swarm dataframe

[mountaindust]

A swarm is essentially a container for many agents of the same type (for speed purposes), but in this function it is rather rudimentary and not well suited for handling arbitrary agent states, properties, etc. - a vary desirable thing to have in such a container. An elegant solution would be to add a Pandas dataframe property, in which the dataframe holds all attributes of each agent (columns are attributes, rows are agents). A list then holds each dataframe in the time history, much as pos_history does now, with the current-time dataframe updated in place. This would also present a natural solution to #23. Then one can easily look at the dataframe associated with a given time point to quickly see the state of every agent in an organized manner.

[mountaindust]

To save space and reduce complexity, there should also be a dictionary to hold attributes shared by all agents within a swarm, rather than storing N copies of redundant data in a dataframe.

[mountaindust]

This is currently being handled on the dataframe branch

[mountaindust]

Next steps probably need to look like this:

• Create a dict called shared_props to hold user-defined properties that lack individual variation. This allows for significant speed gain when, say, drawing n numbers from the same distribution verses n numbers from n distributions
• Use the dataframe to hold everything else. Also can expand its purpose to hold information like initial condition, exit time, and (upon saving) final position for analyzing data. Would need a save_dataframe method for this, which can be re-implemented by a user for whatever data needs are there
• Create a normal distribution jitter function that handles the task of figuring out whether or not the properties are shared or not and reacts accordingly. this can then be used as an example if someone wants to implement jitter or something else with a different distribution than normal.

[mountaindust]

Even better:

• get_prop handles figuring out where the parameter is, and returns either a single value or an ndarray
• the movement methods are then written to handle both scalar (shared) and vector (differentiated) values

[mountaindust]

This is done on the dataframe branch and is passing tests, but need to make sure examples work.

Also need to add a new test to make sure the individual variation case works.