contact_map.frequency_task

Task-based implementation of ContactFrequency.

The overall algorithm is:

1. Identify how we’re going to slice up the trajectory into task-based chunks (block_slices(), default_slices())

2. On each node

   1. Load the trajectory segment (load_trajectory_task())

   2. Run the analysis on the segment (map_task())

3. Once all the results have been collected, combine them (reduce_all_results())

Notes

Includes versions where messages are Python objects and versions (labelled with _json) where messages have been JSON-serialized. However, we don’t yet have a solution for JSON serialization of MDTraj objects, so if JSON serialization is the communication method, the loading of the trajectory and the calculation of the contacts must be combined into a single task.

Functions

block_slices(n_total, n_per_block)	Determine slices for splitting the input array.
default_slices(n_total, n_workers)	Calculate default slices from number of workers.
load_trajectory_task(subslice, file_name, …)	Task for loading file.
map_task(subtrajectory, parameters)	Task to be mapped to all subtrajectories.
map_task_json(subtrajectory, parameters)	JSON-serialized version of map_task()
reduce_all_results(contacts)	Combine multiple ContactFrequency objects into one
reduce_all_results_json(results_of_map)	JSON-serialized version of reduce_all_results()