def backstepwise(data, out, k):
feature_num = len(data[0])
index_array = [i for i in range(feature_num)]
droper = feature_num - k
for i in range(droper):
beta = leastsquare.ls(data[:,np.array(index_array)], out)
yhat = leastsquare.predict(data[:,np.array(index_array)], beta)
theg2 = leastsquare.thegama2(yhat, out, len(index_array))
z = leastsquare.z_score(beta, np.sqrt(theg2), data[:,np.array(index_array)])
min_z = 999999999.
min_idx = None
for j in range(len(z)):
if z[j] < min_z:
min_z = z[j]
min_idx = j
index_array.remove(index_array[min_idx])
beta = leastsquare.ls(data[:,np.array(index_array)], out)
return beta, index_array
def RSS(X, Y, beta):
yhat = leastsquare.predict(X, beta)
return np.sum((yhat-Y)**2)
