def RegionalMetrics(y_pred, y_test, Regions, lons, lats, lons1, lats1,
filenames_train, filenames_test, savefolder, areas_flat):
Ntest, p = y_pred.shape
print(y_pred.shape)
# check Ntest == 1
full_metrics_dict = {}
Nreg = len(Regions)
# Do global metrics on full grid
metrics_dict = CalcMetrics(y_test[0, :], y_pred[0, :], areas_flat)
# Initialise full metric dictionary and add the full global metrics
keys = metrics_dict.keys()
full_metrics_dict = {"Regions": ["Full Grid"] + Regions + ["All Regions"]}
for key in keys:
full_metrics_dict[key] = np.zeros((Nreg + 2))
full_metrics_dict[key][0] = metrics_dict[key]
# Now do regional metrics by taking average over regions
if p > 1000:
y_pred_reg = split_into_regions(y_pred, areas_flat, lons, lats, lons1,
lats1, Regions)
y_test_reg = split_into_regions(y_test, areas_flat, lons, lats, lons1,
lats1, Regions)
else:
y_pred_reg = y_pred
y_test_reg = y_test
for (i, reg) in zip(range(Nreg), Regions):
print(i, y_pred_reg[0, i])
metrics_dict = CalcMetrics(np.array([y_test_reg[0, i]]),
np.array([y_pred_reg[0, i]]),
np.array([1.0]))
for key in metrics_dict.keys():
full_metrics_dict[key][i + 1] = metrics_dict[key]
metrics_dict = CalcMetrics(y_test_reg[0, :], y_pred_reg[0, :],
np.ones(Nreg))
for key in keys:
full_metrics_dict[key][Nreg + 1] = metrics_dict[key]
colnames = ["Regions"] + list(keys)
print(colnames)
print(full_metrics_dict)
df = pd.DataFrame(data=full_metrics_dict, columns=colnames)
print(df)
df.to_csv(savefolder + 'RegionalMetrics.csv')
return full_metrics_dict
def regional_y(self, regions):
print((self.y.shape))
print(self.area_y.shape)
y_regional = split_into_regions(self.y, self.area_y,
self.lons, self.lats, self.lons1, self.lats1,
regions)
self.y = y_regional
self.y_regions = regions
self.area_y = np.ones(len(regions))
self.y_type = 'Regional'
self.info = self.info + '\nRegional y'
def regional_X(self, regions):
# this should only be used if X is still the original X
# add an error catcher here for if regional_X or mean_X etc.
# has been applied already- or if already split into training data
X_regional = split_into_regions(self.X, self.area_X,
self.lons, self.lats, self.lons1, self.lats1,
regions)
self.X = X_regional
self.X_type = 'Regional'
self.area_X = np.ones(len(regions))
self.info = self.info + '\nRegional X'