README

Model and analysis code corresponding to Ritvo, Nguyen, Turk-Browne & Norman (2023).

These models have been developed based on the Emergent framework, developed primarily by the CCN lab at UCDavis. In our simulations, we use our own variant of Emergent, which can be accessed in the neurodiff_leabra and neurodiff_emergent repositories.

• Steps to Installing and Running the Simulations
  1. Setting up your GOPATH and GOROOT
  2. Install go
  3. Install neurodiff_simulations
     • Cloning the repository: Run git clone https://github.com/PrincetonCompMemLab/neurodiff_simulations.git
     • Build the simulation (do this for chanales / favila / schlichting simulations)
       • e.g. cd into emergent-differentiation/chanales
       • run go build
     • Run the simulations:
       • run ./main
  4. Reproducing the plots

Assumptions

• OS: We can guarantee performance on Unix-based platforms such as MacOS and Linux. Performance on Windows operating systems is not guaranteed.
• Any version starting 1.13 should work. We used Go 1.18
• To run our analyses, we use the Slurm scheduling system.

File structure:

You need to set the GOPATH and GOROOT.

• GOROOT is where the go executable is
• GOPATH is where to install go packages.

Note that go forces the separation of GOPATH and GOROOT, so they can't be the same directory. This README will assume you are using a folder named go/ for GOROOT and gocode/ for for GOPATH.

1. Add the following lines to ~/.bashrc

   export GOROOT="PATH/TO/CODE/go"
   export GOPATH="PATH/TO/CODE/gocode"
   export PATH=$GOROOT/bin:$PATH
   

   • replace "PATH/TO/CODE/" with the paths to the actual paths of your go installs

2. Make sure to run:

   source ~/.bashrc
   

Installing Go

Download Go

The first step to setting up emergent is to download Go. Pick your favored binary release here, download it, and run it. The MSI installers for Windows and Mac do all the work for you.

You should download versions 1.13 or later.

Downloading Go on the command line

To download go using the command line, use wget and then the download link. Make sure to download the tar.gz file. For example, if you are downloading Go 1.14 for Linux OS, you can run

wget https://dl.google.com/go/go1.14.1.linux-amd64.tar.gz

The above command downloads a .tar.gz file wherever you ran the command from. You probably want to run it from /PATH/TO/CODE, or else move the tar.gz file there after you download it.

Install Go

The command wget https://dl.google.com/go/go1.14.1.linux-amd64.tar.gz downloaded a .tar.gz file wherever you ran it. Move it to where you want the GOPATH directory to be (using mv). Then unzip the file:

tar -xzf go1.14.1.linux-amd64.tar.gz

That should have created a folder called go/. Change the name to be whatever you called the GOPATH directory (i.e. gocode/).

Then make the directory for the GOROOT:

mkdir PATH/TO/CODE/go

Now you should have two folders that match GOPATH and GOROOT

Test Your Go Installation

You might want to make sure you installed Go successfully. To do this:

1. Create a folder $GOROOT/src/hello

   1. Then create a file named hello.go with the following content:

   package main
   
   import "fmt"
   
   func main() {
   	fmt.Printf("hello, world\\n")
   }
   
   

2. 'Build' the script by executing the command:

   go build
   
   

   within your hello folder. This creates a hello.exe file in your directory that you can then run to execute the code.

3. Run hello.exe.

• If the command returns hello, world, you're golden. If not, try running echo $PATH to see if GOROOT and GOPATH were added correctly.

Install neurodiff_simulations

• Clone the github repo:
  • Run git clone https://github.com/PrincetonCompMemLab/neurodiff_simulations.git
• Build the simulation (do this for chanales / favila / schlichting simulations)
  • cd into neurodiff_simulations
  • run go build
• Run the simulations:
  • run ./main

Running the simulations

Reproducing the plots

• Figures 4, 8, 10: The figs.py script in each simulation directory will generate figures corresponding to the simulations of the Chanales et al., Favila et al., and Schlichting et al. studies
  • In each directory, run python figs.py output_folder_name --data_dir=/PATH/TO/OUTPUT_FOLDER/
    • Replace /PATH/TO/OUTPUT_FOLDER/ and output_folder_name with the path to store the output folder and the name of the output folder, respectively
    • This applies for all figs.py files
    • figs.py will create figures in this directory: /PATH/TO/OUTPUT_FOLDER/output_folder_name/results
    • If data_dir is not specified, the script will save the figures in a subdirectory of the current directory called ./figs
• Figure 6B:
  • In each simulation directory, run python figs.py output_folder_name --data_dir=/PATH/TO/OUTPUT_FOLDER/ --searchvar=LRateOverAll
    • figs.py will create figures in this directory: /PATH/TO/OUTPUT_FOLDER/output_folder_name/results
• Figure 6A:
  • figs.py will create a plot of the learning rate vs. within-pair correlation before and after learning for each condition (e.g. 0/6, 1/6, 2/6, 3/6, 4/6, and 5/6 are the learning conditions of the Chanales et al. study)
  • You can make figs.py make this plot for each learning condition by doing the following:
    • For the Chanales et al. simulation, add the flag --HiddNumOverlapUnits=i where i is in [0,1,2,3,4,5] to the line lines[-4] = f"{analyze_only}./main --mode=batch --saveDirName={saveDirName} --runs={num_exps} --trncyclog=false --tstcyclog=false {cmd_string} \n" in figs.py
    • For the Favila et al. simulation, add the flag --same_diff_flag=i where i is in ['Same', 'Different'] to the line lines[-4] = f"{analyze_only}./main --mode=batch --saveDirName={saveDirName} --runs={num_exps} --trncyclog=false --tstcyclog=false {cmd_string} \n" in figs.py
      • the flag is case sensitive !
    • For the Schlichting et al. simulation, add the flag --blocked_interleave_flag=i where i is in ['Blocked', 'Interleave'] to the line lines[-4] = f"{analyze_only}./main --mode=batch --saveDirName={saveDirName} --runs={num_exps} --trncyclog=false --tstcyclog=false {cmd_string} \n" in figs.py
  • run python figs.py output_folder_name --data_dir=/PATH/TO/OUTPUT_FOLDER/ --searchvar=LRateOverAll