numpy/recurrence-loop
| Rule | recurrence-loop |
|---|---|
| Pack | NumPy |
| Severity | warning |
| Fix | none, edit by hand |
| Library versions | every version |
A loop that computes each value from the one before runs Python once for every step. Use scipy.signal.lfilter for a linear step and np.cumsum for a running total.
log_price = np.empty(days + 1)log_price[0] = firstfor t in range(1, days + 1): log_price[t] = intercept + slope * log_price[t - 1] + sd * rng.standard_normal()shocks = intercept + sd * rng.standard_normal(days)later, _ = scipy.signal.lfilter([1.0], [1.0, -slope], shocks, zi=[slope * first])log_price = np.concatenate([[first], later])What the rule reports
Section titled “What the rule reports”The rule reports a for loop over range that writes x[t] from x[t - 1] of the same array.
How to fix it
Section titled “How to fix it”xpd has no automatic fix, so make the change by hand.
The loop is a linear recurrence, which scipy.signal.lfilter computes in one call. On NumPy
2.5.5 and SciPy 1.18.1 the two forms differed by 4e-15 over 5,000 steps, with the same random
draws. A running sum is np.cumsum, a running product np.cumprod, and a running maximum
np.maximum.accumulate. A step that is not linear in the earlier value has to stay a loop.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
This loop computes $OUT[$T] from $OUT[$T - 1] one step per pass. For a linear step use scipy.signal.lfilter, as in lfilter([1.0], [1.0, -slope], shocks, zi=[slope * first]), and for a running total np.cumsum. Keep the loop only when the step is not linear.Browse the other rules of the NumPy pack.