Set scale_to_one to false in *all* cases

mokapot/peps.py

@@ -1,7 +1,7 @@
 from __future__ import annotations
 
-from typing import TypeVar, Callable, Iterator
 import logging
+from typing import TypeVar, Callable, Iterator
 
 import numpy as np
 import scipy.stats as stats
@@ -76,9 +76,7 @@ def peps_from_scores(
     if pep_function is not None:
         return pep_function(scores, targets, is_tdc)
     else:
-        raise ValueError(
-            f"Unknown pep algorithm in peps_from_scores: {pep_algorithm}"
-        )
+        raise ValueError(f"Unknown pep algorithm in peps_from_scores: {pep_algorithm}")
 
 
 @typechecked
@@ -417,9 +415,7 @@ def peps_from_scores_kde_nnls(
         factor = (~targets).sum() / targets.sum()
     else:
         # Estimate pi0 and estimate number of correct targets
-        factor = estimate_pi0_by_slope(
-            target_pdf, decoy_pdf, pi0_estimation_threshold
-        )
+        factor = estimate_pi0_by_slope(target_pdf, decoy_pdf, pi0_estimation_threshold)
 
     correct = target_pdf - decoy_pdf * factor
     correct = np.clip(correct, 0, None)
@@ -526,9 +522,7 @@ def make_perm_mat(rows, cols):
     tol = 1e-7
     d, _, mode = _nnls(W @ A @ R, W @ k, tol=tol)
     if mode != 1:
-        LOGGER.debug(
-            "\t - Warning: nnls went into loop. Taking last solution."
-        )
+        LOGGER.debug("\t - Warning: nnls went into loop. Taking last solution.")
     p = np.cumsum(R @ d)
 
     if not ascending:
@@ -744,7 +738,7 @@ def peps_from_scores_hist_nnls(
         Scores and targets come from competition, rather than separate search.
     scale_to_one:
         Boolean value indicating whether to scale the PEP estimates to 1 for
-        small score values. Default is True.
+        small score values. Default is False.
 
     Returns
     -------
@@ -763,7 +757,7 @@ def peps_from_scores_hist_nnls(
 def peps_func_from_hist_nnls(
     hist_data: TDHistData,
     is_tdc: bool,
-    scale_to_one: bool = True,
+    scale_to_one: bool = False,
     weight_exponent: float = -1.0,
 ) -> Callable[[np.ndarray[float]], np.ndarray[float]]:
     """Compute a function that calculates the PEP (Posterior Error Probability)
@@ -779,7 +773,7 @@ def peps_func_from_hist_nnls(
     is_tdc:
         Scores and targets come from competition, rather than separate search.
     scale_to_one: Scale the result if the maximum PEP is smaller than 1.
-         (Default value = True)
+         (Default value = False)
     weight_exponent:
          The weight exponent for the `fit_nnls` fit (see there, default 1).
 
@@ -799,9 +793,7 @@ def peps_func_from_hist_nnls(
     # Do monotone fit, minimizing || n - diag(p) * k || with weights n over
     # monotone descending p
     try:
-        pep_est = fit_nnls(
-            n, k, ascending=False, weight_exponent=weight_exponent
-        )
+        pep_est = fit_nnls(n, k, ascending=False, weight_exponent=weight_exponent)
     except RuntimeError as e:
         raise PepsConvergenceError from e
 
@@ -811,6 +803,4 @@ def peps_func_from_hist_nnls(
     # Linearly interpolate the pep estimates from the eval points to the scores
     # of interest (keeping monotonicity) clip in case we went slightly out of
     # bounds
-    return lambda scores: np.clip(
-        np.interp(scores, eval_scores, pep_est), 0, 1
-    )
+    return lambda scores: np.clip(np.interp(scores, eval_scores, pep_est), 0, 1)