def conditional_sample(self, data):
assert isinstance(data, XSeqDataSet)
sample = np.zeros(data.arrays['X'].shape)
input_indices = list(range(self.output_index)) + \
list(range(self.output_index + 1, data.arrays['X'].shape[1], 1))
inputs = data.variable_subsets(input_indices)
outputs = data.variable_subsets([self.output_index])
X_data = inputs
XY_data = XYDataSet()
XY_data.labels['X'] = inputs.labels['X']
XY_data.labels['Y'] = outputs.labels['X']
XY_data.arrays['X'] = inputs.arrays['X']
XY_data.arrays['Y'] = outputs.arrays['X']
X_sample = self.input_distribution.conditional_sample(X_data)
XY_data.arrays['X'] = X_sample
Y_sample = self.output_distribution.conditional_samples(XY_data)
sample[:, input_indices] = X_sample
sample[:, self.output_index] = Y_sample
return sample
def llh(self, data):
assert isinstance(data, XSeqDataSet)
sample = np.zeros(data.arrays['X'].shape)
input_indices = list(range(self.output_index)) + \
list(range(self.output_index + 1, data.arrays['X'].shape[1], 1))
inputs = data.variable_subsets(input_indices)
outputs = data.variable_subsets([self.output_index])
X_data = inputs
XY_data = XYDataSet()
XY_data.labels['X'] = inputs.labels['X']
XY_data.labels['Y'] = outputs.labels['X']
XY_data.arrays['X'] = inputs.arrays['X']
XY_data.arrays['Y'] = outputs.arrays['X']
llh = self.input_distribution.llh(X_data)
llh += self.output_distribution.llh(XY_data)
return llh
def conditional_sample(self, data):
assert isinstance(data, XSeqDataSet)
sample = np.zeros(data.arrays['X'].shape)
input_indices = list(range(self.output_index)) + \
list(range(self.output_index + 1, data.arrays['X'].shape[1], 1))
inputs = data.variable_subsets(input_indices)
outputs = data.variable_subsets([self.output_index])
X_data = inputs
XY_data = XYDataSet()
XY_data.labels['X'] = inputs.labels['X']
XY_data.labels['Y'] = outputs.labels['X']
XY_data.arrays['X'] = inputs.arrays['X']
XY_data.arrays['Y'] = outputs.arrays['X']
X_sample = self.input_distribution.conditional_sample(X_data)
XY_data.arrays['X'] = X_sample
Y_sample = self.output_distribution.conditional_samples(XY_data)
sample[:, input_indices] = X_sample
sample[:, self.output_index] = Y_sample
return sample
def fit(self):
best_llh = -np.Inf
best_distribution = None
# best_index = None
for input_var in range(self.data.arrays['X'].shape[1]):
inputs = self.data.variable_subsets(list(range(input_var)) +
list(range(input_var + 1, self.data.arrays['X'].shape[1], 1)))
outputs = self.data.variable_subsets([input_var])
input_agent = self.input_learner()
input_agent.load_data(inputs)
input_agent.fit()
XY_data = XYDataSet()
XY_data.labels['X'] = inputs.labels['X']
XY_data.labels['Y'] = outputs.labels['X']
XY_data.arrays['X'] = inputs.arrays['X']
XY_data.arrays['Y'] = outputs.arrays['X']
output_agent = self.output_learner()
output_agent.load_data(XY_data)
output_agent.fit()
conditional_distribution = RegressionDAG(input_var, input_agent.conditional_distributions[0],
output_agent.conditional_distributions[0])
score = LLHScorer.score(self.data, conditional_distribution)
if score > best_llh:
best_llh = score
best_distribution = conditional_distribution
# best_index = input_var
# print best_index
self.conditional_distributions = [best_distribution]
def llh(self, data):
assert isinstance(data, XSeqDataSet)
sample = np.zeros(data.arrays['X'].shape)
input_indices = list(range(self.output_index)) + \
list(range(self.output_index + 1, data.arrays['X'].shape[1], 1))
inputs = data.variable_subsets(input_indices)
outputs = data.variable_subsets([self.output_index])
X_data = inputs
XY_data = XYDataSet()
XY_data.labels['X'] = inputs.labels['X']
XY_data.labels['Y'] = outputs.labels['X']
XY_data.arrays['X'] = inputs.arrays['X']
XY_data.arrays['Y'] = outputs.arrays['X']
llh = self.input_distribution.llh(X_data)
llh += self.output_distribution.llh(XY_data)
return llh
def fit(self):
best_llh = -np.Inf
best_distribution = None
# best_index = None
for input_var in range(self.data.arrays['X'].shape[1]):
inputs = self.data.variable_subsets(
list(range(input_var)) +
list(range(input_var + 1, self.data.arrays['X'].shape[1], 1)))
outputs = self.data.variable_subsets([input_var])
input_agent = self.input_learner()
input_agent.load_data(inputs)
input_agent.fit()
XY_data = XYDataSet()
XY_data.labels['X'] = inputs.labels['X']
XY_data.labels['Y'] = outputs.labels['X']
XY_data.arrays['X'] = inputs.arrays['X']
XY_data.arrays['Y'] = outputs.arrays['X']
output_agent = self.output_learner()
output_agent.load_data(XY_data)
output_agent.fit()
conditional_distribution = RegressionDAG(
input_var, input_agent.conditional_distributions[0],
output_agent.conditional_distributions[0])
score = LLHScorer.score(self.data, conditional_distribution)
if score > best_llh:
best_llh = score
best_distribution = conditional_distribution
# best_index = input_var
# print best_index
self.conditional_distributions = [best_distribution]