braindynamicsusyd / spikenet Goto Github PK

Say, it takes XXX mins to simulate a network of XXX neurons and XXX synapses with a firing rate of XXX Hz. Just to give the users an intuitive feeling.
Also, point out how the performance scales with various parameters. Firing rate is usually the number 1 concern.

There is an existing protocol to remove a spike.

They are too outdated that I am considering removing them all together.

For some reason it doesn't work, although it works on redhat 6.

brew has migrated hdf5 and autoconf does seem to be up to date.
As a result, the code cannot be run on a mac if following the readme.

I haven’t added these to the simulation-restarting part yet:
https://github.com/BrainDynamicsUSYD/SpikeNet/blob/master/matlab_interface/writeExtConductanceTimeVariantSettingsHDF5.m
https://github.com/BrainDynamicsUSYD/SpikeNet/blob/master/matlab_interface/writeExtCurrentTimeVariantSettingsHDF5.m

This means if you use these two new features together with restarting, shit will happen.

We need a thorough check there if we still want to use the restarting feature

Yuxi reported that the IE ratio outputted by the simulator is not consistent with the result calculated directly from the sampled data.

Not sure why. Need to look into it. Test Welford_online?

• max(I) in synapses should not be larger than N_pre
• max(J) in synapses should not be larger than N_post

Warning: unrecognized data type: > INIT010

In ReadYG (line 361)
In PostProcessYG (line 30)
Warning: unrecognized data type: > INIT011

In ReadYG (line 361)
In PostProcessYG (line 30)
Warning: unrecognized data type: > INIT009

In ReadYG (line 361)
In PostProcessYG (line 30)

No reason for it to be there.

Qi was confused by "logical vector" and actually used logical vector as input to write2ExtSpikeTInvSettingsHDF5().
What it requires is a double vector with 0 and 1s.

Need to fix this with all the files.

Step 7 in 'Run the demo' has an unnecessary '}' at the end of line 2 for anybody following steps exactly/copy-pasting.

The following function is a hack. It opens and closes the input file every time it is called.
This is because the rest of the code uses HDF5 C++ API, but HDF5 C API is needed here.
This may cause excessive overhead.

Solutions:
(1) Use HDF5 C API for the rest of the code (lots of work!) or
(2) Wait for better HDF5 C++ API error handling.

bool SimuInterface::group_exist_HDF5(const string & filename, const string & name){
hid_t file_id = H5Fopen( filename.c_str(), H5F_ACC_RDONLY, H5P_DEFAULT );
//hbool_t check_object_valid = true;
bool r = H5LTpath_valid(file_id, name.c_str(), true);
H5Fclose(file_id);
return r;
}

When multiple non-default parameter definitions are given to Neurons/Synapses, only the first is implemented.

Well, the good news is that this function is now rarely used.

In almost all cases, these two parameters should be identical but there is no check in the code.
Some form of check should be implemented.
A simple solution is to do it in matlab instead of c++

Sounds like a very good suggestion. Need to confirm if reading from text files is indeed taking too much time.

/usr/bin/m4:configure.ac:10: cannot open `m4/ax_lib_hdf5.m4': No such file or directory
autom4te: /usr/bin/m4 failed with exit status: 1

Since the text-based I/O has been removed from C++, it is natural to revamp the HDF5 I/O protocols as suggested by James. Perhaps Guozhang can do this?

Just don't compile with gcc-6.x.x with hdf5-1.8.16
Use gcc-4.x.x

Matlab that is older than 2013a has different h5read error identifier message for the same type of error, which will cause problem.
Not sure if it will be changed in future versions. At least 2013a-2016b have identical ones.
Why the hell did they change the error identifier message for the same function???