make kriging poly, corr and mfk rho_regr options an enumerate

smt/extensions/mfk.py

@@ -34,9 +34,9 @@ def _initialize(self):
  super(MFK, self)._initialize()
  declare = self.options.declare
 
- declare('rho_regr', 'constant',types=FunctionType,\
- values=('constant', 'linear', 'quadratic'), desc='regr. term')
- declare('theta0', None, types=(list, np.ndarray), \
+ declare('rho_regr', 'constant', types=str,\
+ values=('constant', 'linear', 'quadratic'), desc='Regression function type for rho')
+ declare('theta0', types=(list, np.ndarray), \
  desc='Initial hyperparameters')
  declare('optim_var', False, types = bool, \
  values = (True, False), \
@@ -145,12 +145,12 @@ def _new_train(self):
 
 
  # Regression matrix and parameters
- self.F_all[lvl] = self.options['poly'](self.X_norma)
+ self.F_all[lvl] = self._regression_types[self.options['poly']](self.X_norma)
  self.p_all[lvl] = self.F_all[lvl].shape[1]
 
  # Concatenate the autoregressive part for levels > 0
  if lvl > 0:
- F_rho = self.options['rho_regr'](self.X_norma)
+ F_rho = self._regression_types[self.options['rho_regr']](self.X_norma)
  self.q_all[lvl] = F_rho.shape[1]
  self.F_all[lvl] = np.hstack((F_rho*np.dot(self._predict_intermediate_values(self.X_norma, lvl, descale=False),
  np.ones((1,self.q_all[lvl]))), self.F_all[lvl]))
@@ -194,7 +194,7 @@ def _new_train(self):
  self._new_train()
 
  def _componentwise_distance(self,dx,opt=0):
- d = componentwise_distance(dx,self.options['corr'].__name__,
+ d = componentwise_distance(dx, self.options['corr'],
  self.nx)
  return d
 
@@ -225,8 +225,8 @@ def _predict_intermediate_values(self, X, lvl, descale = True):
  if descale :
  X = (X - self.X_mean) / self.X_std
 ## X = (X - self.X_mean[0]) / self.X_std[0]
- f = self.options['poly'](X)
- f0 = self.options['poly'](X)
+ f = self._regression_types[self.options['poly']](X)
+ f0 = self._regression_types[self.options['poly']](X)
  dx = manhattan_distances(X, Y=self.X_norma_all[0], sum_over_features=False)
  d = self._componentwise_distance(dx)
  # Get regression function and correlation
@@ -236,7 +236,7 @@ def _predict_intermediate_values(self, X, lvl, descale = True):
  beta = self.optimal_par[0]['beta']
  Ft = solve_triangular(C, F, lower=True)
  yt = solve_triangular(C, self.y_norma_all[0], lower=True)
- r_ = self.options['corr'](self.optimal_theta[0], d).reshape(n_eval, self.nt_all[0])
+ r_ = self._correlation_types[self.options['corr']](self.optimal_theta[0], d).reshape(n_eval, self.nt_all[0])
  gamma = self.optimal_par[0]['gamma']
 
 
@@ -247,10 +247,10 @@ def _predict_intermediate_values(self, X, lvl, descale = True):
  for i in range(1,lvl):
  F = self.F_all[i]
  C = self.optimal_par[i]['C']
- g = self.options['rho_regr'](X)
+ g = self._regression_types[self.options['rho_regr']](X)
  dx = manhattan_distances(X, Y=self.X_norma_all[i], sum_over_features=False)
  d = self._componentwise_distance(dx)
- r_ = self.options['corr'](self.optimal_theta[i], d).reshape(n_eval, self.nt_all[i])
+ r_ = self._correlation_types[self.options['corr']](self.optimal_theta[i], d).reshape(n_eval, self.nt_all[i])
  f = np.vstack((g.T*mu[:,i-1], f0.T))
  Ft = solve_triangular(C, F, lower=True)
  yt = solve_triangular(C, self.y_norma_all[i], lower=True)
@@ -339,7 +339,7 @@ def predict_variances_all_levels(self, X):
  beta = self.optimal_par[0]['beta']
  Ft = solve_triangular(C, F, lower=True)
  yt = solve_triangular(C, self.y_norma_all[0], lower=True)
- r_ = self.options['corr'](self.optimal_theta[0], d).reshape(n_eval, self.nt_all[0])
+ r_ = self._correlation_types[self.options['corr']](self.optimal_theta[0], d).reshape(n_eval, self.nt_all[0])
  gamma = self.optimal_par[0]['gamma']
 
  # Scaled predictor
@@ -360,7 +360,7 @@ def predict_variances_all_levels(self, X):
  for i in range(1,nlevel):
  F = self.F_all[i]
  C = self.optimal_par[i]['C']
- g = self.options['rho_regr'](X)
+ g = self._regression_types[self.options['rho_regr']](X)
  dx = manhattan_distances(X, Y=self.X_norma_all[i], sum_over_features=False)
  d = self._componentwise_distance(dx)
  r_ = self.options['corr'](self.optimal_theta[i], d).reshape(n_eval, self.nt_all[i])
@@ -424,28 +424,28 @@ def _predict_derivatives(self, x, kx):
 
  dy_dx = np.zeros((n_eval, lvl))
 
- if self.options['corr'].__name__ != 'squar_exp':
+ if self.options['corr'] != 'squar_exp':
  raise ValueError(
  'The derivative is only available for square exponential kernel')
- if self.options['poly'].__name__ == 'constant':
+ if self.options['poly'] == 'constant':
  df = np.zeros([n_eval,1])
- elif self.options['poly'].__name__ == 'linear':
+ elif self.options['poly'] == 'linear':
  df = np.zeros((n_eval, self.nx + 1))
  df[:,1:] = 1
  else:
  raise ValueError(
  'The derivative is only available for ordinary kriging or '+
  'universal kriging using a linear trend')
  df0 = copy.deepcopy(df)
- if self.options['rho_regr'].__name__ != 'constant' :
+ if self.options['rho_regr'] != 'constant' :
  raise ValueError(
  'The derivative is only available for regression rho constant')
  # Get pairwise componentwise L1-distances to the input training set
  dx = manhattan_distances(x, Y=self.X_norma_all[0], sum_over_features=
  False)
  d = self._componentwise_distance(dx)
  # Compute the correlation function
- r_ = self.options['corr'](self.optimal_theta[0], d).reshape(n_eval,self.nt_all[0])
+ r_ = self._correlation_types[self.options['corr']](self.optimal_theta[0], d).reshape(n_eval,self.nt_all[0])
 
  # Beta and gamma = R^-1(y-FBeta)
  beta = self.optimal_par[0]['beta']
@@ -463,10 +463,10 @@ def _predict_derivatives(self, x, kx):
  for i in range(1,lvl):
  F = self.F_all[i]
  C = self.optimal_par[i]['C']
- g = self.options['rho_regr'](x)
+ g = self._regression_types[self.options['rho_regr']](x)
  dx = manhattan_distances(x, Y=self.X_norma_all[i], sum_over_features=False)
  d = self._componentwise_distance(dx)
- r_ = self.options['corr'](self.optimal_theta[i], d).reshape(n_eval, self.nt_all[i])
+ r_ = self._correlation_types[self.options['corr']](self.optimal_theta[i], d).reshape(n_eval, self.nt_all[i])
  df = np.vstack((g.T*dy_dx[:,i-1], df0.T))
 
  Ft = solve_triangular(C, F, lower=True)

smt/surrogate_models/gekpls.py

@@ -46,6 +46,6 @@ def _compute_pls(self,X,y):
 
  def _componentwise_distance(self,dx,opt=0):
 
- d = componentwise_distance_PLS(dx,self.options['corr'].__name__,
+ d = componentwise_distance_PLS(dx,self.options['corr'],
  self.options['n_comp'],self.coeff_pls)
  return d

smt/surrogate_models/kpls.py

@@ -34,6 +34,6 @@ def _compute_pls(self,X,y):
  return X,y
 
  def _componentwise_distance(self,dx,opt=0):
- d = componentwise_distance_PLS(dx,self.options['corr'].__name__,
+ d = componentwise_distance_PLS(dx, self.options['corr'],
  self.options['n_comp'],self.coeff_pls)
  return d

smt/surrogate_models/kplsk.py

@@ -35,9 +35,9 @@ def _compute_pls(self,X,y):
  def _componentwise_distance(self,dx,opt=0):
  if opt == 0:
  # Kriging step
- d = componentwise_distance(dx,self.options['corr'].__name__,self.nx)
+ d = componentwise_distance(dx,self.options['corr'],self.nx)
  else:
  # KPLS step
- d = componentwise_distance_PLS(dx,self.options['corr'].__name__,
+ d = componentwise_distance_PLS(dx,self.options['corr'],
  self.options['n_comp'],self.coeff_pls)
  return d

smt/surrogate_models/krg.py

@@ -1,8 +1,8 @@
-'''
+"""
 Author: Dr. Mohamed A. Bouhlel <[email protected]>
 
 This package is distributed under New BSD license.
-'''
+"""
 
 from __future__ import division
 import warnings
@@ -15,13 +15,12 @@
 The kriging class.
 """
 
-class KRG(KrgBased):
 
+class KRG(KrgBased):
  def _initialize(self):
  super(KRG, self)._initialize()
- self.name = 'Kriging'
+ self.name = "Kriging"
 
- def _componentwise_distance(self,dx,opt=0):
- d = componentwise_distance(dx,self.options['corr'].__name__,
- self.nx)
+ def _componentwise_distance(self, dx, opt=0):
+ d = componentwise_distance(dx, self.options["corr"], self.nx)
  return d

smt/surrogate_models/krg_based.py

@@ -8,10 +8,6 @@
 TODO:
 - fail_iteration and nb_iter_max to remove from options
 - define outputs['sol'] = self.sol
-
-- debug _train: self_pkl = pickle.dumps(obj)
- cPickle.PicklingError: Can't pickle <type 'function'>: attribute lookup __builtin__.function failed
-
 """
 
 from __future__ import division
@@ -50,11 +46,11 @@ def _initialize(self):
  super(KrgBased, self)._initialize()
  declare = self.options.declare
  supports = self.supports
- declare('poly', 'constant',types=FunctionType,values=('constant', 'linear', 'quadratic'),
- desc='regr. term')
- declare('corr', 'squar_exp', types=FunctionType,values=('abs_exp', 'squar_exp'),
- desc='type of corr. func.')
- declare('data_dir', values=None, types=str,
+ declare('poly', 'constant', values=('constant', 'linear', 'quadratic'),
+ desc='Regression function type')
+ declare('corr', 'squar_exp', values=('abs_exp', 'squar_exp'),
+ desc='Correlation function type')
+ declare('data_dir', types=str,
  desc='Directory for loading / saving cached data; None means do not save or load')
  declare('theta0', [1e-2], types=(list, np.ndarray), desc='Initial hyperparameters')
  self.name = 'KrigingBased'
@@ -92,7 +88,7 @@ def _new_train(self):
  raise Exception("Multiple input features cannot have the same value.")
 
  # Regression matrix and parameters
- self.F = self.options['poly'](self.X_norma)
+ self.F = self._regression_types[self.options['poly']](self.X_norma)
  n_samples_F = self.F.shape[0]
  if self.F.ndim > 1:
  p = self.F.shape[1]
@@ -112,16 +108,13 @@ def _train(self):
  """
  Train the model
  """
- """
  inputs = {'self': self}
  with cached_operation(inputs, self.options['data_dir']) as outputs:
  if outputs:
  self.sol = outputs['sol']
  else:
  self._new_train()
  #outputs['sol'] = self.sol
- """
- self._new_train()
 
  def _reduced_likelihood_function(self, theta):
 
@@ -183,7 +176,7 @@ def _reduced_likelihood_function(self, theta):
  theta = tmp_var[:-1]
  noise = tmp_var[-1]
 
- r = self.options['corr'](theta, self.D).reshape(-1,1)
+ r = self._correlation_types[self.options['corr']](theta, self.D).reshape(-1,1)
 
  R = np.eye(self.nt) * (1. + nugget+ noise)
  R[self.ij[:, 0], self.ij[:, 1]] = r[:,0]
@@ -276,10 +269,10 @@ def _predict_values(self, x):
  False)
  d = self._componentwise_distance(dx)
  # Compute the correlation function
- r = self.options['corr'](self.optimal_theta, d).reshape(n_eval,self.nt)
+ r = self._correlation_types[self.options['corr']](self.optimal_theta, d).reshape(n_eval,self.nt)
  y = np.zeros(n_eval)
  # Compute the regression function
- f = self.options['poly'](x)
+ f = self._regression_types[self.options['poly']](x)
  # Scaled predictor
  y_ = np.dot(f, self.optimal_par['beta']) + np.dot(r,
  self.optimal_par['gamma'])
@@ -314,14 +307,14 @@ def _predict_derivatives(self, x, kx):
  False)
  d = self._componentwise_distance(dx)
  # Compute the correlation function
- r = self.options['corr'](self.optimal_theta, d).reshape(n_eval,self.nt)
+ r = self._correlation_types[self.options['corr']](self.optimal_theta, d).reshape(n_eval,self.nt)
 
- if self.options['corr'].__name__ != 'squar_exp':
+ if self.options['corr'] != 'squar_exp':
  raise ValueError(
  'The derivative is only available for square exponential kernel')
- if self.options['poly'].__name__ == 'constant':
+ if self.options['poly'] == 'constant':
  df = np.array([0])
- elif self.options['poly'].__name__ == 'linear':
+ elif self.options['poly'] == 'linear':
  df = np.zeros((self.nx + 1, self.nx))
  df[1:,:] = 1
  else:
@@ -493,7 +486,7 @@ def minus_reduced_likelihood_function(log10t):
  if exit_function:
  return best_optimal_rlf_value, best_optimal_par, best_optimal_theta
 
- if self.options['corr'].__name__ == 'squar_exp':
+ if self.options['corr'] == 'squar_exp':
  self.options['theta0'] = (best_optimal_theta*self.coeff_pls**2).sum(1)
  else:
  self.options['theta0'] = (best_optimal_theta*np.abs(self.coeff_pls)).sum(1)
@@ -508,24 +501,7 @@ def _check_param(self):
  """
  This function check some parameters of the model.
  """
- # Check regression model
- if not callable(self.options['poly']):
- if self.options['poly'] in self._regression_types:
- self.options['poly'] = self._regression_types[
- self.options['poly']]
- else:
- raise ValueError("regr should be one of %s or callable, "
- "%s was given." % (self._regression_types.keys(),
- self.options['poly']))
- if self.name == 'MFK' and not callable(self.options['rho_regr']):
- if self.options['rho_regr'] in self._regression_types:
- self.options['rho_regr'] = self._regression_types[
- self.options['rho_regr']]
- else:
- raise ValueError("rho_regr should be one of %s or callable, "
- "%s was given." % (self._regression_types.keys(),
- self.options['rho_regr']))
-
+ # FIXME: _check_param should be overriden in corresponding subclasses
  if self.name in ['KPLS', 'KPLSK', 'GEKPLS']:
  d = self.options['n_comp']
  else:
@@ -538,14 +514,6 @@ def _check_param(self):
  raise ValueError('the number of dim %s should be equal to the length of theta0 %s.' % 
  (d, len(self.options['theta0'])))
 
- if not callable(self.options['corr']):
- if self.options['corr'] in self._correlation_types:
- self.options['corr'] = self._correlation_types[self.options['corr']]
- else:
- raise ValueError("corr should be one of %s or callable, "
- "%s was given."% (self._correlation_types.keys(),
- self.options['corr']))
-
  if self.supports['training_derivatives']:
  if not(1 in self.training_points[None]):
  raise Exception('Derivative values are needed for using the GEKPLS model.')