class EWACalibratedForecaster(Forecaster):
"""Produces calibrated estimates by minimizng internal regret"""
def __init__(self, N, eta=None):
super(EWACalibratedForecaster, self).__init__(N+1)
self.F_int = EWAInternalForecaster(N+1, eta)
self.e = np.array([float(i)/N for i in range(N+1)])
self.examples_seen = 0
def predict(self):
w = self.F_int.weights.reshape(self.N,)
ind = np.random.multinomial(1, w)
return ind.dot(self.e)
def expected_prediction(self):
w = self.F_int.weights.reshape(self.N,)
return w.T.dot(self.e)
def observe(self, y_t):
l = np.array([(y_t - e)**2 for e in self.e])
self.F_int.observe(l)
self.examples_seen += 1
@property
def weights(self):
return self.F_int.weights
class EWACalibratedForecaster(Forecaster):
"""Produces calibrated estimates by minimizng internal regret"""
def __init__(self, N, eta=None):
super(EWACalibratedForecaster, self).__init__(N+1)
self.F_int = EWAInternalForecaster(N+1, eta)
self.e = np.array([float(i)/N for i in xrange(N+1)])
self.examples_seen = 0
def predict(self):
w = self.F_int.weights.reshape(self.N,)
ind = np.random.multinomial(1, w)
return ind.dot(self.e)
def expected_prediction(self):
w = self.F_int.weights.reshape(self.N,)
return w.T.dot(self.e)
def observe(self, y_t):
l = np.array([(y_t - e)**2 for e in self.e])
self.F_int.observe(l)
self.examples_seen += 1
@property
def weights(self):
return self.F_int.weights
def __init__(self, N, eta=None): super(EWACalibratedForecaster, self).__init__(N+1) self.F_int = EWAInternalForecaster(N+1, eta) self.e = np.array([float(i)/N for i in range(N+1)]) self.examples_seen = 0
def __init__(self, N, eta=None): super(EWACalibratedForecaster, self).__init__(N+1) self.F_int = EWAInternalForecaster(N+1, eta) self.e = np.array([float(i)/N for i in xrange(N+1)]) self.examples_seen = 0
