Skip to content

Latest commit

 

History

History
158 lines (127 loc) · 8.23 KB

README.md

File metadata and controls

158 lines (127 loc) · 8.23 KB
Raw

Eye Candy

To use, please open Google Chrome 67 or later. As of Dec 2019, you'll need to go to chrome://flags and then enable Experimental Web Platform features [1].

You also need Docker. Clone this repository to a local folder, cd or chdir into it, and execute:

docker-compose build
docker-compose up

Now open 0.0.0.0:3000 in Chrome. be sure to open a 0.0.0.0:3000/stimulus.html window too. Try scrolling to the bottom and hitting start.

[1] This is needed for using ES6 modules in Web workers. Eventually, this will be merged into Chrome and thus no longer required.

About the Eye-Candy Programming Language

Programs are written in the Eye-Candy programming language (EPL), a domain specific language that is embedded in javascript, and designed from the ground up for asynchronous stimulation. The core idea is that an EPL program is a generator function that runs on the server & yields a new stimulus each time it is called. This passes control to the client, which displays the stimulus using GPU-accelerated HTML5 canvas. The results of the stimulation are then returned to the server, which passes the values into the EPL program and returns with the next stimulus. Thus, EPL is designed from the ground-up for closed-loop experiments.

Here's a trivial EPL program:

const metadata = {name: "test", version: "0.1.0"}

function* stimulusGenerator() {
    yield new Wait(1)
    yield new Solid(2, "white")
    yield new Wait(1)
}

More sophisticated example programs can be found in web/programs.

Supported stimuli

A variety of stimulation types are included. These are all defined in web/src/epl/types.

  • Wait (black screen)
  • Solid (full-field flash of any valid HTML color)
  • Bar (single moving square-bar)
  • Chirp (full-field frequency-swept cosine)
  • ChirpAmplitude (full-field amplitude-swept cosine)
  • Grating (square-wave moving grating)
  • SinusoidalGrating (sinusoidal-wave moving grating)
  • Checkerboard
  • Letter (using ETDRS optotype)
  • TiledLetter (repeated ETDRS optotype that files full-field)
  • EyeChart (mimics changing focus on an ETDRS eyechart)
  • Image (display arbitrary images, either pre-rendered or served from the server's /data/eye-candy/images)
  • Video (display video from mounted volume, by defualt this is in /data/eye-candy/videos, see docker-compose)

Deterministic randomness

Programs are initialized with a seed value for the Mersenne Twister, a deterministic pseudo-random number generator. Thus, each time an experiment is run random parameters can be randomly sampled, and by saving the seed we can re-run the exact same experiment.

Machine-readable lab notebooks

yaml-format lab notebooks are automatically generated to ensure reproducible experiments. Here's an example:

---
affiliation: University of Washington Medicine
lab: Van Gelder Lab
people: 'Tyler Benster, Darwin Babino'
purpose: 'Testing eye-candy'
mouseNumber: '1'
species: mouse
mouseType: 'wt'
birthdate: '2018/01/01'
orientation: '0 deg'
dissectionTime: '9:30am'
solution: 'AMES'
retinaWeight: 'bent wire'
meaType: 60MEA200/30iR-ITO

displayMode: video
experimentNumber: '1'
filename: 'R1_E1_eyecandy_test'
seed: '121658968'
program: test
epl: |
  const metadata = {name: "test", version: "0.1.0"}

  let stimuli = []
  let meta = {}

  stimuli.push(new Wait(1, meta))
  stimuli.push(new Solid(2, "white", meta))
  stimuli.push(new Wait(1, meta))


  function* stimulusGenerator() {
      for (s of stimuli) {
          yield s
      }
  }
windowHeight: '328'
windowWidth: '500'
date: 2019-12-02T22:24:41.549Z
version: 0.5
flickerVersion: 0.3
gitSHA: 5c5dee7c891db64a58c8e2785fb6c2a3f71da061

Since we have the seed, windowHeight, windowWidth, and EPL progam, we can recreate the exact stimuli used during the experiment.

encoding stimulus timing with ephys data

flicker.html denotes the timing of every single frame using a change in light intensity. Every frame it oscillates, and the mean amplitude of the oscillation changes according to modulo 3 of the stimulusIndex. In other words, by using a light detector we can decode the timing of each frame as well as the start of each stimulus--and if there is a dropped frame, we can impute missing stimulus start times thanks to the encoding of which stimulusIndex is associated with each frame.

Pre-rendering

Rendering can often be computationally expensive, especially when generating millions of random numbers is required. To allow for pre-rendering complex stimuli like random binary masks that are balanced over time such that each pixel sums to the same value, EPL supports two special objects: preRenderFunc and preRenderArgs. These objects are used on the client (the webbrowser) to render frames that are then cached in the browser using IndexedDB. Thus, EPL programs execute both on the server and on the client. Cached frames can be reused between sessions as preRenderFunc is memoized: two sessions with identical preRenderArgs as well as canvas height and width for stimulus.html will retrieve the previously computed pre-rendered frames from IndexedDB. See web/programs/eyechart-saccade.js or web/programs/celltyping.js for complex examples.

Clearing the cache can be done in Developer Tools (ctrl-shift-c) -> Storage -> IndexedDB -> eyeCandyDB -> right click myStore -> Clear.

saving video of protocol

Some analysis techniques require access to the pixel values for each frame during the experiment. This is supported by clicking save-video instead of start. Eye-candy uses the MediaStream Recoring API to create a .mp4 file of displayed stimuli + saves a frame log with the timing of each frame, as recorded by the client web-browser. Note that currently, this save dialog pops up on the stimulus.html page, wherein at the end of an experiment, you will be prompted to save a .mp4 file as well as a .log (or .frames) file.

How to add new protocol

Two options: for quick testing you can use the custom dropdown selection, or you can create web/src/epl/programs/myProtocol.js. You need to reload index.html for the dropdown to display myProtocol.js, but once it's there, any changes are hot-loaded by the server when you click load.

Dev notes

How to add a new stimulus

  1. add to types.js
  2. add to actions.js
  3. add to dispatchers.js
  4. add to reducers.js
  5. add to static/js/render.js

Or by using Image type, can just define in EPL preRenderFunc! See eyechart-saccade.js for an example.

Handshake overview:

  • Client stimulus.html POST '/hello'
    • gets SID & stores in localStorage b/c cookies get HTTPOnly flag set for no apparent reason and we need to use SID in socket connection, too.
  • Client stimulus.html POST '/window'
    • sets windows[sid] (keeps track of height and width)
  • Client index.html socket 'load'
    • compiles js to program[sid] in vm
  • Server socket emit 'pre-render'
    • The server cannot render, so asks the client to pre-render
  • Client socket emit 'renderResults'
    • client returns pre-render results
  • Client index.html POST '/start-program'
    • client informs server of start program button press
  • Server socket emit 'run'
  • stimulus.js renderLoop does its thing

TODOs

  • replace koa-socket-session with something better supported (1.2.0 fails)
  • update to koa-socket-2
  • Chunk number of stimmuli returned by server to lower number of requsets
  • fix black frame between binary static by buffering one frame & first swapping canvas on rAF and then rendering next frame to off-frame

video

  • Always save video
  • support client-side image specification

BUGS

  • Stimulus should change on global timer (no increasing drift from estimate, make sure not adding frame on stim switch?)

minified src:

https://github.com/ramda/ramda/blob/master/dist/ramda.min.js