def fivo(cell, inputs, seq_lengths, num_samples=1, resampling_criterion=ess_criterion, parallel_iterations=30, swap_memory=True, random_seed=None): # batch_size represents the number of particle filters running in parallel. batch_size = tf.shape(seq_lengths)[0] max_seq_len = tf.reduce_max(seq_lengths) data_dim = tf.shape(inputs[0])[-1] seq_mask = tf.transpose( tf.sequence_mask(seq_lengths, maxlen=max_seq_len, dtype=tf.float32), perm=[1, 0]) # Each sequence in the batch will be the input data for a different # particle filter. The batch will be laid out as: # particle 1 of particle filter 1 # particle 1 of particle filter 2 # ... # particle 1 of particle filter batch_size # particle 2 of particle filter 1 # ... # particle num_samples of particle filter batch_size if num_samples > 1: inputs, seq_mask = nested.tile_tensors([inputs, seq_mask], [1, num_samples]) # inputs: [max_seq_len, batch_size*num_samples, ...] (Duong) inputs_ta, mask_ta = nested.tas_for_tensors([inputs, seq_mask], max_seq_len) t0 = tf.constant(0, tf.int32) init_states = cell.zero_state(batch_size * num_samples, tf.float32) ta_names = ['log_weights', 'log_ess', 'resampled'] tas = [tf.TensorArray(tf.float32, max_seq_len, name='%s_ta' % n) for n in ta_names] log_weights_acc = tf.zeros([num_samples, batch_size], dtype=tf.float32) log_p_hat_acc = tf.zeros([batch_size], dtype=tf.float32) kl_acc = tf.zeros([num_samples * batch_size], dtype=tf.float32) accs = (log_weights_acc, log_p_hat_acc, kl_acc) def while_predicate(t, *unused_args): return t < max_seq_len def while_step(t, rnn_state, tas, accs): """Implements one timestep of FIVO computation.""" log_weights_acc, log_p_hat_acc, kl_acc = accs cur_inputs, cur_mask = nested.read_tas([inputs_ta, mask_ta], t) # cur_inputs: slice at time t, [batch_size*num_samples, ...] (Duong) # Run the cell for one step. log_q_z, log_p_z, log_p_x_given_z, kl, new_state, new_rnn_out\ = cell(cur_inputs, rnn_state, cur_mask, ) # Compute the incremental weight and use it to update the current # accumulated weight. kl_acc += kl * cur_mask log_alpha = (log_p_x_given_z + log_p_z - log_q_z) * cur_mask # ELBO (Duong) log_alpha = tf.reshape(log_alpha, [num_samples, batch_size]) log_weights_acc += log_alpha # Calculate the effective sample size. ess_num = 2 * tf.reduce_logsumexp(log_weights_acc, axis=0) ess_denom = tf.reduce_logsumexp(2 * log_weights_acc, axis=0) log_ess = ess_num - ess_denom # Calculate the ancestor indices via resampling. Because we maintain the # log unnormalized weights, we pass the weights in as logits, allowing # the distribution object to apply a softmax and normalize them. resampling_dist = tf.contrib.distributions.Categorical( logits=tf.transpose(log_weights_acc, perm=[1, 0])) ancestor_inds = tf.stop_gradient( resampling_dist.sample(sample_shape=num_samples, seed=random_seed)) # Because the batch is flattened and laid out as discussed # above, we must modify ancestor_inds to index the proper samples. # The particles in the ith filter are distributed every batch_size rows # in the batch, and offset i rows from the top. So, to correct the indices # we multiply by the batch_size and add the proper offset. Crucially, # when ancestor_inds is flattened the layout of the batch is maintained. offset = tf.expand_dims(tf.range(batch_size), 0) ancestor_inds = tf.reshape(ancestor_inds * batch_size + offset, [-1]) noresample_inds = tf.range(num_samples * batch_size) # Decide whether or not we should resample; don't resample if we are past # the end of a sequence. should_resample = resampling_criterion(num_samples, log_ess, t) should_resample = tf.logical_and(should_resample, cur_mask[:batch_size] > 0.) float_should_resample = tf.to_float(should_resample) ancestor_inds = tf.where( tf.tile(should_resample, [num_samples]), ancestor_inds, noresample_inds) new_state = nested.gather_tensors(new_state, ancestor_inds) # Update the TensorArrays before we reset the weights so that we capture # the incremental weights and not zeros. ta_updates = [log_weights_acc, log_ess, float_should_resample] new_tas = [ta.write(t, x) for ta, x in zip(tas, ta_updates)] # For the particle filters that resampled, update log_p_hat and # reset weights to zero. log_p_hat_update = tf.reduce_logsumexp( log_weights_acc, axis=0) - tf.log(tf.to_float(num_samples)) log_p_hat_acc += log_p_hat_update * float_should_resample log_weights_acc *= (1. - tf.tile(float_should_resample[tf.newaxis, :], [num_samples, 1])) new_accs = (log_weights_acc, log_p_hat_acc, kl_acc) return t + 1, new_state, new_tas, new_accs _, _, tas, accs = tf.while_loop(while_predicate, while_step, loop_vars=(t0, init_states, tas, accs), parallel_iterations=parallel_iterations, swap_memory=swap_memory) log_weights, log_ess, resampled = [x.stack() for x in tas] final_log_weights, log_p_hat, kl = accs # Add in the final weight update to log_p_hat. log_p_hat += (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) log_weights = tf.transpose(log_weights, perm=[0, 2, 1]) return log_p_hat, kl, log_weights, log_ess, resampled
def elbo(cell, inputs, seq_lengths, num_samples=1, parallel_iterations=30, swap_memory=True): batch_size = tf.shape(seq_lengths)[0] max_seq_len = tf.reduce_max(seq_lengths) data_dim = tf.shape(inputs[0])[-1] seq_mask = tf.transpose( tf.sequence_mask(seq_lengths, maxlen=max_seq_len, dtype=tf.float32), perm=[1, 0]) if num_samples > 1: inputs, seq_mask = nested.tile_tensors([inputs, seq_mask], [1, num_samples]) inputs_ta, mask_ta = nested.tas_for_tensors([inputs, seq_mask], max_seq_len) t0 = tf.constant(0, tf.int32) init_states = cell.zero_state(batch_size * num_samples, tf.float32) init_inputs, init_mask = nested.read_tas([inputs_ta, mask_ta], t0) ta_names = ['log_weights', 'log_ess'] tas = [tf.TensorArray(tf.float32, max_seq_len, name='%s_ta' % n) for n in ta_names] log_weights_acc = tf.zeros([num_samples, batch_size], dtype=tf.float32) kl_acc = tf.zeros([num_samples * batch_size], dtype=tf.float32) accs = (log_weights_acc, kl_acc) def while_predicate(t, *unused_args): return t < max_seq_len def while_step(t, rnn_state, tas, accs): """Implements one timestep of IWAE computation.""" log_weights_acc, kl_acc = accs cur_inputs, cur_mask = nested.read_tas([inputs_ta, mask_ta], t) # Run the cell for one step. log_q_z, log_p_z, log_p_x_given_z, kl, new_state, new_rnn_out\ = cell(cur_inputs, rnn_state, cur_mask, ) # Compute the incremental weight and use it to update the current # accumulated weight. kl_acc += kl * cur_mask log_alpha = (log_p_x_given_z + log_p_z - log_q_z) * cur_mask log_alpha = tf.reshape(log_alpha, [num_samples, batch_size]) log_weights_acc += log_alpha # Calculate the effective sample size. ess_num = 2 * tf.reduce_logsumexp(log_weights_acc, axis=0) ess_denom = tf.reduce_logsumexp(2 * log_weights_acc, axis=0) log_ess = ess_num - ess_denom # Update the Tensorarrays and accumulators. ta_updates = [log_weights_acc, log_ess] new_tas = [ta.write(t, x) for ta, x in zip(tas, ta_updates)] new_accs = (log_weights_acc, kl_acc) return t + 1, new_state, new_tas, new_accs _, _, tas, accs = tf.while_loop(while_predicate, while_step, loop_vars=(t0, init_states, tas, accs), parallel_iterations=parallel_iterations, swap_memory=swap_memory) log_weights, log_ess = [x.stack() for x in tas] final_log_weights, kl = accs log_p_hat = (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) log_weights = tf.transpose(log_weights, perm=[0, 2, 1]) return log_p_hat, kl, log_weights, log_ess
def iwae(cell, inputs, seq_lengths, num_samples=1, parallel_iterations=30, swap_memory=True): """Computes the IWAE lower bound on the log marginal probability. This method accepts a stochastic latent variable model and some observations and computes a stochastic lower bound on the log marginal probability of the observations. The IWAE estimator is defined by averaging multiple importance weights. For more details see "Importance Weighted Autoencoders" by Burda et al. https://arxiv.org/abs/1509.00519. When num_samples = 1, this bound becomes the evidence lower bound (ELBO). Args: cell: A callable that implements one timestep of the model. See models/vrnn.py for an example. inputs: The inputs to the model. A potentially nested list or tuple of Tensors each of shape [max_seq_len, batch_size, ...]. The Tensors must have a rank at least two and have matching shapes in the first two dimensions, which represent time and the batch respectively. At each timestep 'cell' will be called with a slice of the Tensors in inputs. seq_lengths: A [batch_size] Tensor of ints encoding the length of each sequence in the batch (sequences can be padded to a common length). num_samples: The number of samples to use. parallel_iterations: The number of parallel iterations to use for the internal while loop. swap_memory: Whether GPU-CPU memory swapping should be enabled for the internal while loop. Returns: log_p_hat: A Tensor of shape [batch_size] containing IWAE's estimate of the log marginal probability of the observations. kl: A Tensor of shape [batch_size] containing the kl divergence from q(z|x) to p(z), averaged over samples. log_weights: A Tensor of shape [max_seq_len, batch_size, num_samples] containing the log weights at each timestep. Will not be valid for timesteps past the end of a sequence. log_ess: A Tensor of shape [max_seq_len, batch_size] containing the log effective sample size at each timestep. Will not be valid for timesteps past the end of a sequence. """ batch_size = tf.shape(seq_lengths)[0] max_seq_len = tf.reduce_max(seq_lengths) seq_mask = tf.transpose( tf.sequence_mask(seq_lengths, maxlen=max_seq_len, dtype=tf.float32), perm=[1, 0]) if num_samples > 1: inputs, seq_mask = nested.tile_tensors([inputs, seq_mask], [1, num_samples]) inputs_ta, mask_ta = nested.tas_for_tensors([inputs, seq_mask], max_seq_len) t0 = tf.constant(0, tf.int32) init_states = cell.zero_state(batch_size * num_samples, tf.float32) ta_names = ['log_weights', 'log_ess'] tas = [tf.TensorArray(tf.float32, max_seq_len, name='%s_ta' % n) for n in ta_names] log_weights_acc = tf.zeros([num_samples, batch_size], dtype=tf.float32) kl_acc = tf.zeros([num_samples * batch_size], dtype=tf.float32) accs = (log_weights_acc, kl_acc) def while_predicate(t, *unused_args): return t < max_seq_len def while_step(t, rnn_state, tas, accs): """Implements one timestep of IWAE computation.""" log_weights_acc, kl_acc = accs cur_inputs, cur_mask = nested.read_tas([inputs_ta, mask_ta], t) # Run the cell for one step. log_q_z, log_p_z, log_p_x_given_z, kl, new_state = cell( cur_inputs, rnn_state, cur_mask, ) # Compute the incremental weight and use it to update the current # accumulated weight. kl_acc += kl * cur_mask log_alpha = (log_p_x_given_z + log_p_z - log_q_z) * cur_mask log_alpha = tf.reshape(log_alpha, [num_samples, batch_size]) log_weights_acc += log_alpha # Calculate the effective sample size. ess_num = 2 * tf.reduce_logsumexp(log_weights_acc, axis=0) ess_denom = tf.reduce_logsumexp(2 * log_weights_acc, axis=0) log_ess = ess_num - ess_denom # Update the Tensorarrays and accumulators. ta_updates = [log_weights_acc, log_ess] new_tas = [ta.write(t, x) for ta, x in zip(tas, ta_updates)] new_accs = (log_weights_acc, kl_acc) return t + 1, new_state, new_tas, new_accs _, _, tas, accs = tf.while_loop( while_predicate, while_step, loop_vars=(t0, init_states, tas, accs), parallel_iterations=parallel_iterations, swap_memory=swap_memory) log_weights, log_ess = [x.stack() for x in tas] final_log_weights, kl = accs log_p_hat = (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) log_weights = tf.transpose(log_weights, perm=[0, 2, 1]) return log_p_hat, kl, log_weights, log_ess
def fivo(cell, inputs, seq_lengths, num_samples=1, resampling_criterion=ess_criterion, parallel_iterations=30, swap_memory=True, random_seed=None): """Computes the FIVO lower bound on the log marginal probability. This method accepts a stochastic latent variable model and some observations and computes a stochastic lower bound on the log marginal probability of the observations. The lower bound is defined by a particle filter's unbiased estimate of the marginal probability of the observations. For more details see "Filtering Variational Objectives" by Maddison et al. https://arxiv.org/abs/1705.09279. When the resampling criterion is "never resample", this bound becomes IWAE. Args: cell: A callable that implements one timestep of the model. See models/vrnn.py for an example. inputs: The inputs to the model. A potentially nested list or tuple of Tensors each of shape [max_seq_len, batch_size, ...]. The Tensors must have a rank at least two and have matching shapes in the first two dimensions, which represent time and the batch respectively. At each timestep 'cell' will be called with a slice of the Tensors in inputs. seq_lengths: A [batch_size] Tensor of ints encoding the length of each sequence in the batch (sequences can be padded to a common length). num_samples: The number of particles to use in each particle filter. resampling_criterion: The resampling criterion to use for this particle filter. Must accept the number of samples, the effective sample size, and the current timestep and return a boolean Tensor of shape [batch_size] indicating whether each particle filter should resample. See ess_criterion and related functions defined in this file for examples. parallel_iterations: The number of parallel iterations to use for the internal while loop. Note that values greater than 1 can introduce non-determinism even when random_seed is provided. swap_memory: Whether GPU-CPU memory swapping should be enabled for the internal while loop. random_seed: The random seed to pass to the resampling operations in the particle filter. Mainly useful for testing. Returns: log_p_hat: A Tensor of shape [batch_size] containing FIVO's estimate of the log marginal probability of the observations. kl: A Tensor of shape [batch_size] containing the sum over time of the kl divergence from q_t(z_t|x) to p_t(z_t), averaged over particles. Note that this includes kl terms from trajectories that are culled during resampling steps. log_weights: A Tensor of shape [max_seq_len, batch_size, num_samples] containing the log weights at each timestep of the particle filter. Note that on timesteps when a resampling operation is performed the log weights are reset to 0. Will not be valid for timesteps past the end of a sequence. log_ess: A Tensor of shape [max_seq_len, batch_size] containing the log effective sample size of each particle filter at each timestep. Will not be valid for timesteps past the end of a sequence. resampled: A Tensor of shape [max_seq_len, batch_size] indicating when the particle filters resampled. Will be 1.0 on timesteps when resampling occurred and 0.0 on timesteps when it did not. """ # batch_size represents the number of particle filters running in parallel. batch_size = tf.shape(seq_lengths)[0] max_seq_len = tf.reduce_max(seq_lengths) seq_mask = tf.transpose( tf.sequence_mask(seq_lengths, maxlen=max_seq_len, dtype=tf.float32), perm=[1, 0]) # Each sequence in the batch will be the input data for a different # particle filter. The batch will be laid out as: # particle 1 of particle filter 1 # particle 1 of particle filter 2 # ... # particle 1 of particle filter batch_size # particle 2 of particle filter 1 # ... # particle num_samples of particle filter batch_size if num_samples > 1: inputs, seq_mask = nested.tile_tensors([inputs, seq_mask], [1, num_samples]) inputs_ta, mask_ta = nested.tas_for_tensors([inputs, seq_mask], max_seq_len) t0 = tf.constant(0, tf.int32) init_states = cell.zero_state(batch_size * num_samples, tf.float32) ta_names = ['log_weights', 'log_ess', 'resampled'] tas = [tf.TensorArray(tf.float32, max_seq_len, name='%s_ta' % n) for n in ta_names] log_weights_acc = tf.zeros([num_samples, batch_size], dtype=tf.float32) log_p_hat_acc = tf.zeros([batch_size], dtype=tf.float32) kl_acc = tf.zeros([num_samples * batch_size], dtype=tf.float32) accs = (log_weights_acc, log_p_hat_acc, kl_acc) def while_predicate(t, *unused_args): return t < max_seq_len def while_step(t, rnn_state, tas, accs): """Implements one timestep of FIVO computation.""" log_weights_acc, log_p_hat_acc, kl_acc = accs cur_inputs, cur_mask = nested.read_tas([inputs_ta, mask_ta], t) # Run the cell for one step. log_q_z, log_p_z, log_p_x_given_z, kl, new_state = cell( cur_inputs, rnn_state, cur_mask, ) # Compute the incremental weight and use it to update the current # accumulated weight. kl_acc += kl * cur_mask log_alpha = (log_p_x_given_z + log_p_z - log_q_z) * cur_mask log_alpha = tf.reshape(log_alpha, [num_samples, batch_size]) log_weights_acc += log_alpha # Calculate the effective sample size. ess_num = 2 * tf.reduce_logsumexp(log_weights_acc, axis=0) ess_denom = tf.reduce_logsumexp(2 * log_weights_acc, axis=0) log_ess = ess_num - ess_denom # Calculate the ancestor indices via resampling. Because we maintain the # log unnormalized weights, we pass the weights in as logits, allowing # the distribution object to apply a softmax and normalize them. resampling_dist = tf.contrib.distributions.Categorical( logits=tf.transpose(log_weights_acc, perm=[1, 0])) ancestor_inds = tf.stop_gradient( resampling_dist.sample(sample_shape=num_samples, seed=random_seed)) # Because the batch is flattened and laid out as discussed # above, we must modify ancestor_inds to index the proper samples. # The particles in the ith filter are distributed every batch_size rows # in the batch, and offset i rows from the top. So, to correct the indices # we multiply by the batch_size and add the proper offset. Crucially, # when ancestor_inds is flattened the layout of the batch is maintained. offset = tf.expand_dims(tf.range(batch_size), 0) ancestor_inds = tf.reshape(ancestor_inds * batch_size + offset, [-1]) noresample_inds = tf.range(num_samples * batch_size) # Decide whether or not we should resample; don't resample if we are past # the end of a sequence. should_resample = resampling_criterion(num_samples, log_ess, t) should_resample = tf.logical_and(should_resample, cur_mask[:batch_size] > 0.) float_should_resample = tf.to_float(should_resample) ancestor_inds = tf.where( tf.tile(should_resample, [num_samples]), ancestor_inds, noresample_inds) new_state = nested.gather_tensors(new_state, ancestor_inds) # Update the TensorArrays before we reset the weights so that we capture # the incremental weights and not zeros. ta_updates = [log_weights_acc, log_ess, float_should_resample] new_tas = [ta.write(t, x) for ta, x in zip(tas, ta_updates)] # For the particle filters that resampled, update log_p_hat and # reset weights to zero. log_p_hat_update = tf.reduce_logsumexp( log_weights_acc, axis=0) - tf.log(tf.to_float(num_samples)) log_p_hat_acc += log_p_hat_update * float_should_resample log_weights_acc *= (1. - tf.tile(float_should_resample[tf.newaxis, :], [num_samples, 1])) new_accs = (log_weights_acc, log_p_hat_acc, kl_acc) return t + 1, new_state, new_tas, new_accs _, _, tas, accs = tf.while_loop( while_predicate, while_step, loop_vars=(t0, init_states, tas, accs), parallel_iterations=parallel_iterations, swap_memory=swap_memory) log_weights, log_ess, resampled = [x.stack() for x in tas] final_log_weights, log_p_hat, kl = accs # Add in the final weight update to log_p_hat. log_p_hat += (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) log_weights = tf.transpose(log_weights, perm=[0, 2, 1]) return log_p_hat, kl, log_weights, log_ess, resampled
def iwae(cell, inputs, seq_lengths, num_samples=1, parallel_iterations=30, swap_memory=True): """Computes the IWAE lower bound on the log marginal probability. This method accepts a stochastic latent variable model and some observations and computes a stochastic lower bound on the log marginal probability of the observations. The IWAE estimator is defined by averaging multiple importance weights. For more details see "Importance Weighted Autoencoders" by Burda et al. https://arxiv.org/abs/1509.00519. When num_samples = 1, this bound becomes the evidence lower bound (ELBO). Args: cell: A callable that implements one timestep of the model. See models/vrnn.py for an example. inputs: The inputs to the model. A potentially nested list or tuple of Tensors each of shape [max_seq_len, batch_size, ...]. The Tensors must have a rank at least two and have matching shapes in the first two dimensions, which represent time and the batch respectively. At each timestep 'cell' will be called with a slice of the Tensors in inputs. seq_lengths: A [batch_size] Tensor of ints encoding the length of each sequence in the batch (sequences can be padded to a common length). num_samples: The number of samples to use. parallel_iterations: The number of parallel iterations to use for the internal while loop. swap_memory: Whether GPU-CPU memory swapping should be enabled for the internal while loop. Returns: log_p_hat: A Tensor of shape [batch_size] containing IWAE's estimate of the log marginal probability of the observations. kl: A Tensor of shape [batch_size] containing the kl divergence from q(z|x) to p(z), averaged over samples. log_weights: A Tensor of shape [max_seq_len, batch_size, num_samples] containing the log weights at each timestep. Will not be valid for timesteps past the end of a sequence. log_ess: A Tensor of shape [max_seq_len, batch_size] containing the log effective sample size at each timestep. Will not be valid for timesteps past the end of a sequence. """ batch_size = tf.shape(seq_lengths)[0] max_seq_len = tf.reduce_max(seq_lengths) seq_mask = tf.transpose(tf.sequence_mask(seq_lengths, maxlen=max_seq_len, dtype=tf.float32), perm=[1, 0]) if num_samples > 1: inputs, seq_mask = nested.tile_tensors([inputs, seq_mask], [1, num_samples]) inputs_ta, mask_ta = nested.tas_for_tensors([inputs, seq_mask], max_seq_len) t0 = tf.constant(0, tf.int32) init_states = cell.zero_state(batch_size * num_samples, tf.float32) ta_names = ['log_weights', 'log_ess'] tas = [ tf.TensorArray(tf.float32, max_seq_len, name='%s_ta' % n) for n in ta_names ] log_weights_acc = tf.zeros([num_samples, batch_size], dtype=tf.float32) kl_acc = tf.zeros([num_samples * batch_size], dtype=tf.float32) accs = (log_weights_acc, kl_acc) def while_predicate(t, *unused_args): return t < max_seq_len def while_step(t, rnn_state, tas, accs): """Implements one timestep of IWAE computation.""" log_weights_acc, kl_acc = accs cur_inputs, cur_mask = nested.read_tas([inputs_ta, mask_ta], t) # Run the cell for one step. log_q_z, log_p_z, log_p_x_given_z, kl, new_state = cell( cur_inputs, rnn_state, cur_mask, ) # Compute the incremental weight and use it to update the current # accumulated weight. kl_acc += kl * cur_mask log_alpha = (log_p_x_given_z + log_p_z - log_q_z) * cur_mask log_alpha = tf.reshape(log_alpha, [num_samples, batch_size]) log_weights_acc += log_alpha # Calculate the effective sample size. ess_num = 2 * tf.reduce_logsumexp(log_weights_acc, axis=0) ess_denom = tf.reduce_logsumexp(2 * log_weights_acc, axis=0) log_ess = ess_num - ess_denom # Update the Tensorarrays and accumulators. ta_updates = [log_weights_acc, log_ess] new_tas = [ta.write(t, x) for ta, x in zip(tas, ta_updates)] new_accs = (log_weights_acc, kl_acc) return t + 1, new_state, new_tas, new_accs _, _, tas, accs = tf.while_loop(while_predicate, while_step, loop_vars=(t0, init_states, tas, accs), parallel_iterations=parallel_iterations, swap_memory=swap_memory) log_weights, log_ess = [x.stack() for x in tas] final_log_weights, kl = accs log_p_hat = (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) log_weights = tf.transpose(log_weights, perm=[0, 2, 1]) return log_p_hat, kl, log_weights, log_ess
def fivo(cell, inputs, seq_lengths, num_samples=1, resampling_criterion=ess_criterion, parallel_iterations=30, swap_memory=True, random_seed=None): """Computes the FIVO lower bound on the log marginal probability. This method accepts a stochastic latent variable model and some observations and computes a stochastic lower bound on the log marginal probability of the observations. The lower bound is defined by a particle filter's unbiased estimate of the marginal probability of the observations. For more details see "Filtering Variational Objectives" by Maddison et al. https://arxiv.org/abs/1705.09279. When the resampling criterion is "never resample", this bound becomes IWAE. Args: cell: A callable that implements one timestep of the model. See models/vrnn.py for an example. inputs: The inputs to the model. A potentially nested list or tuple of Tensors each of shape [max_seq_len, batch_size, ...]. The Tensors must have a rank at least two and have matching shapes in the first two dimensions, which represent time and the batch respectively. At each timestep 'cell' will be called with a slice of the Tensors in inputs. seq_lengths: A [batch_size] Tensor of ints encoding the length of each sequence in the batch (sequences can be padded to a common length). num_samples: The number of particles to use in each particle filter. resampling_criterion: The resampling criterion to use for this particle filter. Must accept the number of samples, the effective sample size, and the current timestep and return a boolean Tensor of shape [batch_size] indicating whether each particle filter should resample. See ess_criterion and related functions defined in this file for examples. parallel_iterations: The number of parallel iterations to use for the internal while loop. Note that values greater than 1 can introduce non-determinism even when random_seed is provided. swap_memory: Whether GPU-CPU memory swapping should be enabled for the internal while loop. random_seed: The random seed to pass to the resampling operations in the particle filter. Mainly useful for testing. Returns: log_p_hat: A Tensor of shape [batch_size] containing FIVO's estimate of the log marginal probability of the observations. kl: A Tensor of shape [batch_size] containing the sum over time of the kl divergence from q_t(z_t|x) to p_t(z_t), averaged over particles. Note that this includes kl terms from trajectories that are culled during resampling steps. log_weights: A Tensor of shape [max_seq_len, batch_size, num_samples] containing the log weights at each timestep of the particle filter. Note that on timesteps when a resampling operation is performed the log weights are reset to 0. Will not be valid for timesteps past the end of a sequence. log_ess: A Tensor of shape [max_seq_len, batch_size] containing the log effective sample size of each particle filter at each timestep. Will not be valid for timesteps past the end of a sequence. resampled: A Tensor of shape [max_seq_len, batch_size] indicating when the particle filters resampled. Will be 1.0 on timesteps when resampling occurred and 0.0 on timesteps when it did not. """ # batch_size represents the number of particle filters running in parallel. batch_size = tf.shape(seq_lengths)[0] max_seq_len = tf.reduce_max(seq_lengths) seq_mask = tf.transpose(tf.sequence_mask(seq_lengths, maxlen=max_seq_len, dtype=tf.float32), perm=[1, 0]) # Each sequence in the batch will be the input data for a different # particle filter. The batch will be laid out as: # particle 1 of particle filter 1 # particle 1 of particle filter 2 # ... # particle 1 of particle filter batch_size # particle 2 of particle filter 1 # ... # particle num_samples of particle filter batch_size if num_samples > 1: inputs, seq_mask = nested.tile_tensors([inputs, seq_mask], [1, num_samples]) inputs_ta, mask_ta = nested.tas_for_tensors([inputs, seq_mask], max_seq_len) t0 = tf.constant(0, tf.int32) init_states = cell.zero_state(batch_size * num_samples, tf.float32) ta_names = ['log_weights', 'log_ess', 'resampled'] tas = [ tf.TensorArray(tf.float32, max_seq_len, name='%s_ta' % n) for n in ta_names ] log_weights_acc = tf.zeros([num_samples, batch_size], dtype=tf.float32) log_p_hat_acc = tf.zeros([batch_size], dtype=tf.float32) kl_acc = tf.zeros([num_samples * batch_size], dtype=tf.float32) accs = (log_weights_acc, log_p_hat_acc, kl_acc) def while_predicate(t, *unused_args): return t < max_seq_len def while_step(t, rnn_state, tas, accs): """Implements one timestep of FIVO computation.""" log_weights_acc, log_p_hat_acc, kl_acc = accs cur_inputs, cur_mask = nested.read_tas([inputs_ta, mask_ta], t) # Run the cell for one step. log_q_z, log_p_z, log_p_x_given_z, kl, new_state = cell( cur_inputs, rnn_state, cur_mask, ) # Compute the incremental weight and use it to update the current # accumulated weight. kl_acc += kl * cur_mask log_alpha = (log_p_x_given_z + log_p_z - log_q_z) * cur_mask log_alpha = tf.reshape(log_alpha, [num_samples, batch_size]) log_weights_acc += log_alpha # Calculate the effective sample size. ess_num = 2 * tf.reduce_logsumexp(log_weights_acc, axis=0) ess_denom = tf.reduce_logsumexp(2 * log_weights_acc, axis=0) log_ess = ess_num - ess_denom # Calculate the ancestor indices via resampling. Because we maintain the # log unnormalized weights, we pass the weights in as logits, allowing # the distribution object to apply a softmax and normalize them. resampling_dist = tf.contrib.distributions.Categorical( logits=tf.transpose(log_weights_acc, perm=[1, 0])) ancestor_inds = tf.stop_gradient( resampling_dist.sample(sample_shape=num_samples, seed=random_seed)) # Because the batch is flattened and laid out as discussed # above, we must modify ancestor_inds to index the proper samples. # The particles in the ith filter are distributed every batch_size rows # in the batch, and offset i rows from the top. So, to correct the indices # we multiply by the batch_size and add the proper offset. Crucially, # when ancestor_inds is flattened the layout of the batch is maintained. offset = tf.expand_dims(tf.range(batch_size), 0) ancestor_inds = tf.reshape(ancestor_inds * batch_size + offset, [-1]) noresample_inds = tf.range(num_samples * batch_size) # Decide whether or not we should resample; don't resample if we are past # the end of a sequence. should_resample = resampling_criterion(num_samples, log_ess, t) should_resample = tf.logical_and(should_resample, cur_mask[:batch_size] > 0.) float_should_resample = tf.to_float(should_resample) ancestor_inds = tf.where(tf.tile(should_resample, [num_samples]), ancestor_inds, noresample_inds) new_state = nested.gather_tensors(new_state, ancestor_inds) # Update the TensorArrays before we reset the weights so that we capture # the incremental weights and not zeros. ta_updates = [log_weights_acc, log_ess, float_should_resample] new_tas = [ta.write(t, x) for ta, x in zip(tas, ta_updates)] # For the particle filters that resampled, update log_p_hat and # reset weights to zero. log_p_hat_update = tf.reduce_logsumexp( log_weights_acc, axis=0) - tf.log(tf.to_float(num_samples)) log_p_hat_acc += log_p_hat_update * float_should_resample log_weights_acc *= ( 1. - tf.tile(float_should_resample[tf.newaxis, :], [num_samples, 1])) new_accs = (log_weights_acc, log_p_hat_acc, kl_acc) return t + 1, new_state, new_tas, new_accs _, _, tas, accs = tf.while_loop(while_predicate, while_step, loop_vars=(t0, init_states, tas, accs), parallel_iterations=parallel_iterations, swap_memory=swap_memory) log_weights, log_ess, resampled = [x.stack() for x in tas] final_log_weights, log_p_hat, kl = accs # Add in the final weight update to log_p_hat. log_p_hat += (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) log_weights = tf.transpose(log_weights, perm=[0, 2, 1]) return log_p_hat, kl, log_weights, log_ess, resampled
def create_eval_graph(inputs, targets, lengths, model, config): parallel_iterations = 30 swap_memory = True batch_size = tf.shape(lengths)[0] num_samples = config.num_samples max_seq_len = tf.reduce_max(lengths) init_states = model.zero_state(batch_size * num_samples, tf.float32) seq_mask = tf.transpose(tf.sequence_mask(lengths, maxlen=max_seq_len, dtype=tf.float32), perm=[1, 0]) if num_samples > 1: inputs_tmp, seq_mask = nested.tile_tensors( [(inputs, targets), seq_mask], [1, num_samples]) inputs_ta, mask_ta = nested.tas_for_tensors([inputs_tmp, seq_mask], max_seq_len) else: inputs_ta, mask_ta = nested.tas_for_tensors( [(inputs, targets), seq_mask], max_seq_len) t0 = tf.constant(0, tf.int32) init_states = model.zero_state(batch_size * num_samples, tf.float32) ta_names = [ 'log_weights_t', 'sampleds', 'trues', 'rnn_states', 'rnn_latents', 'rnn_outs' ] tas = [ tf.TensorArray(tf.float32, max_seq_len, name='%s_ta' % n) for n in ta_names ] log_weights_acc = tf.zeros([num_samples, batch_size], dtype=tf.float32) log_p_hat_acc = tf.zeros([batch_size], dtype=tf.float32) kl_acc = tf.zeros([num_samples * batch_size], dtype=tf.float32) if config.bound == "elbo": accs = (log_weights_acc, kl_acc) elif config.bound == "fivo": accs = (log_weights_acc, log_p_hat_acc, kl_acc) target_sampled0 = tf.zeros( shape=[batch_size * num_samples, config.data_dim], dtype=tf.float32) target_true0 = tf.zeros(shape=[batch_size * num_samples, config.data_dim], dtype=tf.float32) init_targets = (target_sampled0, target_true0) def while_predicate(t, *unused_args): return t < max_seq_len resampling_criterion = bounds.ess_criterion def while_step(t, rnn_state, tas, accs, while_samples): """Implements one timestep of IWAE computation.""" if config.bound == "elbo": log_weights_acc, kl_acc = accs elif config.bound == "fivo": log_weights_acc, log_p_hat_acc, kl_acc = accs cur_inputs, cur_mask = nested.read_tas([inputs_ta, mask_ta], t) if config.missing_data: cur_inputs = tf.cond( tf.logical_and(t < max_seq_len - 6, t >= max_seq_len - 18), lambda: while_samples, lambda: cur_inputs) # Run the cell for one step. log_q_z, log_p_z, log_p_x_given_z, kl, new_state, new_rnn_out, dists_return\ = model(cur_inputs, rnn_state, cur_mask, return_value = "probs" ) new_sample0 = dists.sample_from_probs(dists_return, config.lat_bins, config.lon_bins, config.sog_bins, config.cog_bins) new_sample0 = tf.cast(new_sample0, tf.float32) new_sample_ = (new_sample0, tf.zeros_like(new_sample0, dtype=tf.float32)) # Compute the incremental weight and use it to update the current # accumulated weight kl_acc += kl * cur_mask log_alpha = (log_p_x_given_z + log_p_z - log_q_z) * cur_mask log_alpha = tf.reshape(log_alpha, [config.num_samples, batch_size]) log_weights_acc += log_alpha # Calculate the effective sample size. ess_num = 2 * tf.reduce_logsumexp(log_weights_acc, axis=0) ess_denom = tf.reduce_logsumexp(2 * log_weights_acc, axis=0) log_ess = ess_num - ess_denom if config.bound == "fivo": # Calculate the ancestor indices via resampling. Because we maintain the # log unnormalized weights, we pass the weights in as logits, allowing # the distribution object to apply a softmax and normalize them. resampling_dist = tf.contrib.distributions.Categorical( logits=tf.transpose(log_weights_acc, perm=[1, 0])) ancestor_inds = tf.stop_gradient( resampling_dist.sample(sample_shape=num_samples, seed=config.random_seed)) # Because the batch is flattened and laid out as discussed # above, we must modify ancestor_inds to index the proper samples. # The particles in the ith filter are distributed every batch_size rows # in the batch, and offset i rows from the top. So, to correct the indices # we multiply by the batch_size and add the proper offset. Crucially, # when ancestor_inds is flattened the layout of the batch is maintained. offset = tf.expand_dims(tf.range(batch_size), 0) ancestor_inds = tf.reshape(ancestor_inds * batch_size + offset, [-1]) noresample_inds = tf.range(num_samples * batch_size) # Decide whether or not we should resample; don't resample if we are past # the end of a sequence. should_resample = resampling_criterion(num_samples, log_ess, t) should_resample = tf.logical_and(should_resample, cur_mask[:batch_size] > 0.) float_should_resample = tf.to_float(should_resample) ancestor_inds = tf.where(tf.tile(should_resample, [num_samples]), ancestor_inds, noresample_inds) new_state = nested.gather_tensors(new_state, ancestor_inds) new_sample_ = nested.gather_tensors(new_sample_, ancestor_inds) # Update the Tensorarrays and accumulators. ta_updates = [ log_alpha, new_sample_[0], new_sample_[1], new_state[0], new_state[1], new_rnn_out ] # ta_updates = [log_weights_acc, log_ess] new_tas = [ta.write(t, x) for ta, x in zip(tas, ta_updates)] if config.bound == "fivo": # For the particle filters that resampled, update log_p_hat and # reset weights to zero. log_p_hat_update = tf.reduce_logsumexp( log_weights_acc, axis=0) - tf.log(tf.to_float(num_samples)) log_p_hat_acc += log_p_hat_update * float_should_resample log_weights_acc *= (1. - tf.tile( float_should_resample[tf.newaxis, :], [num_samples, 1])) new_accs = (log_weights_acc, log_p_hat_acc, kl_acc) elif config.bound == "elbo": new_accs = (log_weights_acc, kl_acc) return t + 1, new_state, new_tas, new_accs, new_sample_ _, _, tas, accs, new_sample = tf.while_loop( while_predicate, while_step, loop_vars=(t0, init_states, tas, accs, init_targets), parallel_iterations=parallel_iterations, swap_memory=swap_memory) #log_weights, log_ess = [x.stack() for x in tas] log_weights, track_sample, track_true, \ rnn_state_tf, rnn_latent_tf, rnn_out_tf = [x.stack() for x in tas] #log_weights, log_ess, resampled = [x.stack() for x in tas] if config.bound == "fivo": final_log_weights, log_p_hat, kl = accs # Add in the final weight update to log_p_hat. log_p_hat += (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) elif config.bound == "elbo": final_log_weights, kl = accs log_p_hat = (tf.reduce_logsumexp(final_log_weights, axis=0) - tf.log(tf.to_float(num_samples))) kl = tf.reduce_mean(tf.reshape(kl, [num_samples, batch_size]), axis=0) ll_per_seq = log_p_hat ll_per_t = ll_per_seq / tf.to_float(lengths) # ll_per_t = tf.reduce_mean(ll_per_seq / tf.to_float(lengths)) # ll_per_seq = tf.reduce_mean(ll_per_seq) return track_sample, track_true, log_weights, ll_per_t, \ final_log_weights/tf.to_float(lengths), rnn_state_tf, rnn_latent_tf, rnn_out_tf