def adapt_filter(self, obs_mat):
'''
Adaptive filter by using a fixed observation matrix and dynamic transition
matrix
Parameters:
----
obs_mat :np.array()
the fixed observation matrix
Return:
----
the filtered result in dataframe
'''
from filterpy.kalman import KalmanFilter
measure_dat = self.data['2017-05-01 01:00:00':].iloc[
self.train_size:, :]
#seperating lists
train_lst = DataImporter.sepDat2(self.train_window, self.predict_size,
measure_dat)
measure_lst = DataImporter.sepDat(
self.predict_size, measure_dat.iloc[self.train_window:, :])
#estimating coefficients
import warnings
warnings.filterwarnings("ignore")
param_lst = [
VAR(i).fit(1, trend='nc').params.values for i in train_lst
]
#calculating the kf results
res_lst = []
#initialize filter params
kf = KalmanFilter(dim_x=self.dim, dim_z=self.dim)
kf.x = self.init_val
kf.P = self.init_prob
kf.R = self.R
kf.Q = self.Q
kf.H = obs_mat
#filtering
for i in range(len(measure_lst)):
#changing params of transition mat
kf.F = param_lst[i].T
#doing one batch
x, p, _, _ = kf.batch_filter(measure_lst[i].values)
res = pd.DataFrame(x)
res.columns = self.data.columns
res.index = measure_lst[i].index
res_lst.append(res)
return KFfilter.merge(pd.concat(res_lst), measure_dat)
