def regression_kriging_evaluation(year, features_=['all']):
from pykrige.rk import RegressionKriging
from sklearn.model_selection import LeaveOneOut
from sklearn.svm import SVR
from sklearn.linear_model import LinearRegression
from sklearn.ensemble import RandomForestRegressor
import warnings
warnings.filterwarnings("ignore", category=DeprecationWarning)
svr_model = SVR(kernel='rbf', C=10, gamma=0.001)
lr_model = LinearRegression()
rf_model = RandomForestRegressor(n_estimators=2)
Y, D, P, Tf, Gd = extract_raw_samples(year, crime_t=['total'])
coords = get_centroid_ca()
errors = []
for k in range(77):
X_train, X_test, Y_train, Y_test = build_features(Y, D, P, Tf, Y, Gd, Y, k, features=features_, taxi_norm="bydestination")
if k == 0:
print X_train.shape
coords_train = np.delete(coords, k, axis=0)
coords_test = np.array(coords)[k,None]
m_rk = RegressionKriging(regression_model=rf_model)
m_rk.fit(X_train, coords_train, Y_train)
z = m_rk.predict(X_test, coords_test)
errors.append(abs(Y_test - z[0]))
print np.mean(errors), np.mean(errors)/np.mean(Y)
return errors
# print("[DEBUG] p_test[:10] : {}".format(p_test[:10]))
# print("[DEBUG] x_train[:10] : {}".format(x_train[:10]))
# print("[DEBUG] x_test[:10] : {}".format(x_test[:10]))
# print("[DEBUG] target_train[:10] : {}".format(target_train[:10]))
# print("[DEBUG] target_test[:10] : {}".format(target_test[:10]))
# assert False
for m in models:
print('=' * 40)
print('regression model:', m.__class__.__name__)
m_rk = RegressionKriging(regression_model=m, n_closest_points=10)
m_rk.fit(p_train, x_train, target_train)
print('Regression Score: ',
m_rk.regression_model.score(p_test, target_test))
print('RK score: ', m_rk.score(p_test, x_test, target_test))
print("predict: {}".format(m_rk.predict(p_test, x_test)))
print("truth value : {}".format(target_test))
# fit(self, p, x, y)
# | fit the regression method and also Krige the residual
# |
# | Parameters
# | ----------
# | p: ndarray
# | (Ns, d) array of predictor variables (Ns samples, d dimensions)
# | for regression
# | x: ndarray
# | ndarray of (x, y) points. Needs to be a (Ns, 2) array
# | corresponding to the lon/lat, for example 2d regression kriging.
# | array of Points, (x, y, z) pairs of shape (N, 3) for 3d kriging
# | y: ndarray
x_y = x_grd[:, :-1]
##numpy array of surface elevation
z = x_grd[:, -1:]
#%%
#constract a randomforest model
rf_model = RandomForestRegressor(n_estimators=100)
#fit the regression kriging model
m_rk = RegressionKriging(regression_model=rf_model,
n_closest_points=2,
method='ordinary',
variogram_model='spherical')
#%%
m_rk.fit(p, x, sub1.iloc[:, 39])
pred_sph = m_rk.predict(z, x_y)
#%%
sph = pred_sph
#%%
#filter values
sph[sph < 0.3] = 0
sph.resize(src.height, src.width)
show(sph)
#%%
with rio.Env():
# Write an array as a raster band to a new 8-bit file. For
# the new file's profile, we start with the profile of the source
profile = src.profile