Python utils.logSumExp() Examples

The following are 30 code examples of utils.logSumExp(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module utils , or try the search function .
Example #1
Source File: rebar.py    From models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF = self._create_loss()
    self.optimizerLoss = tf.reduce_mean(self.optimizerLoss)

    # Setup optimizer
    grads_and_vars = self.optimizer_class.compute_gradients(self.optimizerLoss)
    self._create_train_op(grads_and_vars)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #2
Source File: rebar.py    From object_detection_with_tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF = self._create_loss()
    self.optimizerLoss = tf.reduce_mean(self.optimizerLoss)

    # Setup optimizer
    grads_and_vars = self.optimizer_class.compute_gradients(self.optimizerLoss)
    self._create_train_op(grads_and_vars)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #3
Source File: rebar.py    From object_detection_with_tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #4
Source File: rebar.py    From object_detection_with_tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #5
Source File: rebar.py    From object_detection_with_tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #6
Source File: rebar.py    From object_detection_with_tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective = self._create_loss()

    # Create additional updates for control variates and temperature
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV')))

    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #7
Source File: rebar.py    From object_detection_with_tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective, variance_objective_grad = self._create_loss()
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV'))
                 + [(variance_objective_grad, self.pre_temperature_variable)])
    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #8
Source File: rebar.py    From g-tensorflow-models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF = self._create_loss()
    self.optimizerLoss = tf.reduce_mean(self.optimizerLoss)

    # Setup optimizer
    grads_and_vars = self.optimizer_class.compute_gradients(self.optimizerLoss)
    self._create_train_op(grads_and_vars)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #9
Source File: rebar.py    From g-tensorflow-models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #10
Source File: rebar.py    From g-tensorflow-models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #11
Source File: rebar.py    From g-tensorflow-models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #12
Source File: rebar.py    From g-tensorflow-models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective = self._create_loss()

    # Create additional updates for control variates and temperature
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV')))

    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #13
Source File: rebar.py    From g-tensorflow-models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective, variance_objective_grad = self._create_loss()
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV'))
                 + [(variance_objective_grad, self.pre_temperature_variable)])
    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #14
Source File: rebar.py    From object_detection_kitti with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective, variance_objective_grad = self._create_loss()
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV'))
                 + [(variance_objective_grad, self.pre_temperature_variable)])
    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #15
Source File: rebar.py    From models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #16
Source File: rebar.py    From models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #17
Source File: rebar.py    From models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #18
Source File: rebar.py    From models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective = self._create_loss()

    # Create additional updates for control variates and temperature
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV')))

    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #19
Source File: rebar.py    From models with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective, variance_objective_grad = self._create_loss()
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV'))
                 + [(variance_objective_grad, self.pre_temperature_variable)])
    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #20
Source File: rebar.py    From multilabel-image-classification-tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF = self._create_loss()
    self.optimizerLoss = tf.reduce_mean(self.optimizerLoss)

    # Setup optimizer
    grads_and_vars = self.optimizer_class.compute_gradients(self.optimizerLoss)
    self._create_train_op(grads_and_vars)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #21
Source File: rebar.py    From multilabel-image-classification-tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #22
Source File: rebar.py    From multilabel-image-classification-tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #23
Source File: rebar.py    From multilabel-image-classification-tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #24
Source File: rebar.py    From multilabel-image-classification-tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective = self._create_loss()

    # Create additional updates for control variates and temperature
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV')))

    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #25
Source File: rebar.py    From multilabel-image-classification-tensorflow with MIT License 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective, variance_objective_grad = self._create_loss()
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV'))
                 + [(variance_objective_grad, self.pre_temperature_variable)])
    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #26
Source File: unit.py    From NeuralFactorGraph with MIT License 5 votes vote down vote up 
def assertEqualMarginals(self, graph, all_sequences, sent_likelihood):
		"""
		Check factor/variable marginals are approximately equal 
		to marginals obtained from brute force inference
		"""

		# Check variable marginals
		threshold = 0.01
		eq = True

		denom = -float('inf')
		maxDiff = -float('inf')

		for s, sequence in enumerate(all_sequences):
			denom = utils.logSumExp(sent_likelihood[s], denom)

		# Iterate over all timesteps
		for t in range(graph.T):
			for tag in self.model.uniqueTags:
				tagBeliefs = graph.getVarByTimestepnTag(t, tag.idx).belief.cpu().data.numpy()
				for labelIdx in range(tag.size()):
					num = -float('inf')
					for s, sequence in enumerate(all_sequences):
						if sequence[t][tag.idx]==labelIdx:
							num = utils.logSumExp(sent_likelihood[s], num)

					# Check difference
					# maxDiff = max(maxDiff, np.max(np.abs(tagBeliefs[labelIdx]- np.exp(num-denom))))
					tagLogProb = np.exp(num-denom)
					maxDiff = max(maxDiff, np.max(np.abs(np.exp(tagBeliefs[labelIdx]) - tagLogProb)))
					if maxDiff > threshold:
						eq = False

		if not eq:
			print("Marginals not equal. Max difference of %f" %maxDiff)
		else:
			print("Passed unit test!")

		sys.exit(0)
Example #27
Source File: rebar.py    From yolo_v2 with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #28
Source File: rebar.py    From yolo_v2 with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #29
Source File: rebar.py    From yolo_v2 with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads = self._create_loss()
    self._create_train_op(loss_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))
Example #30
Source File: rebar.py    From yolo_v2 with Apache License 2.0 5 votes vote down vote up 
def _create_network(self):
    logF, loss_grads, variance_objective = self._create_loss()

    # Create additional updates for control variates and temperature
    eta_grads = (self.optimizer_class.compute_gradients(variance_objective,
                                                        var_list=tf.get_collection('CV')))

    self._create_train_op(loss_grads, eta_grads)

    # Create IWAE lower bound for evaluation
    self.logF = self._reshape(logF)
    self.iwae = tf.reduce_mean(U.logSumExp(self.logF, axis=1) -
                               tf.log(tf.to_float(self.n_samples)))