[WIP/ENH] - Add metrics in linear space

specparam/metrics/definitions.py

@@ -14,35 +14,52 @@
     measure='mae',
     description='Mean absolute error of the model fit to the data.',
     func=compute_mean_abs_error,
+    space='log',
 )
 
 error_mse = Metric(
     category='error',
     measure='mse',
     description='Mean squared error of the model fit to the data.',
-    func=compute_mean_squared_error
+    func=compute_mean_squared_error,
+    space='log',
 )
 
 error_rmse = Metric(
     category='error',
     measure='rmse',
     description='Root mean squared error of the model fit to the data.',
     func=compute_root_mean_squared_error,
+    space='log',
 )
 
 error_medae = Metric(
     category='error',
     measure='medae',
     description='Median absolute error of the model fit to the data.',
     func=compute_median_abs_error,
+    space='log',
+)
+
+error_maelin = Metric(
+    category='error',
+    measure='maelin',
+    description='Mean absolute error of the model fit to the data, in linear space.',
+    func=compute_mean_abs_error,
+    space='linear',
 )
 
 # Collect available error metrics
 ERROR_METRICS = {
+
+    # log spacing
     'mae' : error_mae,
     'mse' : error_mse,
     'rmse' : error_rmse,
     'medae' : error_medae,
+
+    # linear spacing
+    'maelin' : error_maelin,
 }
 
 ###################################################################################################
@@ -53,6 +70,15 @@
     measure='rsquared',
     description='R-squared between the model fit and the data.',
     func=compute_r_squared,
+    space='log',
+)
+
+gof_rsquaredlin = Metric(
+    category='gof',
+    measure='rsquaredlin',
+    description='R-squared between the model fit and the data, in linear space.',
+    func=compute_r_squared,
+    space='linear',
 )
 
 gof_adjrsquared = Metric(
@@ -62,12 +88,18 @@
     func=compute_adj_r_squared,
     kwargs={'n_params' : lambda data, results: \
             results.params.periodic.params.size + results.params.aperiodic.params.size},
+    space='log',
 )
 
 # Collect available error metrics
 GOF_METRICS = {
+
+    # log spacing
     'rsquared' : gof_rsquared,
     'adjrsquared' : gof_adjrsquared,
+
+    # linear spacing
+    'rsquaredlin' : gof_rsquaredlin,
 }
 
 ###################################################################################################

specparam/metrics/metric.py

@@ -18,20 +18,23 @@ class Metric():
         Description of the metric.
     func : callable
         The function that computes the metric.
+    space : {'log', 'linear'}
+        Spacing of the data & model to use for metric evaluation.
     kwargs : dictionary
         Additional keyword argument to compute the metric.
         Each key should be the name of the additional argument.
         Each value should be a lambda function that takes 'data' & 'results'
         and returns the desired parameter / computed value.
     """
 
-    def __init__(self, category, measure, description, func, kwargs=None):
+    def __init__(self, category, measure, description, func, space='log', kwargs=None):
         """Initialize metric."""
 
         self.category = category
         self.measure = measure
         self.description = description
         self.func = func
+        self.space = space
         self.result = np.nan
         self.kwargs = {} if not kwargs else kwargs
 
@@ -76,7 +79,10 @@ def compute_metric(self, data, results):
         for key, lfunc in self.kwargs.items():
             kwargs[key] = lfunc(data, results)
 
-        self.result = self.func(data.power_spectrum, results.model.modeled_spectrum, **kwargs)
+        self.result = self.func(
+            data.get_data('full', space=self.space),
+            results.model.get_component('full', space=self.space),
+            **kwargs)
 
 
     def reset(self):

specparam/tests/metrics/test_metric.py

@@ -20,7 +20,7 @@ def test_metric(tfm):
 def test_metric_kwargs(tfm):
 
     metric = Metric('gof', 'ar2', 'Description.', compute_adj_r_squared,
-                    {'n_params' : lambda data, results: \
+                    kwargs={'n_params' : lambda data, results: \
                         results.params.periodic.params.size + results.params.aperiodic.params.size})
 
     assert isinstance(metric, Metric)