RuntimeWarning: overflow encountered in exp

overflow encountered in exp

$ python - <<'PY'
import numpy as np
np.exp(1000)
PY
/home/user/.local/lib/python3.10/site-packages/numpy/core/_methods.py:...: RuntimeWarning: overflow encountered in exp
  return umr_minimum(a, axis, None, keepdims)
# result will be inf

Why this happens

Exponential grows very quickly; inputs beyond float range produce infinities. NumPy emits a RuntimeWarning rather than crashing.

Fix

Clip or scale inputs, use log-space computations when possible (log-sum-exp trick), or use higher precision libraries.

Wrong code

import numpy as np
np.exp(1000)

Fixed code

import numpy as np
import math
# use clipping
np.exp(np.clip(1000, None, 700))

# or compute in log-space
def logsumexp(a):
    m = np.max(a)
    return m + np.log(np.sum(np.exp(a - m)))