def init_ss_learner(self, learner=None, params={'alpha':0.1}):
###############################################################
# Setup semi-supervised co-cluster learning models
###############################################################
self.ss_models = np.empty((2,), dtype = 'object') # For Dyadic data
self.cumsum_obj = np.zeros((2,))
self.obj_ss = [np.empty((self.M,self.K)),np.empty((self.N,self.L))]
for i in range(2):
if type(params) is dict:
self.ss_models[i] = learner_class(learner, params)
else:
self.ss_models[i] = learner_class(learner, params[i])
def train_process(iterable_data):
#import sys
#sys.path.insert(0, '/home/neeraj/.eclipse/org.eclipse.platform_3.8_155965261/plugins/org.python.pydev_2.7.1.2012100913/pysrc/')
#import pydevd;pydevd.settrace()
(rc_list,R,C,I,J,rowAttr,colAttr,Z,learner_params,learner,train_loss_type,num_observations,M,N) = iterable_data
#pydevd.settrace()
# Gives an error for Logistic because in the current version of sklearn coef_ cannot be written to
# Initialize the model
model = learner_class(learner, learner_params)
print 'learning model'
#model.model.coef_ = models_coefs[0]
#model.model.intercept_ = models_coefs[1]
# Filter out irrelevant rows
r = rc_list[0]
c = rc_list[1]
row_index = R[I]==r
filtered_I = I[row_index]
filtered_J = J[row_index]
filtered_Z = Z[row_index]
# Filter out irrelevant cols
col_index = C[filtered_J]==c
filtered_I = filtered_I[col_index]
filtered_Z = filtered_Z[col_index]
filtered_J = filtered_J[col_index]
if len(filtered_Z)>0:
# Build the current covariates matrix
# Extract and Normalize covariates
flat_index = (sp.coo_matrix((np.arange(num_observations), (I, J)), shape=(M,N))).tocsr()
flat_index = np.array(flat_index[filtered_I, filtered_J], copy=False).ravel()
covariates = np.hstack((rowAttr[filtered_I], colAttr[filtered_J]))
# call learner
model.fit(covariates, filtered_Z)
if train_loss_type == 'negloglik':
total_error = train_loss_dict[train_loss_type](filtered_Z, model.model.predict_proba(covariates)).sum()
else:
total_error = train_loss_dict[train_loss_type](filtered_Z, model.model.predict(covariates)).sum()
return total_error,r,c,[model.model.coef_,model.model.intercept_]
def init_learner(self, learner=None, params={'alpha':0.1}, train_loss=None, test_loss=None):
###############################################################
# Setup Output learning models
###############################################################
if learner is None: # uses scikits.learn internal bias
if self.D>0:
learner = "ridge"
else:
learner = "mean"
self.models = np.empty((self.K, self.L), dtype = 'object')
for r in range(self.K):
for c in range(self.L):
self.models[r,c] = learner_class(learner, params)
if train_loss is None:
train_loss = "sq_err"
if test_loss is None:
test_loss = "mse"
self.train_loss_type = train_loss
self.train_loss = train_loss_dict[train_loss]
self.test_loss = test_loss_dict[test_loss]
self.learner = learner
self.learner_params = params
