in Python are expensive. Fuse operations when possible.

Bad:

y = np.sqrt(np.abs(x))

np.abs creates a new array, then np.sqrt creates another.

Better:

np.sqrt(x**2)  # single fused operation (if applicable)

Or in Numba you can do:

from numba import njit

@njit
def f(x):
    return np.sqrt(x**2)  # no repeated Python calls

---

4. No allocations in the hot path

Preallocate arrays instead of creating new ones repeatedly.

Bad:

for i in range(1000):
    tmp = np.zeros(1000)  # repeated allocation
    tmp += i

Good:

tmp = np.zeros(1000)
for i in range(1000):
    tmp[:] = i  # reuse buffer

---

5. Avoid branches (if/else)

Branches are slow in vectorized code; use masking or lookup tables.

Bad:

y = np.zeros_like(x)
for i in range(len(x)):
    if x[i] > 0:
        y[i] = x[i]
    else:
        y[i] = 0

Good:

y = np.where(x > 0, x, 0)

Or with boolean masking:

y = x.copy()
y[x < 0] = 0

---

6. Avoid conversions

Converting between types or objects costs a lot. Keep dtype consistent.

Bad:

arr = np.array([1,2,3], dtype=np.int32)
arr_float = arr.astype(np.float64)  # expensive

Good:

arr = np.array([1,2,3], dtype=np.float64)

Or preallocate with the correct type to begin with.

---

7. Avoid I/O or logging

Never print, log, or read/write files inside the hot path.

Bad:

for i in range(1000000):
    print(i)  # kills performance

Good:

result = x**2 + 3  # compute first
# log or save only once outside hot loop