Demo Data:
Auditory Data:
Raw Data (1.8 MB) (.wav and spike times)Preprocessed Data (3.0 MB) (spectrograms and rasters)
Visual Data:
Raw Data (3.5 MB)Preprocessed Data (14.4 MB)
STRFPAK code:
STRFPAK 5.3 is now available
Version 5.3 has a new checksum feature which uses the built-in md5 function of your Mac or Linux box (as long as ssl can be found in /usr/bin/openssl). This new checksum is faster, less memory-intensive and less likely to generate hash collisions.The documentation has also been cleaned up slightly.
STRFPAK 5.2 has arrived
Version 5.2 has all the new features of version 5.2 beta 2 (see below), plus the following numerical enhancements:- Two improvements have been made in how the autocorrelation
is calculated.
- STRFPAK now automatically decides if it is faster to calculate the autocorrelation with a time-domain or a Fourier-domain method. Fourier-domain methods are faster for STRFs with a large number of latencies, while time-domain methods are faster for STRFs with fewer latencies. STRFPAK thus far has used only Fourier-domain stimulus autocorrelation, but time-domain algorithms are faster when the number of latencies is less than the logarithm of the number of time samples.
- If a Fourier-domain algorithm is faster, STRFPAK now calculates the Fourier transform of each spatial stimulus band only once, then computes the inverse FFT of the product of the pre-computed Fourier transforms of the stimulus, reducing the total number of Fourier transforms needed by a factor of 3.
- STRFPAK now also uses a time-domain algorithm for stimulus-response cross correlation where it is faster.
STRFPAK 5.2 beta 2 is available for download
STRFPAK 5.2 beta 2 is now available. It now has batch mode support for saved subspaces and negative firing rates. Numerical issues have been fixed, and a few GUI improvements have been made over version 5.2 beta.Local Caches
One large improvement over previous versions is that version 5.2 beta 2 caches cell-specific files in a local directory instead of in the main cache directory. If your performance is taking a hit because you have thousands of cached files in a central directory, erase the contents ofyour cache directory and try STRFPAK 5.2 beta 2.Saved subspaces
Message April 10 2007: the numerical flaw in the beta code has been fixed in version 5.2 beta 2.If you have presented stimuli from two different classes (like natural sounds and noise stimuli) and find you have different STRFs, you might want to compare the STRFs to see what stimulus-dependent nonlinearities are in the neuron's transfer function. Until now, you could not be sure if differences in the STRF were due to real nonlinearities or to the fact that stimuli with different statistics cause STRFs to be projected into different subspaces when tolerance values are applied. Now, every time STRFPAK is run, pointers to the cached subspace are saved. Future STRFPAK sessions can choose to project STRFs into the saved subspaces (if desired) so as to apply exactly the same bias to STRFs as in the original run. Select a saved subspace from the "Special options" panel, new to STRFPAK 5.2.
Negative firing rates
Until version 5.2, STRFPAK has assumed that firing rates always have to be positive. This restriction is appropriate for typical STRFPAK runs, but if users choose to use STRFPAK for other purposes, the positive-rate requirement can be disabled in the "Special options" panel, new to STRFPAK 5.2.Beta testing
Please keep me informed as to whether version 5.2 beta 2 works as advertised.STRFPAK 5.1.1 here
Download the latest stable version of STRFPAK: 5.1.1. 5.1.1 contains two bug fixes over 5.1.0: it fixes a bug with calculating separable receptive fields, and it is less sensitive to the format of input spike time .dat and .txt files.STRFPAK 5.1.0 is more flexible and stable than previous versions. In particular, it now handles auditory data with unusual sampling frequencies (like those needed for bat audition, or the 24 414 Hz sampling frequency imposed by some stimulus-presenting rigs) in a better way. Stimuli are now upsampled to the nearest multiple of the PSTH sampling rate to avoid any possible beat effects between these two sampling frequencies.
Here is a point-form list of all the improvements version 5.1.0 has over version 5.0.5:
- Better handling of diverse sound sampling rates
- Preprocessed auditory data now simply in .mat files (download the sample preprocessed files for examples)
- For Windows machines, now STRFPAK is less picky about the directory-separating character (it can be "/" or "\", but "\" is still recommended for maximum backwards compatibility)
- Thanks to Yale, Noopur, Channing, Angela, and Johanna for your feedback, suggestions, and contributed code.
Older Versions of STRFPAK:
Download STRFPAK 4.4.1 , containing a bug fix for version 4.4, which was incompatible with some older MATLAB versions, and has the following improvements over version 4.3:
- Various speed and reliability improvements.
- If you validate, the best STRF is now saved in a file called "best_strf.mat".
- Fixed a bug with "Display StimStat" when using a cache. Now StimStat works properly always.
- Added user-defined function support for 1-D (auditory) stims. Now, you can use your favourite home-grown function (spectrograms, pitch detectors, etc.) natively with STRFPAK. Just make sure
- Your function is in the your path and that it takes just the stimulus as an input (i.e. hard-code all the parameter values you want).
- Your function returns the transformation you want as the first output argument.
- Loading data is faster thanks to:
- Automatic recognition of .wav files and matching spike files in the current directory.
- New hashing protocol makes hashing the transformed stimuli unnecessary.
- If the automatic recognition of .wav files works with your file names, running a cell in batch mode can now be as easy as typing "cd <data directory> ; STRFPAK_script;".
- Made STRFPAK more resilient to unequal numbers of trials for different stimuli.
New since version 4.0.2
It's much faster:
- It allows you to cache results so that you only need to calculate stimulus autocorrelations once, EVER, per stimulus. STRFs for new neurons but familiar stimuli can take about a minute to calculate.
- New numerical algorithms speed up validation by about a factor of 10 and goodness of fit estimation by a factor of 2. When you add a tolerance value, now you don't have to repeat any calculation you have done.
- It can now remember the parameters you specify, so you do not have to select the same options repeatedly.
It's more stable:
- We've fixed a difficulty where some versions of Matlab cannot open the preprocessing window.
- Support for user-processed stimuli is improved.
The STRFPAK_script file works and is easier to use than ever:
- We have a new system which allows anyone familiar with Matlab to analyze many neurons in batch mode once they have run through the GUI version of STRFPAK for a typical neuron. Now, no changes to the STRFPAK_script.m file are needed - just save the locations of the data files to a .mat file in a specified format and STRFPAK_script will work with the new neural data. See the automatically-generated STRFPAK_script.m file after a STRFPAK run for more details.
Download STRFPAK-4.0.2.zip (0.5 MB)
To run this software start MATLAB, add the STRFPAK directory to your path, move to the data directory and at the prompt type:
>> strfpak
STRFPAK Manuals
STRFPAK_4 User Guide [PDF (3.3 MB)] (not totally current)
