R-Forge Project: Reverse correlation image classification - News

rcicr development moved to github

rdotsch@users.r-forge.r-project.org (Ron Dotsch) — Mon, 09 Jan 2017 12:14:31 GMT

The rcicr development repository has moved to Github. See here: https://github.com/rdotsch/rcicr

rcicr 0.3.4.1 on CRAN now

rdotsch@users.r-forge.r-project.org (Ron Dotsch) — Wed, 13 Jul 2016 13:02:51 GMT

A new version of rcicr is available on CRAN now. It fixes some minor bugs and more importantly: it adds several features, like a new noise type (Gabor), a setting to change the number of spatial scales of the noise, and a new default scaling method, called independent scaling.

Here's the full list of changes since version 0.3.0:

NEW FEATURES:
* Added independent scaling. Works like matched scaling but maintains the zero point. Is the new recommended default for generating a single base image.
* Added Gabor noise option. Generate stimuli with generateStimuli2IFC(..., noise_type='gabor'). Almost all functions have been affected in refactoring.
* Added option to specify number of spatial scales
* Added computeCumulativeCICorrelation. Computes cumulative trial CIs correlations with final/target CI. Use for instance for plotting
* Added label parameter to batchGenerateCI, which adds custom identifiable labels to jpegs
* Added targetpath parameter to CI generation functions, so CIs will not by default be placed in working directory, but in ./cis subdirectory
* Added batchGenerateCI to deal with data from any reverse correlation task that does not use inverse noise. This function can be used for a 1IFC, 2IFC with independently generated stimuli, 3IFC, etc.

CHANGES:
* Changed output of autoscale: scaling factor is no longer part of the list of classification images that is returned
* Changed progressbar to dplyr progress bar, no longer dependent on tcltk (so no more dependency on xquartz on OS X). Also adds time estimation.
* Suppressed warning messages if directory for stimulus or ci output already exists.

FIXES:
* Bugfix: images not saved as jpegs when scaling is set to anything other than autoscale in batchGenerateCI and batchGenerateCI2IFC
* Bugfix: When using pre 0.3.0 Rdata file, the parameter matrix will now be truncated to the appropriate numbers of parameters
* Bugfix: NA's in by parameter of batchGenerateCI2IFC() get ignored instead of making R crash. Thank you Elze Ufkes for reporting this bug!
* Bugfix: target path error message for generateCI2IFC()
* Fixed targetci bug (should be empty list by default instead of boolean) in computeCumulativeCICorrelation
curves of trial-final/target CI correlations to estimate how many trials are necessary in your task
* Bugfix in generateCINoise: added possibility of just one stimulus/response

REFACTORING:
* Renamed generateCI2IFC to generateCI and moved it to rcicr.R as it is a general function, and not just for 2IFC reverse correlation tasks
* Created generateCI2IFC wrapper function (wraps generateCI) for backwards compatibility
* Renamed old generateCI to generateCINoise to differentiate between generateCI and generateCINoise

Ron

rcicr 0.3.0 submitted: fixes important bug

rdotsch@users.r-forge.r-project.org (Ron Dotsch) — Fri, 23 Jan 2015 13:25:49 GMT

Dear rcicr users,

I just submitted rcicr version 0.3.0 to CRAN. It's already available for download as development version. I strongly recommend to update immediately if you are planning on generating new reverse correlation stimuli.

Pre 0.3.0 versions of rcicr had a small bug that caused at least one and probably two sinusoids of the 4092 sinusoid patches not to be rendered (for sure one in the highest spatial frequency and perhaps one in the lowest spatial frequency). The visual effect of this bug is practically unnoticeable, but of course it had to be fixed.

rcicr 0.3.0 is backwards compatible: if you install it, you will be able to analyze data collected with stimuli created pre 0.3.0 without any problems. The package will automatically treat the data appropriately (which means that if you generated stimuli with a pre 0.3.0 version, the latest version will show the same behavior as would the pre 0.3.0 version in the analysis).

I am working on including the following new features in future versions:
- Support for multiple images per trial
- Support for multiple response options per trial
- Support for Gabor and Gaussian noise in addition to sinusoid noise
- Computing pixel-wise correlations between CIs
- Masking stimuli and CIs
- Support for pixel and cluster tests

These features should all be implemented some time during the summer break. If you have any additional feature requests, please let me know!

Best wishes,

Ron

rcicr 0.2.3 supports batch generation of CIs

rdotsch@users.r-forge.r-project.org (Ron Dotsch) — Fri, 29 Aug 2014 16:21:19 GMT

Freshly submitted to R-Forge. You can now conveniently use the batchGenerateCI2IFC function to generate CIs for multiple subjects or conditions.

Quick instructions on using rcicr posted

rdotsch@users.r-forge.r-project.org (Ron Dotsch) — Sun, 03 Aug 2014 12:01:24 GMT

Follow this link: http://ron.dotsch.org/rcicr/

rcicr 0.2.1 now available on CRAN

rdotsch@users.r-forge.r-project.org (Ron Dotsch) — Sun, 03 Aug 2014 11:59:29 GMT

rcicr 0.2.1 is now available on CRAN (OS X version for Mavericks still awaiting to be built though).