def TrainTestFull(X, y, Names, TestName, lons, lats, lons1, lats1, area_flat,
plot_dir):
(X_train, X_test, y_train, y_test, names_train,
names_test) = split_set(X, y, Names, TestName)
# Train
(m, kern) = TrainModel(X_train, y_train)
# Predict
y_pred, cov = TestModel(m, X_test, y_test)
# Plot
PlotMap(y_test, y_pred, lons, lats, names_train, names_test, plot_dir,
area_flat)
regions_all = ['Global'] + list(RegionsList)
print(y_pred.shape)
for v in range(len(TestName)):
Validation(y_test[v:v + 1, ],
y_pred[v:v + 1, :],
cov[v:v + 1, :],
lons,
lats,
lons1,
lats1,
names_train,
names_test,
plot_dir + TestName[v] + '_',
area_flat,
metric_regions=regions_all)
Save(names_train, names_test, X_train, X_test, y_train, y_test, y_pred,
lons, lats, lons1, lats1, area_flat, m, kern,
plot_dir + TestName[0] + '_')
def TrainTestPCA(X,
y,
Names,
TestName,
lons,
lats,
lons1,
lats1,
area_flat,
plot_dir,
n_comp=None,
pca_inputs=True,
pca_outputs=True):
(X_train, X_test, y_train, y_test, names_train,
names_test) = split_set(X, y, Names, TestName)
# PCA
if pca_inputs is True:
X_PCA = PCA(n_comp).fit(X_train)
X_trans = X_PCA.transform(X_train)
X_test_trans = X_PCA.transform(X_test)
else:
X_trans = X_train
X_test_trans = X_test
if pca_outputs is True:
y_PCA = PCA(n_comp).fit(y_train)
y_trans = y_PCA.transform(y_train)
y_test_trans = y_PCA.transform(y_test)
else:
y_trans = y_train
y_test_trans = y_test
# Train
(m, kern) = TrainModel(X_trans, y_trans)
y_pred, cov = TestModel(m, X_test_trans, y_test_trans)
# Inverse PC
if pca_outputs is True:
y_pred_full = y_PCA.inverse_transform(y_pred)
cov = y_PCA.inverse_transform(cov)
else:
y_pred_full = y_pred
# Plot
regions_all = ['Global'] + list(RegionsList)
PlotMap(y_test, y_pred_full, lons, lats, names_train, names_test, plot_dir,
area_flat)
Validation(y_test,
y_pred_full,
cov,
lons,
lats,
lons1,
lats1,
names_train,
names_test,
plot_dir + TestName[0] + '_',
area_flat,
metric_regions=regions_all)
Save(names_train, names_test, X_train, X_test, y_train, y_test,
y_pred_full, lons, lats, lons1, lats1, area_flat, m, kern,
plot_dir + TestName[0] + '_')
def TrainTestScale(X, y, Names, TestName, lons, lats, lons1, lats1, area_flat,
plot_dir):
(X_train, X_test, y_train, y_test, names_train,
names_test) = split_set(X, y, Names, TestName)
# Scale
scaler_x = StandardScaler()
scaler_x.fit(X_train)
X_train_norm = scaler_x.transform(X_train)
X_test_norm = scaler_x.transform(X_test)
scaler_y = StandardScaler()
scaler_y.fit(y_train)
y_train_norm = scaler_y.transform(y_train)
y_test_norm = scaler_y.transform(y_test)
# Train
(m, kern) = TrainModel(X_train_norm, y_train_norm)
# Predict
y_pred_norm, cov = TestModel(m, X_test_norm, y_test_norm)
y_pred = scaler_y.inverse_transform(y_pred_norm)
# Plot
PlotMap(y_test, y_pred, lons, lats, names_train, names_test, plot_dir,
area_flat)
regions_all = ['Global'] + list(RegionsList)
print(y_pred.shape)
for v in range(len(TestName)):
Validation(y_test[v:v + 1, ],
y_pred[v:v + 1, :],
cov[v:v + 1, :],
lons,
lats,
lons1,
lats1,
names_train,
names_test,
plot_dir + TestName[v] + '_',
area_flat,
metric_regions=regions_all)
Save(names_train, names_test, X_train, X_test, y_train, y_test, y_pred,
lons, lats, lons1, lats1, area_flat, m, kern,
plot_dir + TestName[0] + '_')
area_flat = area.flatten()
print(area_flat.shape,X_SfcTemp.shape)
regions = ['Europe','Africa','US','South_America','East_Asia','India']
# Save to
plot_dir = '/work/lm2612/GPPCA/'
# Test data set
print(Names)
for i in range(len(Names)):
print("Training Stage:")
TestName = Names[i]
(X_train,X_test,y_train,y_test,names_train,names_test) = split_set(X,y,Names,TestName)
# PCA
X_PCA = PCA().fit(X_train)
y_PCA = PCA().fit(y_train)
X_trans = X_PCA.transform(X_train)
y_trans = y_PCA.transform(y_train)
# GP Model
(N,p) = X_trans.shape
kern = GPy.kern.Matern52(p,ARD=True)+GPy.kern.Linear(p,ARD=True)
print("kernel used: {}".format(kern))
m = GPy.models.GPRegression(X_trans,y_trans,kern)
m.optimize()
def TrainTestRegional(X,
y,
Names,
TestName,
lons,
lats,
lons1,
lats1,
area_flat,
plot_dir,
regions=RegionOnlyList,
reg_inputs=True,
reg_outputs=True):
(X_train, X_test, y_train, y_test, names_train,
names_test) = split_set(X, y, Names, TestName)
# PCA
if reg_inputs is True:
print(regions)
X_trans = split_into_regions(X_train,
Area(lons1, lats1).flatten(), lons, lats,
lons1, lats1, regions)
X_test_trans = split_into_regions(X_test,
Area(lons1, lats1).flatten(), lons,
lats, lons1, lats1, regions)
else:
X_trans = X_train
X_test_trans = X_test
if reg_outputs is True:
print(regions)
y_trans = split_into_regions(y_train,
Area(lons1, lats1).flatten(), lons, lats,
lons1, lats1, regions)
y_test_trans = split_into_regions(y_test,
Area(lons1, lats1).flatten(), lons,
lats, lons1, lats1, regions)
else:
y_trans = y_train
y_test_trans = y_test
# Train
(m, kern) = TrainModel(X_trans, y_trans)
y_pred, cov = TestModel(m, X_test_trans, y_test_trans)
# Inverse PC
if reg_outputs is True:
areas = np.ones(len(regions))
RegionalMetrics(y_pred, y_test_trans, regions, lons, lats, lons1,
lats1, names_train, names_test,
plot_dir + TestName[0] + '_', areas)
GPError(cov, y_pred, y_test_trans, regions, lons, lats, lons1, lats1,
names_train, names_test, plot_dir + TestName[0] + '_', areas)
else:
y_pred_full = y_pred
# Plot
regions_all = ['Global'] + list(RegionsList)
print(TestName[0])
PlotMap(y_test, y_pred_full, lons, lats, names_train, names_test,
plot_dir, area_flat)
Validation(y_test,
y_pred_full,
cov,
lons,
lats,
lons1,
lats1,
names_train,
names_test,
plot_dir + TestName[0] + '_',
area_flat,
metric_regions=regions_all)
Save(names_train, names_test, X_train, X_test, y_train, y_test, y_pred,
lons, lats, lons1, lats1, area_flat, m, kern,
plot_dir + TestName[0] + '_')