scipy/log-of-special
| Rule | log-of-special |
|---|---|
| Pack | SciPy |
| Severity | warning |
| Fix | unsafe, with xpd check --fix --unsafe-fixes |
| Library versions | every version |
A scipy.special function such as ndtr underflows to 0 far in a tail, so its log is -inf there. Call the log form, such as log_ndtr.
loglike = np.log(special.ndtr(q * xb)).sum()loglike = special.log_ndtr(q * xb).sum()What the rule reports
Section titled “What the rule reports”The rule reports np.log of a scipy.special function that has a log form of its own, under
any import of scipy.special. The table shows where the old form overflowed or lost digits.
| Old form | Where the old form failed | Write instead |
|---|---|---|
np.log(special.ndtr(x)) |
-inf below -37.75 | special.log_ndtr(x) |
np.log(special.gamma(x)) |
inf at 200 | special.gammaln(x) |
np.log(special.factorial(n)) |
inf at 200 | special.gammaln(n + 1) |
np.log(special.beta(a, b)) |
-inf at a = b = 1e6 | special.betaln(a, b) |
np.log(special.expit(x)) |
3 or more digits lost above 8 and below -710 | special.log_expit(x) |
np.log(special.softmax(x)) |
-inf for an entry more than 745 below the largest | special.log_softmax(x) |
How to fix it
Section titled “How to fix it”xpd check --fix --unsafe-fixes applies the fixes of this rule. Check the result, because an unsafe fix can change what the code does.
The unsafe fix writes special.log_ndtr(x), special.gammaln(x), special.gammaln(n + 1), special.betaln(a, b), special.log_expit(x), special.log_softmax(x). log_expit needs SciPy 1.8 and log_softmax SciPy 1.5. A function imported by name is fixed only in a file that imports the log form from scipy.special.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
$SP.ndtr($X) underflows to 0 below about -37.5, so its log is -inf there. At -40 that lost all 16 digits where $SP.log_ndtr lost 0.15. Write $SP.log_ndtr($X).Browse the other rules of the SciPy pack.