a4c4a4f2d33081806397784f96b113d5b3b32c9a,tensorforce/core/memories/prioritized_replay.py,PrioritizedReplay,tf_retrieve_timesteps,#PrioritizedReplay#Any#,149

Before Change

    def tf_retrieve_timesteps(self, n):
        num_timesteps = (self.memory_index - self.episode_indices[0] - 2) % self.capacity + 1
        n = tf.minimum(x=n, y=num_timesteps)

        indices = tf.zeros(shape=(n,), dtype=tf.int32)
        // Only compute once outside loop.
        sum_priorities = tf.reduce_sum(input_tensor=self.priorities, axis=0)
        loop_index = tf.get_variable(
            name="loop-index",
            dtype=tf.int32,
            initializer=0,
            trainable=False
        )

        def cond(sum_priorities, priorities, indices, loop_index, n):
            return tf.less(loop_index, n)

        def sampling_fn(sum_priorities, priorities, indices, loop_index, n):
            // 3 cases:
            // 1. not_sampled_index < len(self.observations):
            // 2. sum_priorities / self.capacity < util.epsilon:
            // 3. all other cases

            def true_fn():
                // tf cond on
                // 1. not_sampled_index < len(self.observations):
                // -> fetch not sampled index
                // 2. sum_priorities / self.capacity < util.epsilon:
                // -> randomly sample
                pass

            def false_fn():
                // Priority sampling loop.
                pass

            index = tf.cond(
                pred=tf.logical_or(x=(self.no_priority_index < n), y=(sum_priorities / self.capacity < util.epsilon)),
                true_fn=true_fn,
                false_fn=false_fn
            )

        tf.while_loop(
            cond=cond,
            body=sampling_fn,
            loop_vars=[sum_priorities, self.priorities, indices, loop_index, num_timesteps]
        )

        // Save batch indices.
        assignment = tf.assign(ref=self.batch_indices, value=indices)

After Change

        // Vectorized sampling.
        sum_priorities = tf.reduce_sum(input_tensor=self.priorities, axis=0)
        sample = tf.random_uniform(shape=(num_priority_elements,),dtype=tf.float32)
        indices = tf.constant(value=0, shape=(num_priority_elements,))

        def cond(loop_index, sample, priorities):
            return tf.reduce_all(input_tensor=(sample <= 0))

Instances