dgreenberg/sbmvivo

Sequential binding model for AP inference

sbmvivo's Issues

GPU memory allocation issues

Hi David,

I'm trying to run inference on one of your in vivo data segments. Do you have a sense how much GPU memory is needed for that? I am getting the following error message:

Error using cuda5s_mex_init
GPU memory allocation failed: error code:

Are there any parameters that I can adjust to trade-off available GPU memory against compute time?

Best,

-Peter

Warning and error related to number of unique samples

sum.infer throws the following series of warnings and errors:

Total data length of 20000 exceeds maximum length of 12500, using partial data from 1 segments.

initial MC std.: 459.202203, tempering distribution with temperature 1452.124869
Warning: preliminary sampling phase did not complete
In aimh_gk (line 160)
In est_params_temperedmode (line 74)
In estimate_parameters (line 72)
In sbm.run_alg (line 30)
In sbm.infer (line 49)
Error using gmfit_gk (line 14)
at least five unique samples required

Error in aimh_gk (line 165)
gstar = gmfit_gk(samples);

Error in est_params_temperedmode (line 74)
[samples, targetlogpvals, acceptance_ratio, accepted, gstar, extradata, first_update,
preliminary_phase_end] = aimh_gk(targetlogpfunc, proposal, p0, temperature,
opts.max_iter_paramest, displayfunc, box_constraints);

Error in estimate_parameters (line 72)
[Ppe,prevPpe,prevlikpe] = est_params_temperedmode(Vpe, Ppe, dtpe, Fpe, nA2Dpe);

Error in sbm.run_alg (line 30)
[P, prevP, prevlik] = estimate_parameters(V, P, dt, F, nA2D);

Error in sbm.infer (line 49)
[M, params, opts] = sbm.run_alg(f, dt, indicator, opts, params, [], [], nA2D);

• Can segments with more than 12500 points not be processed?
• What does it mean that the preliminary sampling phase has not been completed?
• Rerunning the same command on the same data several times results in varying numbers of unique_samples, ranging for 1-6 in my hands. Every time the unique number of samples is <5, I get the above error. Please advise.

-Peter

sbm.infer crashes because 'moment' not defined

sbm.infer attempts to access a struct array (?) named 'moment', which is not defined:

References to this variable appear only in lines 59 and 61 of sbm.infer:

59   stepsize = dt(s) / moments(s).nsteps;
60    st_steps = M(s).gst.mu / stepsize - 1;  % get the times of non-discretized detected APs in simulation  steps. zero is exactly one frame (dt) before the first fluorescence measurement
61    st_frames = st_steps / moments(s).nsteps;  % get the times of non-discretized detected APs in frames. zero is exactly one frame (dt) before the first fluorescence measurement

The error message is:

Undefined variable "moments" or class "moments".

Error in sbm.infer (line 59)
stepsize = dt(s) / moments(s).nsteps;

Where is moments(:).nsteps initialized?

-Peter

MEX file compilation on windows and mac

In attempt to run sbm.infer() I get the following:
Undefined function or variable 'quickrand'.
Error in sample_initial_states (line 16)
randseed = quickrand(urand, randseed);