def _preProcessingData(self, x, y):
for i in range(len(y.data)):
for k in range(len(x.data[0])):
if (type(x.data[i][k]) is not Rdm):
x.data[i][k] = Rdmia.number(x.data[i][k])
if (type(y.data[i]) is not Rdm):
y.data[i][0] = Rdmia.number(y.data[i][0])
def rosenbrock(x): # rosen.m
""" http://en.wikipedia.org/wiki/Rosenbrock_function """
# a sum of squares, so LevMar (scipy.optimize.leastsq) is pretty good
x = np.asarray_chkfinite(x)
x0 = x[:-1]
x1 = x[1:]
return (sum((1.0 - x0)**2.0) + rdmia.number(100.0) * sum(
(x1 - x0**2.0)**2.0))
def makeList(steps, dim):
result = []
r = []
for k in steps:
for i in range(dim):
r.append(rdmia.number(k))
result.append(r)
r = []
return result
def predict(self, x):
r_list = []
if (self._isMult):
for val in range(len(x)):
resultLow = sum([
self._predictorsLow.data[i] * x[val][i].lower()
for i in range(len(self._predictorsLow.data))
])
resultUp = sum([
self._predictorsUp.data[i] * x[val][i].upper()
for i in range(len(self._predictorsUp.data))
])
r_list.append(Rdmia.number(resultLow, resultUp))
else:
for val in range(len(x)):
result = sum([
self._predictors.data[i] * x[val][i]
for i in range(len(self._predictors.data))
])
r_list.append(result)
#r = qm.midpoint(self._predictors.data[0] + self._predictors.data[1]*x.data[val][1])
return r_list
def fit(self, x_data, y_data, kernel='std', alpha=0.1):
#x_1 = (inv_by_adjugate(midPointMatrix(x.T.dot(x))))
#self._preProcessingData(x,y)
#First part of method minimum square
if (kernel == 'std'):
x = Rdmia.array(x_data)
y = Rdmia.array(y_data)
x_1 = x.T.dot(x).midPointMatrix
x_1 = x_1.I
#y_2 = midPointMatrix(x.T.dot(y))
y_2 = x.T.dot(y).midPointMatrix
#Calculated predictors beta
z = (x_1).dot(y_2)
self._predictors = Rdmia.array(
[z.data[i][0] for i in range(len(z.data))])
elif (kernel == 'multidimensional'):
self._isMult = True
opt_minium = 0.0
opt_maximum = 0.0
for i in np.arange(0.0, 1.0, alpha):
x = Rdmia.arrayAlpha(x_data, i)
y_min, y_max = Rdmia.arrayMinMax(y_data)
x_1 = x.T.dot(x)
x_1 = x_1.I
#y_2 = midPointMatrix(x.T.dot(y))
y_2_min = x.T.dot(y_min)
y_2_max = x.T.dot(y_max)
#Calculated predictors beta
z_min = (x_1).dot(y_2_min)
z_max = (x_1).dot(y_2_max)
#print(z.data)
#self._predictorsLow = (Rdmia.array([z_min.data[i][0] for i in range(len(z_min.data))]))
#self._predictorsUp = (Rdmia.array([z_max.data[i][0] for i in range(len(z_max.data))]))
opt_min, opt_max = self._optimizations(x_data, (Rdmia.array(
[z_min.data[i][0]
for i in range(len(z_min.data))])), (Rdmia.array(
[z_max.data[i][0] for i in range(len(z_max.data))])),
y_min, y_max)
if (i == 0):
self._predictorsLow = (Rdmia.array(
[z_min.data[i][0] for i in range(len(z_min.data))]))
self._predictorsUp = (Rdmia.array(
[z_max.data[i][0] for i in range(len(z_max.data))]))
opt_minimum = opt_min
opt_maximum = opt_max
elif (opt_min < opt_minium):
self._predictorsLow = (Rdmia.array(
[z_min.data[i][0] for i in range(len(z_min.data))]))
opt_minimum = opt_min
elif (opt_max < opt_maximum):
self._predictorsUp = (Rdmia.array(
[z_max.data[i][0] for i in range(len(z_max.data))]))
opt_maximum = opt_max
#Calculte Status
y_pred = self.predict(x_data)
self._coefDetermination = self._coefDetermination(y_data, y_pred)
self._rmse = self._RMSE(y_data, y_pred)
print("### STATISTICS OF MODEL ###")
print("# RMSE: ", self._rmse)
print("# Coefficiente of Determination: ", self._coefDetermination)
print("#")
def _mean(self, d):
mean = Rdmia.number(0.0)
for val in d:
mean += val[0]
return mean / len(d)
def _RMSE(self, y, y_pred):
lower = 0.0
upper = 0.0
tot = 0.0
for val in range(len(y)):
lower += (y[val][0].lower() - y_pred[val].lower())
upper += (y[val][0].upper() - y_pred[val].upper())
lower = (lower**2.0 / len(y))**(1.0 / 2.0)
upper = (upper**2.0 / len(y))**(1.0 / 2.0)
return lower, upper
if __name__ == "__main__":
Rdmia.setDotPrecision(2)
'''
x = Rdmia.array([[Rdmia.number(1),Rdmia.number(90,100)],
[Rdmia.number(1),Rdmia.number(90,130)],
[Rdmia.number(1),Rdmia.number(140,180)],
[Rdmia.number(1),Rdmia.number(110,142)],
[Rdmia.number(1),Rdmia.number(90,100)],
[Rdmia.number(1),Rdmia.number(130,160)],
[Rdmia.number(1),Rdmia.number(60,100)],
[Rdmia.number(1),Rdmia.number(130,160)],
[Rdmia.number(1),Rdmia.number(110,190)],
[Rdmia.number(1),Rdmia.number(138,180)],
[Rdmia.number(1),Rdmia.number(110,150)]])
x = [[Rdmia.number(1),Rdmia.number(90,100)],
r = []
return result
#-------------------------------------------------------------------------------
#-------------------------------------------------------------------------------
if __name__ == "__main__": # standalone test --
import sys
import json
dims = [2, 3] #dimensions
nstep = 50 #iterations
seed = 1
init = 1
rdmia.setDotPrecision(0.1)
print("PRECISION: ", rdmia.precision())
problems = {}
# to change these params in sh or ipython, run this.py a=1 b=None c=[3] ...
for arg in sys.argv[1:]:
exec(arg)
np.set_printoptions(
threshold=20,
edgeitems=5,
linewidth=120,
suppress=True,
formatter=dict(float=lambda x: "%.2g" % x)) # float arrays %.2g
np.random.seed(seed)