diff --git a/toolkit.py b/toolkit.py
index 8970eb8620a48e7c1f735e9cca6727ef235c2a62..db11d81d5ea79819871acec39275c61f85a854ec 100755
--- a/toolkit.py
+++ b/toolkit.py
@@ -1047,22 +1047,32 @@ class ClassificationProject(object):
logger.debug("Plotting sig (min={}, max={}) from {}".format(np.min(sig), np.max(sig), sig))
# calculate percentiles to get a heuristic for the range to be plotted
+ x_total = np.concatenate([bkg, sig])
+ w_total = np.concatenate([bkg_weights, sig_weights])
plot_range = weighted_quantile(
- np.concatenate([bkg, sig]),
+ x_total,
[0.01, 0.99],
- sample_weight=np.concatenate([bkg_weights, sig_weights]),
+ sample_weight=w_total,
)
logger.debug("Calculated range based on percentiles: {}".format(plot_range))
+ bins = 50
+
+ # check if we have a distribution of integer numbers (e.g. njet or something categorical)
+ # in that case we want to have a bin for each number
+ if (x_total == x_total.astype(int)).all():
+ plot_range = (math.floor(plot_range[0])-0.5, math.ceil(plot_range[1])+0.5)
+ bins = int(plot_range[1]-plot_range[0])
+
try:
- centers_sig, hist_sig, _ = self.get_bin_centered_hist(sig, bins=50, range=plot_range, weights=sig_weights)
- centers_bkg, hist_bkg, _ = self.get_bin_centered_hist(bkg, bins=50, range=plot_range, weights=bkg_weights)
+ centers_sig, hist_sig, _ = self.get_bin_centered_hist(sig, bins=bins, range=plot_range, weights=sig_weights)
+ centers_bkg, hist_bkg, _ = self.get_bin_centered_hist(bkg, bins=bins, range=plot_range, weights=bkg_weights)
except ValueError:
# weird, probably not always working workaround for a numpy bug
plot_range = (float("{:.3f}".format(plot_range[0])), float("{:.3f}".format(plot_range[1])))
logger.warn("Got a value error during plotting, maybe this is due to a numpy bug - changing range to {}".format(plot_range))
- centers_sig, hist_sig, _ = self.get_bin_centered_hist(sig, bins=50, range=plot_range, weights=sig_weights)
- centers_bkg, hist_bkg, _ = self.get_bin_centered_hist(bkg, bins=50, range=plot_range, weights=bkg_weights)
+ centers_sig, hist_sig, _ = self.get_bin_centered_hist(sig, bins=bins, range=plot_range, weights=sig_weights)
+ centers_bkg, hist_bkg, _ = self.get_bin_centered_hist(bkg, bins=bins, range=plot_range, weights=bkg_weights)
width = centers_sig[1]-centers_sig[0]
ax.bar(centers_bkg, hist_bkg, color="b", alpha=0.5, width=width)
@@ -1139,11 +1149,7 @@ class ClassificationProject(object):
lumifactor=None, apply_class_weight=True,
invert_activation=False):
if invert_activation:
- if not self.activation_function_output == "sigmoid":
- raise NotImplementedError("Inverse function of {} not supported yet - "
- "currently only sigmoid"
- .format(self.activation_function_output))
- trf = lambda y : np.log(y/(1-y))
+ trf = self.get_inverse_act_fn()
else:
trf = lambda y : y
fig, ax = plt.subplots()
@@ -1191,7 +1197,7 @@ class ClassificationProject(object):
return save_show(plt, fig, os.path.join(self.project_dir, "scores.pdf"))
- def plot_significance_hist(self, lumifactor=1., significance_function=None, plot_opts=dict(bins=50, range=(0, 1))):
+ def plot_significance_hist(self, lumifactor=1., significance_function=None, plot_opts=dict(bins=50, range=(0, 1)), invert_activation=False):
"""
Plot significances based on a histogram of scores
@@ -1199,10 +1205,15 @@ class ClassificationProject(object):
logger.info("Plot significances")
- centers_sig_train, hist_sig_train, rel_errors_sig_train = self.get_bin_centered_hist(self.scores_train[self.y_train==1].reshape(-1), weights=self.w_train[self.y_train==1], **plot_opts)
- centers_bkg_train, hist_bkg_train, rel_errors_bkg_train = self.get_bin_centered_hist(self.scores_train[self.y_train==0].reshape(-1), weights=self.w_train[self.y_train==0], **plot_opts)
- centers_sig_test, hist_sig_test, rel_errors_sig_test = self.get_bin_centered_hist(self.scores_test[self.y_test==1].reshape(-1), weights=self.w_test[self.y_test==1], **plot_opts)
- centers_bkg_test, hist_bkg_test, rel_errors_bkg_test = self.get_bin_centered_hist(self.scores_test[self.y_test==0].reshape(-1), weights=self.w_test[self.y_test==0], **plot_opts)
+ if invert_activation:
+ trf = self.get_inverse_act_fn()
+ else:
+ trf = lambda y : y
+
+ centers_sig_train, hist_sig_train, rel_errors_sig_train = self.get_bin_centered_hist(trf(self.scores_train[self.y_train==1].reshape(-1)), weights=self.w_train[self.y_train==1], **plot_opts)
+ centers_bkg_train, hist_bkg_train, rel_errors_bkg_train = self.get_bin_centered_hist(trf(self.scores_train[self.y_train==0].reshape(-1)), weights=self.w_train[self.y_train==0], **plot_opts)
+ centers_sig_test, hist_sig_test, rel_errors_sig_test = self.get_bin_centered_hist(trf(self.scores_test[self.y_test==1].reshape(-1)), weights=self.w_test[self.y_test==1], **plot_opts)
+ centers_bkg_test, hist_bkg_test, rel_errors_bkg_test = self.get_bin_centered_hist(trf(self.scores_test[self.y_test==0].reshape(-1)), weights=self.w_test[self.y_test==0], **plot_opts)
significances_train = []
significances_test = []
@@ -1270,7 +1281,15 @@ class ClassificationProject(object):
return s_sumw, np.sqrt(s_sumw2), b_sumw, np.sqrt(b_sumw2), scores_sorted[threshold_idxs]
- def plot_significance(self, significance_function=None, maxsteps=1000, lumifactor=1., vectorized=False):
+ def get_inverse_act_fn(self):
+ if not self.activation_function_output == "sigmoid":
+ raise NotImplementedError("Inverse function of {} not supported yet - "
+ "currently only sigmoid"
+ .format(self.activation_function_output))
+ return lambda y : np.log(y/(1-y))
+
+
+ def plot_significance(self, significance_function=None, maxsteps=1000, lumifactor=1., vectorized=False, invert_activation=False):
"""
Plot the significance when cutting on all posible thresholds and plot against signal efficiency.
"""
@@ -1279,6 +1298,11 @@ class ClassificationProject(object):
vectorized = True
significance_function = poisson_asimov_significance
+ if invert_activation:
+ trf = self.get_inverse_act_fn()
+ else:
+ trf = lambda y : y
+
fig, ax = plt.subplots()
ax2 = ax.twinx()
prop_cycle = plt.rcParams['axes.prop_cycle']
@@ -1288,6 +1312,7 @@ class ClassificationProject(object):
(self.scores_test, self.y_test, self.w_test, "test")],
colors
):
+ scores = trf(scores)
s_sumws, s_errs, b_sumws, b_errs, thresholds = self.calc_s_ds_b_db(scores, y, w)
stepsize = int(len(s_sumws))/int(maxsteps)
if stepsize == 0: