Skip to content

mloopREADME.md

Latest commit

 

History

History
111 lines (74 loc) · 5.21 KB

mloopREADME.md

File metadata and controls

111 lines (74 loc) · 5.21 KB

M LOOP with DexTeR

shady code brought to you by Jonathan

Using PyDex, we can communicate with Dexter in a simple way without ever needing to alter the LabVIEW code. One application of this is to allow for optimisation of the experiment via tools such as mloop

This implementation uses a somewhat convoluted structure, shown is this diagram: mloopflow

How to run

  1. Setup Experiment as per normal

  2. Launch Dexter 1.3 Microscope Edition and JM version of image analyser. Make sure some atoms are present and imaging etc is nominal.

  3. Save the current DExTer sequence to the shared directory, as mloopsequence.xml

  4. Configure the two config files: exp_config.txt is instructions for mloop, i.e. the bounds on parameters mloopdict.json tells python which parameter to vary

  5. Run the mloop.bat file. This will launch the python parts in separate command prompt windows.

  6. Now press python mode on the DExTer terminal. If everything is working you should see a message "Communication Established..." on the TCP box. If you are happy to proceed open the communicator terminal and press enter

  7. M-LOOP will now run the experiment. Once finished safely exit python mode by typing c then enter into the communicator terminal.

Editing mloop behaviour

Change exp_config.txt, and mloopdict.json to edit which variables are being optimised, and over what range.

Parameters

mloopdict.json is a special text file which stores the python dict, and looks like:

{"Param 1": 
    {"type": "timestep", "timestep": 0, "value": 0},
 "Param 2": {"type": "slowanalogue", "timestep": [12, 13, 14], "channelname": "E/W shims (X)", "value": 0}}

Copy this syntax strictly, varying the objects after the ":" to change the type and timestep of your parameter. I have a feeling channelname will be case and space sensitive so be careful...

Currently these two options are the only types of parameter supported. let me know if you can think of another which would be useful. Most obviously Fast analogues should be implemented.

Ranges and optimisation

exp_config.txt Controls M-LOOP. It is important that you keep the layout the same (i.e. the line spaces) and only change things on the right hand side of equals signs. I've highlighted below the most important section:

#Parameter settings
num_params = 2                #number of parameters
min_boundary = [-1,-1]        #minimum boundary
max_boundary = [1,1]          #maximum boundary
first_params = [0.5,0.5]      #first parameters to try
trust_region = 0.4         	#maximum % move distance from best params

#Halting conditions
max_num_runs = 1000                       #maximum number of runs
max_num_runs_without_better_params = 50  
target_cost = 0.01           

To change cost function change mloopoutput.m Currently it is set to maximise atom number.

Note the boundrys are in format min_boundary = [min 1, min2] max_boundary = [max 1,max 2]

The other parts are to do with the algorithm mloop uses, and how it stores data. These are doumented at https://m-loop.readthedocs.io/en/latest/examples.html

Analysing Data

M-LOOP stores data for each run with timestamp based names, If you are interested in analysing a run further copy this data out put it somewhere e.g. on the shared drive.

One can reproduce the mloop plots using the mloop python code as follows

import mloop.visualizations as mlv
import matplotlib.pyplot as plt

mlv.configure_plots()
mlv.create_controller_visualizations('controller_archive_2016-08-23_13-59.mat',file_type='mat')
mlv.create_gaussian_process_learner_visualizations('learner_archive_2016-08-18_12-18.pkl',file_type='pkl')

plt.show()

TODO: Expand this API to produce sensible plots with correct units etc.

Code documentation

M-LOOP

see https://m-loop.readthedocs.io/en/latest/install.html

M-LOOP communicates by writing files in the dir that you run it in. The directory is set by the batch file.

PyDex

mloopcommunicator is a script written in the PyDEX API, which

  1. Sets up a TCP server for communication with Dexter
  2. loops looking for a exp_input from mloop, then reads this using the translator library of pydex
  3. Using translator.mloopmodify() interprets exp_input based on the mloopdict.json configuration file
  4. Sends on a modified DExTer sequence to be run by DExTer

The script contains hardcoded paths which must be correct for things to work.

translator.translate.mloopmodify

Modifies the experimental sequence based on parameters read from mloop txt file, interpreted by mloopdict nested dictionary.

Supports two types of parameter, specified by the 'type' string. This is case sensitive.

  1. 'timestep' : Use the number as the length of a timestep
  2. 'slowanalogue': Use the number as a Voltage. This has aditional arguments "timestep" : list of timestep numbers over which to modify the volatage "channel name" : Name of the slow analouge to modift

MATLAB

The cost function from the analysed images is provided by mloopoutput.m. This uses a path specified in the config for the image analyser. This shouldn't change but is simple to fix if it does.

This script is run immediately after ana_sp1, i.e. the code which analyses the image.