[COMPARISON] add some labels to the graph. add trend type...

[COMPARISON] add some labels to the graph. add trend type agreement/classification graph. refactor column name

experiment/meteo_france_data/stations_data/main_station_comparison.py

@@ -2,7 +2,7 @@ from experiment.meteo_france_data.scm_models_data.visualization.study_visualizat
     ALL_ALTITUDES_WITH_20_STATIONS_AT_LEAST
 from experiment.meteo_france_data.stations_data.comparison_analysis import ComparisonAnalysis
 from experiment.meteo_france_data.stations_data.visualization.comparisons_visualization.comparisons_visualization import \
-    ComparisonsVisualization
+    ComparisonsVisualization, path_backup_csv_file
 
 
 def visualize_all_stations():
@@ -48,13 +48,14 @@ def wrong_example3():
 
 
 def quick_metric_analysis():
-    ComparisonsVisualization.visualize_metric()
+    ComparisonsVisualization.visualize_metric(csv_filepath=path_backup_csv_file)
+    # ComparisonsVisualization.visualize_metric()
 
 if __name__ == '__main__':
     # wrong_example3()
     # visualize_fast_comparison()
-    visualize_all_stations()
-    # quick_metric_analysis()
+    # visualize_all_stations()
+    quick_metric_analysis()
     # wrong_example2()
     # visualize_non_nan_station()
     # example()

experiment/meteo_france_data/stations_data/visualization/comparisons_visualization/comparisons_visualization.py

+import os.path as op
+from sklearn.metrics import confusion_matrix
 from collections import OrderedDict
 from itertools import chain
 from typing import Dict, List
@@ -20,12 +22,23 @@ from extreme_estimator.extreme_models.utils import r, safe_run_r_estimator, ro
 from spatio_temporal_dataset.coordinates.abstract_coordinates import AbstractCoordinates
 from utils import classproperty
 
-MAE_COLUMN_NAME = 'mean absolute difference'
 
-DISTANCE_COLUMN_NAME = 'distance'
-path_df_location_to_value_csv_example = r'/home/erwan/Documents/projects/spatiotemporalextremes/experiment/meteo_france_data/stations_data/csv/example.csv'
 
 
+path = r'/home/erwan/Documents/projects/spatiotemporalextremes/experiment/meteo_france_data/stations_data/csv'
+path_last_csv_file = op.join(path, 'example.csv')
+path_backup_csv_file = op.join(path, 'example_backup.csv')
+
+
+def get_reanalysis_column_name(station_column_name):
+    return station_column_name + ' ' + REANALYSE_STR
+
+WELL_CLASSIFIED_CNAME = 'well classified'
+DISTANCE_COLUMN_NAME = 'distance'
+TREND_TYPE_CNAME = 'display trend type'
+SAFRAN_TREND_TYPE_CNAME = get_reanalysis_column_name('display trend type')
+MAE_COLUMN_NAME = 'mean absolute difference'
+
 class ComparisonsVisualization(VisualizationParameters):
 
     def __init__(self, altitudes=None, keep_only_station_without_nan_values=False, margin=150,
@@ -99,24 +112,39 @@ class ComparisonsVisualization(VisualizationParameters):
         return df
 
     @classmethod
-    def visualize_metric(cls, df=None):
+    def visualize_metric(cls, df=None, csv_filepath=path_last_csv_file):
         # Load or update df value from example file
         if df is None:
-            df = pd.read_csv(path_df_location_to_value_csv_example, index_col=[0, 1, 2])
+            df = pd.read_csv(csv_filepath, index_col=[0, 1, 2])
         else:
-            df.to_csv(path_df_location_to_value_csv_example)
-
-        # Compute some column like a classication boolean
-
-        # Display some score spatially
-        df_score = df.groupby([MASSIF_COLUMN_NAME]).mean()
-        s_mae = df_score[MAE_COLUMN_NAME]
-        massif_name_to_value = s_mae.to_dict()
-        AbstractStudy.visualize_study(massif_name_to_value=massif_name_to_value,
+            df.to_csv(csv_filepath)
+
+        if TREND_TYPE_CNAME in df.columns:
+            # Display the confusion matrix
+            print(df[TREND_TYPE_CNAME].values)
+            print(AbstractUnivariateTest.three_main_trend_types())
+            m = confusion_matrix(y_true=df[TREND_TYPE_CNAME].values,
+                                 y_pred=df[SAFRAN_TREND_TYPE_CNAME].values,
+                                 labels=AbstractUnivariateTest.three_main_trend_types())
+            print(m)
+
+            # Display the classification score per massif
+            df[WELL_CLASSIFIED_CNAME] = df[TREND_TYPE_CNAME] == df[SAFRAN_TREND_TYPE_CNAME]
+            serie_classificaiton = df.groupby([MASSIF_COLUMN_NAME]).mean()[WELL_CLASSIFIED_CNAME] * 100
+            AbstractStudy.visualize_study(massif_name_to_value=serie_classificaiton.to_dict(),
+                                          default_color_for_missing_massif='b',
+                                          cmap=plt.cm.Greens,
+                                          vmin=0,
+                                          vmax=100,
+                                          label='agreement on trend type classification (%)' )
+        # Display the mae score
+        serie_mae = df.groupby([MASSIF_COLUMN_NAME]).mean()[MAE_COLUMN_NAME]
+        AbstractStudy.visualize_study(massif_name_to_value=serie_mae.to_dict(),
                                       default_color_for_missing_massif='b',
                                       cmap=plt.cm.Reds,
-                                      vmin=s_mae.min(),
-                                      vmax=s_mae.max())
+                                      vmin=0,
+                                      vmax=65,
+                                      label='average absolute difference between annual maxima snowfall (mm)')
 
     def _visualize_ax_main(self, plot_function, comparison: ComparisonAnalysis, massif, ax=None, show=False):
         if ax is None:
@@ -164,7 +192,7 @@ class ComparisonsVisualization(VisualizationParameters):
 
             if isinstance(plot_ordered_value_dict, dict):
                 if REANALYSE_STR in i:
-                    plot_station_ordered_value_dict = OrderedDict([(k + ' ' + REANALYSE_STR, v)
+                    plot_station_ordered_value_dict = OrderedDict([(get_reanalysis_column_name(k), v)
                                                                    for k, v in plot_ordered_value_dict.items()])
                 else:
                     ordered_value_dict.update(plot_ordered_value_dict)
@@ -214,7 +242,7 @@ class ComparisonsVisualization(VisualizationParameters):
             most_likely_trend_type = trend_test_res[best_idx][0]
             display_trend_type = AbstractUnivariateTest.get_display_trend_type(real_trend_type=most_likely_trend_type)
             label += "\n {} starting in {}".format(display_trend_type, most_likely_year)
-            ordered_dict['display trend type'] = display_trend_type
+            ordered_dict[TREND_TYPE_CNAME] = display_trend_type
             ordered_dict['most likely year'] = most_likely_year
             # Display the nllh against the starting year
             step = 1

experiment/trend_analysis/univariate_test/abstract_univariate_test.py

@@ -49,6 +49,10 @@ class AbstractUnivariateTest(object):
         return [cls.POSITIVE_TREND, cls.NEGATIVE_TREND,
                 cls.SIGNIFICATIVE_POSITIVE_TREND, cls.SIGNIFICATIVE_NEGATIVE_TREND, cls.NO_TREND]
 
+    @classmethod
+    def three_main_trend_types(cls):
+        return [cls.SIGNIFICATIVE_NEGATIVE_TREND, cls.NON_SIGNIFICATIVE_TREND, cls.SIGNIFICATIVE_POSITIVE_TREND]
+
     @classmethod
     def display_trend_type_to_style(cls):
         d = OrderedDict()