def get_sensor():
row = str(request.args.get('row'))
pos = str(request.args.get('pos'))
resp = dynamo_client.get_item(
TableName='SoilSensor',
Key={
'Row': {'S': row},
'PositionInRow': {'S': pos}
}
)
soil_data = np.array([float(i['N']) for i in resp['Item']['payload']['M']['moistureSensor']['L']]).reshape(-1, 1)
light_data = np.array([float(i['N']) for i in resp['Item']['payload']['M']['lightSensor']['L']]).reshape(-1, 1)
x = np.arange(len(soil_data)).reshape(-1, 1)
predictive_range = np.arange(len(soil_data), len(soil_data) + 72).reshape(-1, 1)
soil_kernel = kernels.ExpSineSquared()
soil_gp = GaussianProcessRegressor(kernel=soil_kernel)
soil_gp.fit(x, soil_data)
print(f"Soil MLL: {soil_gp.log_marginal_likelihood_value_}")
predicted_soil = soil_gp.predict(predictive_range)
light_kernel = kernels.ExpSineSquared()
light_gp = GaussianProcessRegressor(kernel=light_kernel)
light_gp.fit(x, light_data)
print(f"Light MLL: {light_gp.log_marginal_likelihood_value_}")
predicted_light = light_gp.predict(predictive_range)
resp = {
"soil_observed": soil_data.flatten().tolist(),
"soil_predicted": predicted_soil.flatten().tolist(),
"light_observed": light_data.flatten().tolist(),
"light_predicted": predicted_light.flatten().tolist()
}
return json.dumps(resp, indent=4)
def make_exp_sine_squared(n_dim_obs=5, n_dim_lat=0, T=1, **kwargs):
"""Make precision matrices using a temporal sine kernel."""
from regain.bayesian.gaussian_process_ import sample as samplegp
from sklearn.gaussian_process import kernels
L, K_HO = make_ell(n_dim_obs, n_dim_lat)
periodicity = kwargs.get("periodicity", np.pi)
length_scale = kwargs.get("length_scale", 2)
epsilon = kwargs.get("epsilon", 0.5)
sparse = kwargs.get("sparse", True)
temporal_kernel = kernels.ExpSineSquared(periodicity=periodicity,
length_scale=length_scale)(
np.arange(T)[:, None])
u = samplegp(temporal_kernel, p=n_dim_obs * (n_dim_obs - 1) // 2)[0]
K, K_obs = [], []
for uu in u.T:
theta = squareform(uu)
if sparse:
# sparsify
theta[np.abs(theta) < epsilon] = 0
theta += np.diag(np.sum(np.abs(theta), axis=1) + 0.01)
K.append(theta)
assert is_pos_def(theta)
theta_observed = theta - L
assert is_pos_def(theta_observed)
K_obs.append(theta_observed)
thetas = np.array(K)
return thetas, np.array(K_obs), np.array([L] * T)
def main():
# Specify type_ and region of data
type_ = 'organic'
region = 'WestTexNewMexico'
# Specify the kernel functions; please see the paper for the rationale behind the choices
kernel = Kernels.ExpSineSquared(length_scale=20., periodicity=365.) \
+ 0.8 * Kernels.RationalQuadratic(alpha=20., length_scale=80.) \
+ Kernels.WhiteKernel(.2)
# Fit gp model and plot
run_gp(kernel, n_restarts_optimizer=10, type_=type_, region=region)
'k_Constant':
lambda: kers.ConstantKernel(),
'k_WN':
lambda: kers.WhiteKernel(),
'k_RBF':
lambda: kers.RBF(),
'k_RQ':
lambda: kers.RationalQuadratic(),
'k_mat0':
lambda: kers.Matern(nu=0.5),
'k_mat1':
lambda: kers.Matern(nu=1.5),
'k_mat2':
lambda: kers.Matern(nu=2.5),
'k_sine':
lambda: kers.ExpSineSquared(),
#now combination kernels
'k1':
lambda: kers.Sum(kers.ConstantKernel(), kers.ExpSineSquared()),
'k2':
lambda: kers.Product(kers.ConstantKernel(), kers.ExpSineSquared()),
'k3':
lambda: kers.Sum(
kers.ConstantKernel(),
kers.Product(kers.ConstantKernel(), kers.ExpSineSquared())),
'k4':
lambda: kers.Sum(kers.ConstantKernel(),
kers.Product(kers.RBF(), kers.ExpSineSquared())),
'k5':
lambda: kers.Sum(
kers.Product(kers.ConstantKernel(), kers.RBF()),
def _periodic_kernel(X, Y=None, inverse_width=1):
return kernels.ExpSineSquared(length_scale=inverse_width)(X, Y=Y)