uci-carl / carlsim3 Goto Github PK

To use the parameter-tuning interface, users have to make use of 3 or 4 C++ classes, all of which are undocumented.

There is also some (outdated) Javadoc buried in there somewhere for the Java side of the PTI framework. It's only relevant to people who are writing their own ECJ extensions (the User Guide chapter on ECJ covers all users really need to know), but, for the sake of completeness, there might be a way to merge that documentation with our doxygen build if we felt like it.

Chapter 1 has outdated information re: installation as it still mentions the zip file from the CARL website. Update by explaining how to fork/clone the repo, and how to use stable vs master branch.
You can basically copy the installation instructions in the README.md.

ParameterInstances.cpp uses the stoi() function, a C++11 feature that isn't available by default on most Linux distros.

For a while now, we've had a Makefile for building gtest tests for the C++ component of the parameter-tuning interface (PTI).

The Java portion of PTI also has unit tests, however. I've only ever run them through Netbeans myself, but it would be nice to have a Makefile option so it's convenient for users to run the JUnit tests (if they feel so moved).

Ubunto 14.04, Tegra X1 platform. Syetem Query test passed. NVCC version 7.0. CUDA_7.0 samples runs good

Not having the CUDA samples installed will lead to an error at compile time, because we rely on the file helper_cuda.h, which usually resides in /usr/local/cuda/samples/common/inc. Just happened to one user.

We should mention this prominently in User Guide, Chapter 1.

• Need to document about installing ECJ (basically the steps, after downloading the .jar file from the hyperlink, which include unzipping it, using 'make' on it, and moving it to the appropriate directory).
• Need to mention the additional software tools required to support ECJ such as 'openjdk', 'ant' etc. Also mention the software VERSIONS that work and the ones that don't, since this is a very common issue.
• Say how you can verify that ECJ is installed correctly, and if not, what steps to take in order to solve the issue or reinstall ECJ.
• Remove comment about tools/ecj_pti/user.mk since in the current version of CARLsim there is only one 'user.mk' file which already contains the appropriately set ECJ environment variables. But should mention that the variables (specifically $ECJ_JAR) might need to be modified if a different or newer version of ECJ is installed other than ECJ 22 (or else the 'ec package not found' error persists).
• Maybe add some general tips and troubleshooting techniques about download and installation

In the example in User Guide, the first argument of setESTDP is presynaptic group (g1) but described as postsynaptic group.

In the example in Tutorial, the first argument is postsynaptic group.

So which one is right? Presynaptic or postsynaptic?

And if it's postsynaptic group, how can I implement the following model:
There are three neuron groups A/B/C. Neuron group A connects to B, neuron group C also connects to B. How can I use plasticity on A->B synapses, and fix synapses of C->B?

Thanks a lot!

Show off the functionality of the Motion Energy library and how it integrates with CARLsim and the OAT.

It is still not clear how to set up STDP. The User Guide chapter barely mentions the different setSTDP flavors, and the only code snippet that is shown (on DA-STDP) uses the outdated CARLsim::setSTDP version. It is never explained what setSTDP does compared to setESTDP and setISTDP. We should include clear code examples to show which commands to use in which situations.

For the CARLsim team: The example multi-compartment network in projects/compModelTest from an older version of the feat/Compartment branch should be cleaned up and turned into an official Tutorial.

Files are here:
https://github.com/UCI-CARL/CARLsim/tree/7bb2a90fe11e0d22378c3c693d630f095ae51c62/projects/compModelTest

The Regression Suite chapter is a stub. It needs instructions on:

• how to install Google Test (Unix/Windows)
• how to run the tests and verify output (see all green, passing)
• what the different test cases do and how to add a new one

Also, some of this stuff is outdated. For example, we only provide test cases to A) test core functionality of CARLsim. We don't provide test cases to B) test the reproducibility of published results, and C) to benchmark simulation speed within this context. Those are separate things.

It would be nice to have a Tutorial that focuses on spatial receptive fields (e.g., Gaussian) and how to investigate created connectivity patterns with the ConnectionMonitor class of the OAT.

Maybe implement a DOG edge detector as suggested by one of our users...

Using Doxygen on Windows produces an error: Here is the error message when type "doxygen.exe CARLsim.config.doxy" in command prompt

Generating index page...
explicit link request to 'ch1_getting_started' in layout file 'CARLsimLayout.xml' could not be resolved
...

This might be caused by "@ref" in the layout file:

<tab type="usergroup" visible="yes" title="1. Getting Started" url="@ref ch1_getting_started">

Needs to be replaced with:

<tab type="usergroup" visible="yes" title="1. Getting Started" url="\\ref ch1_getting_started">

However, the double-backslash \\ doesn't work in Linux... There we need a single backslash.
Maybe we can have two different layout and config files; one for Windows, one for Unix.

Fix.

The InteractiveSpikeGen class (tools/spike_generators/interactive_spikegen.h) needs documentation and a corresponding section in User Guide Chapter 6, possibly 6.1.4 (make "Generating Custom Spikes" the last subsection of 6.1).

This SpikeGen is similar to PeriodicSpikeGenerator: It produces spike with a certain ISI. However, it's "interactive" because it lets you update the firing rates during the simulation. It also lets you set a different firing rate per neuron.
It also lets you set a "quota" (some number of spikes) to schedule. Each scheduled spike (with appropriate ISI) decrements the quota. Once the quota reaches zero, no more spikes will be scheduled.

Maybe we could also merge this functionality with the existing PeriodicSpikeGenerator...

The PTI makefiles need updated to reflect the new organization of CARLsim3's configuration.

The README for the parameter-tuning interface is outdated and useless. It needs rewritten to provide installation instructions and pointers to the User's Guide/Tutorial.

The ConnectionReader class of the Offline Analysis Toolbox in MATLAB has some undocumented code.

There are a number of methods in the CARLsim API (carlsim/interface/include/carlsim.h) that need documentation.

Since we include gtest as a submodule, building tests is easy and convenient.

But the README suggests that building the tests is as simple as running make test after checkout. This won't work until gtest has been built—that that takes a few steps.

We should update the README to direct users to the gtest README, and to build the gtest library in the designated path.

Need to document the API and functionality of dopamine-modulated STDP in Chapter 5 of the User Guide.

It looks like there is no way of creating setup where single neuron connects to two others, where one synapse is excitatory, while the second one inhibitory. This is due to the fact that connection type for group is always established based on presynaptic neuron.

The PrePostGroupSpikeGen class (tools/spike_generators/pre_post_spikegen.h) needs documentation.

I think it generates spikes with a specific pre-post delay (which is useful for STDP testing), but it is undocumented.
Also, it needs a short subsection in Chapter 6 of the User Guide, possibly subsection 6.1.4 (make "Genearting Custom Spikes" the last subsection of 6.1).