Latest ideas

[kushalkolar]

We're using improv to handle process & memory management. Visualization data will be sent to jupyter using zmq.

Need to write and decide the following:

1. Acquisition information. Probably just a json dict that is submitted via zmq from matlab after the improv setup is invoked. We want:

   • mesmerize batch dir or database name
   • animal_id
   • channel information
   • shape of each channel
   • dtype
   • framerate
   • laser power
   • zoom
   • all other scanimage info
   • date

2. Per-frame acquisition header of size 5,000 bytes with each frame that is sent via zmq from matlab/scanimage:

   1. frame_index - uint32
   2. trial_index - uint32
   3. trigger_state - uint32
   4. timestamp - float32
   5. Many empty bytes for future use We could just define these using a yaml file or something.

Note: 5,000 byte header is less than 1% the size of (512 512 2) bytes which is the size of a 512 x 512 int16 frame.

3. Matlab class using parfeval to send data using zmq. The frame data and per-frame acquisition information are sent together in a single zmq message as raw bytes in this format:

   • header: 5,000 bytes
   • channel_a: shape_x shape_y dtype.nbytes
   • channel_b: shape_x shape_y dtype.nbytes

4. MesmerizeWriter to write incoming data as a "batch item"

5. How to keep the cnmf Estimates object and read from it during live visualization.

[kushalkolar]

zmq brainstorming

[kushalkolar]

Probably better to use a store for:

1. Storing acquisition metadata so it can be accessed by all actors
2. Storing UUID and params for this acquisition