Skip to main content

Tools for neuroscience experiments

Project description

toon

image image

Description

Additional tools for neuroscience experiments, including:

  • A framework for polling input devices on a separate process.
  • A framework for keyframe-based animation.
  • High-resolution clocks.

Everything should work on Windows/Mac/Linux.

Install

Current release:

pip install toon

Development version (requires compilation-- C++11 (and if using MSVC, >= 2015 for proper std::chrono implementation)):

pip install git+https://github.com/aforren1/toon

See the demos/ folder for usage examples (note: some require additional packages).

Overview

Input

toon provides a framework for polling from input devices, including common peripherals like mice and keyboards, with the flexibility to handle less-common devices like eyetrackers, motion trackers, and custom devices (see toon/input/ for examples). The goal is to make it easier to use a wide variety of devices, including those with sampling rates >1kHz, with minimal performance impact on the main process.

We use the built-in multiprocessing module to control a separate process that hosts the device, and, in concert with numpy, to move data to the main process via shared memory. It seems that under typical conditions, we can expect single read() operations to take less than 500 microseconds (and more often < 100 us). See demos/bench_plot.py for an example of measuring user-side read performance.

Typical use looks like this:

from toon.input import MpDevice
from mydevice.mouse import Mouse
from timeit import default_timer

device = MpDevice(Mouse())

with device:
    t1 = default_timer() + 10
    while default_timer() < t1:
        res = device.read()
        # alternatively, unpack immediately
        # time, data = device.read()
        if res:
            time, data = res # unpack (or access via res.time, res.data)
            # N-D array of data (0th dim is time)
            print(data)
            # 1D array of times
            print(time)

Creating a custom device is relatively straightforward, though there are a few boxes to check.

from ctypes import c_double

class MyDevice(BaseDevice):
    # optional: give a hint for the buffer size (we'll allocate 1 sec worth of this)
    sampling_frequency = 500

    # this can either be introduced at the class level, or during __init__
    shape = (3, 3)
    # ctype can be a python type, numpy dtype, or ctype
    # including ctypes.Structures
    ctype = c_double

    # optional. Do not start device communication here, wait until `enter`
    def __init__(self):
        pass

    ## Use `enter` and `exit`, rather than `__enter__` and `__exit__`
    # optional: configure the device, start communicating
    def enter(self):
        pass

    # optional: clean up resources, close device
    def exit(self):
        pass

    # required
    def read(self):
        # See demos/ for examples of sharing a time source between the processes
        time = self.clock()
        # store new data with a timestamp
        data = get_data()
        return time, data

This device can then be passed to a toon.input.MpDevice, which preallocates the shared memory and handles other details.

A few things to be aware of for data returned by MpDevice:

  • If there's no data for a given read, None is returned.
  • The returned data is a copy of the local copy of the data. If you don't need copies, set use_views=True when instantiating the MpDevice.
  • If receiving batches of data when reading from the device, you can return a list of (time, data) tuples.
  • You can optionally use device.start()/device.stop() instead of a context manager.
  • You can check for remote errors at any point using device.check_error(), though this automatically happens after entering the context manager and when reading.
  • In addition to python types/dtypes/ctypes, devices can return ctypes.Structures (see input tests or the example_devices folder for examples).

Animation

This is still a work in progress, though I think it has some utility as-is. It's a port of the animation component in the Magnum framework, though lacking some of the features (e.g. Track extrapolation, proper handling of time scaling).

Example:

from math import sin, pi

from time import sleep
from timeit import default_timer
import matplotlib.pyplot as plt
from toon.anim import Track, Player
# see toon/anim/easing.py for all available easings
from toon.anim.easing import LINEAR, ELASTIC_IN

class Circle(object):
    x = 0
    y = 0

circle = Circle()
# list of (time, value)
keyframes = [(0.2, -0.5), (0.5, 0), (3, 0.5)]
x_track = Track(keyframes, easing=LINEAR)

# we can reuse keyframes
y_track = Track(keyframes, easing=ELASTIC_IN)

player = Player(repeats=3)

# directly modify an attribute
player.add(x_track, 'x', obj=circle)

def y_cb(val, obj):
    obj.y = val

# modify via callback
player.add(y_track, y_cb, obj=circle)

t0 = default_timer()
player.start(t0)
vals = []
times = []
while player.is_playing:
    t = default_timer()
    player.advance(t)
    times.append(t)
    vals.append([circle.x, circle.y])
    # sleep(1/60)

plt.plot(times, vals)
plt.show()

Other notes:

  • Non-numeric attributes, like color strings, can also be modified in this framework (easing is ignored).
  • The Player can also be used as a mixin, in which case the obj argument can be omitted from player.add() (see the demos/ folder for examples).
  • Multiple objects can be modified simultaneously by feeding a list of objects into player.add().

Utilities

The util module includes high-resolution clocks/timers via std::chrono::steady_clock. The class is called MonoClock, and an instantiation called mono_clock is created upon import. Usage:

from toon.util import mono_clock, MonoClock

clk = mono_clock # re-use pre-instantiated clock
clk2 = MonoClock(relative=False) # time relative to whenever the system's clock started

t0 = clk.get_time()

Another utility currently included is a priority function, which tries to improve the determinism of the calling script. This is derived from Psychtoolbox's Priority (doc here). General usage is:

from toon.util import priority

res = priority(1)
if not res:
    raise ValueError('Failed to raise priority.')

# ...do stuff...

priority(0)

The input should be a 0 (no priority/cancel), 1 (higher priority), or 2 (realtime). If the requested level is rejected, the function will return False. The exact implementational details depend on the host operating system. All implementations disable garbage collection.

Windows

  • Uses SetPriorityClass and SetThreadPriority/AvSetMmMaxThreadCharacteristics.
  • level = 2 only seems to work if running Python as administrator.

MacOS

  • Only disables/enables garbage collection; I don't have a Mac to test on.

Linux

  • Sets the scheduler policy and parameters sched_setscheduler.
  • If level == 2, locks the calling process's virtual address space into RAM via mlockall.
  • Any level > 0 seems to fail unless the user is either superuser, or has the right capability. I've used setcap: sudo setcap cap_sys_nice=eip <path to python> (disable by passing sudo setcap cap_sys_nice= <path>). For memory locking, I've used Psychtoolbox's 99-psychtoolboxlimits.conf and added myself to the psychtoolbox group.

Your mileage may vary on whether these actually improve latency/determinism. When in doubt, measure! Read the warnings here.

Notes about checking whether parts are working:

Windows

  • In the task manager under details, right-clicking on python and mousing over "Set priority" will show the current priority level. I haven't figured out how to verify the Avrt threading parts are working.

Linux

  • Check mlockall with cat /proc/{python pid}/status | grep VmLck
  • Check priority with top -c -p $(pgrep -d',' -f python)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

toon-0.15.1.zip (328.2 kB view hashes)

Uploaded Source

Built Distributions

toon-0.15.1-cp38-cp38-win_amd64.whl (221.5 kB view hashes)

Uploaded CPython 3.8 Windows x86-64

toon-0.15.1-cp38-cp38-manylinux2010_x86_64.whl (1.1 MB view hashes)

Uploaded CPython 3.8 manylinux: glibc 2.12+ x86-64

toon-0.15.1-cp38-cp38-manylinux1_x86_64.whl (1.1 MB view hashes)

Uploaded CPython 3.8

toon-0.15.1-cp38-cp38-macosx_10_9_x86_64.whl (221.3 kB view hashes)

Uploaded CPython 3.8 macOS 10.9+ x86-64

toon-0.15.1-cp37-cp37m-win_amd64.whl (220.9 kB view hashes)

Uploaded CPython 3.7m Windows x86-64

toon-0.15.1-cp37-cp37m-manylinux2010_x86_64.whl (1.0 MB view hashes)

Uploaded CPython 3.7m manylinux: glibc 2.12+ x86-64

toon-0.15.1-cp37-cp37m-manylinux1_x86_64.whl (1.0 MB view hashes)

Uploaded CPython 3.7m

toon-0.15.1-cp37-cp37m-macosx_10_9_x86_64.whl (221.4 kB view hashes)

Uploaded CPython 3.7m macOS 10.9+ x86-64

toon-0.15.1-cp36-cp36m-win_amd64.whl (212.7 kB view hashes)

Uploaded CPython 3.6m Windows x86-64

toon-0.15.1-cp36-cp36m-manylinux2010_x86_64.whl (952.4 kB view hashes)

Uploaded CPython 3.6m manylinux: glibc 2.12+ x86-64

toon-0.15.1-cp36-cp36m-manylinux1_x86_64.whl (952.4 kB view hashes)

Uploaded CPython 3.6m

toon-0.15.1-cp36-cp36m-macosx_10_9_x86_64.whl (214.0 kB view hashes)

Uploaded CPython 3.6m macOS 10.9+ x86-64

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page