Speedup and output (#38)

* Make UniformModel more efficient Signed-off-by: Matthias Kümmmerer <[email protected]> * Verbosity parameter in CrossvalMultipleRegularizations Signed-off-by: Matthias Kümmmerer <[email protected]> --------- Signed-off-by: Matthias Kümmmerer <[email protected]>
matthias-k · Nov 18, 2023 · 0a48f5d · 0a48f5d
1 parent a66a5c5
commit 0a48f5d
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 8 deletions.
diff --git a/pysaliency/baseline_utils.py b/pysaliency/baseline_utils.py
@@ -356,11 +356,12 @@ def _normalize_regularization_factors(args):
 
 class CrossvalMultipleRegularizations(object):
     """ Class for computing crossvalidation scores of a fixation KDE with multiple regularization models"""
-    def __init__(self, stimuli, fixations, regularization_models: OrderedDict, crossvalidation):
+    def __init__(self, stimuli, fixations, regularization_models: OrderedDict, crossvalidation, verbose=False):
         self.stimuli = stimuli
         self.fixations = fixations
 
         self.cv = crossvalidation
+        self.verbose = verbose
 
         X_areas = fixations_to_scikit_learn(
             self.fixations, normalize=stimuli,
@@ -406,19 +407,19 @@ def score(self, log_bandwidth, *args, **kwargs):
             bandwidth=10**log_bandwidth,
             regularizations=10**log_regularizations,
             regularizing_log_likelihoods=self.regularization_log_likelihoods),
-            self.X, cv=self.cv, verbose=1).sum() / len(self.X) / np.log(2)
+            self.X, cv=self.cv, verbose=self.verbose).sum() / len(self.X) / np.log(2)
         val += np.log2(self.mean_area)
         return val
 
 
 class CrossvalGoldMultipleRegularizations(CrossvalMultipleRegularizations):
-    def __init__(self, stimuli, fixations, regularization_models):
+    def __init__(self, stimuli, fixations, regularization_models, verbose=False):
         if fixations.subject_count > 1:
             crossvalidation_factory = ScikitLearnImageSubjectCrossValidationGenerator
         else:
             crossvalidation_factory = ScikitLearnWithinImageCrossValidationGenerator
 
-        super().__init__(stimuli, fixations, regularization_models, crossvalidation_factory=crossvalidation_factory)
+        super().__init__(stimuli, fixations, regularization_models, crossvalidation_factory=crossvalidation_factory, verbose=verbose)
 
 
 # baseline models

diff --git a/pysaliency/models.py b/pysaliency/models.py
@@ -407,10 +407,13 @@ def _log_density(self, stimulus):
         return np.zeros((stimulus.shape[0], stimulus.shape[1])) - np.log(stimulus.shape[0]) - np.log(stimulus.shape[1])
 
     def log_likelihoods(self, stimuli, fixations, verbose=False):
-        lls = []
-        for n in fixations.n:
-            lls.append(-np.log(stimuli.shapes[n][0]) - np.log(stimuli.shapes[n][1]))
-        return np.array(lls)
+        stimulus_shapes = np.zeros((len(stimuli), 2), dtype=int)
+        stimulus_indices = sorted(np.unique(fixations.n))
+        for stimulus_index in stimulus_indices:
+            stimulus_shapes[stimulus_index] = stimuli.stimulus_objects[stimulus_index].size
+
+        stimulus_log_likelihoods = -np.log(stimulus_shapes).sum(axis=1)
+        return stimulus_log_likelihoods[fixations.n]
 
 
 class MixtureModel(Model):