- Multiple input and multiple output (as opposed to functions where inputs and outputs are always synchronised)
- Less memory (because of streaming)
MiMo is a multi-input multi-output Python streaming library. It allows users to define a stream with multiple inputs and multiple outputs and run them completely from beginning to end. Back-pressure has also been implemented to prevent too much memory from being used.
There are two core components in MiMo; the
Stream and the
Workflow. Streams to the computational processing but do not handle how the data is passed between streams. Workflows pass the data around streams but do no processing of their own. To create a workflow, the user needs to implement streams and pipe them together.
Implementing a stream can be done through inheriting a sub-class from the
Stream class or creating a
Stream class with a custom function as the
fn parameter. The following code shows two implementations of a stream that will produce the numbers from 0 to 99.
from mimo import Stream # Method 1 (inheritance) class MyStream(Stream): IN =  OUT = ['entity'] async def run(self, ins, outs): for item in iter(range(100)): await outs.entity.push(item) # Method 2 (constructor) my_stream = Stream(outs=['entity], fn=my_stream_fn) async def my_stream_fn(ins, outs, state): for item in iter(range(100)): await outs.entity.push(item)
There are a few things to note about the
1. It must be asynchronous, ie. it must be defined wth the
async def keywords.
2. It takes two parameters,
outs, that contain the input streams and the output streams. The names of the input and output streams are defined by the
OUT member variables or overridden using the
outs of the initialisation function. Accessing the input and output streams can be done through the attributes. From the example above, accessing the
entity output stream can be done with
3. Input streams can be popped and peeked and this must be done using the
await keyword. Input streams haven't been used in the above example, but the entities in the stream can be accessed one at a time with the functions
peek. Popping an entity will remove it from the input stream, and peeking will look at the top-most entity without removing it from the stream. Input streams can also be iterated using the
async for looping construct.
4. Output streams can be pushed and must also use the
await keyword. Pushing an entity to an output stream will make it available to any connected downstream streams.
Workflows are created by piping streams together. First a workflow must be instantiated and populated with the desired streams. The steps returned by populating a workflow can then be used to make the connections between the streams using the
pipe function. The function returns the stream being piped to, so
pipe calls can be chained. The workflow can be run by calling the
from mimo import Stream, Workflow def main(): workflow = Workflow() step1 = workflow.add_stream(Stream(outs=['a'])) step2 = workflow.add_stream(Stream(['b'], ['c'])) step3 = workflow.add_stream(Stream(['d'])) step1.pipe(step2).pipe(step3) print(str(workflow)) workflow.start() if __name__ == '__main__': import sys sys.exit(main())