
Q1. I am using a version of Ubuntu that does not have gcc-4.3 in its repository (>11.04), what can I compile the code?
A1: There is full detail procedure on how to install gcc-4.3 on the documentation (/doc folder) and the website of software.

Q2. How can I save the basis vectors as images?
A2. There is an option for gondola called "-show" which saves the basis as images. Please see the documentation of the gondola for more detail.

Q3. How can I rank basis vectors according to their importances in the classification?
A3. We have provided a script called WekaRank. Before using it, meke sure you check its documentation. To use it, you need to have Weka extension as it is specified in the INSTALL.txt

Q4. How can I use the GONDOLA inside of MATLAB? Or, how can use functions inside of MATLAB?
A4. You can use gondola's API inside of MATLAB. Since there are some C/C++ files to compile, first build the software to make sure that the mex file are built. Then, you need to add the ".m" files to your path plus the "/lib" folder in the build folder structure. In addition, you have to make sure that "LIBLINEAR" mex files are also in you path. In case you are using MOSEK should be installed properly (check MOSEK website). 

Q5.How can get probability of the class label instead of just class labels?
A5. The Jython scripts coming with the GONDOLA are capable of producing probability instead of class labels however the default configuration files are set in a way to produce just class labels. You are interested to make probabilities, you need to the following:
    a. After making folder structure and configuration files, make a copy of the big file (the one specifying experiments), let's call it ProbExp_5foldCV.config.
    b. Modify the values in front of  "classifierBinFn" which is [PATH_TO_BUILD_FOLDER]/TwoClass_classifier.py and change it to [PATH_TO_BUILD_FOLDER]/wekaClassifierWithProbability.py.
    c. You need to repeat steps 3,4,5 of the launcher scripts (finding the best parameters, applying on test and train lists and summarization).
   
    

Q6. How can I expedite the method (ie run it faster)?
A6. There are a few ways to make GONDOLA to run faster:
    a. Run GONDOLA in the multi-threaded way. For this you need to make sure that multi-threading is activated during compilation. This feature is OFF by default. To do so, in the cmake toggle the advance options (by pressing "t" in the ccmake environment) and remove " -R -singleCompThread" in the BASIS_MCC_FLAGS.  And re-build GONDOLA.
  
    b. Make sure that there are as few useless values (e.g. air which is zero background around brain images) in the image via appropriate cropping.
    
    c. Use down-sampling the config file. The default value is 2. We do not suggest to increase it any further unless you are dealing with very high-res images are resolution of the basis vectors do not matter to you.
    
    

Q7. How can I get better interpretability?
A7. Basis (ie columns of B) are between zero and one but because if scaling of the data, they barely reach the upper bound of the feasible set (they barely have max 1). I suggest when you overlay basis, set athreshold and make the values below that threshold transparent. You can easily do that using AFNI or MIPAV for medical images.

Q8. How can I choose the parameters?
A8. The default parameters should work in general however if you want to have a good insight about how to choose the parameters, see the TMI papers.

Q9. I don't have "mcc" (MATLAB compiler) , how can I use GONDOLA?
A9. The main goal of compiling GONDOLA with mcc was to make it independent from MATLAB license. If you don't have "mcc", you should still be able to use GONDOLA API's. you should build GONDOLA mex files but not the whole package and the rest is similar to answer to Q4 (see A4).

 
Q10. Is multi-class classification is supported in GONDOLA 1.0?
A10. GONDOLA1.0 does not fully support multi-class although we have a multi-class variant. Please contact me if you are interested (Kayhan Batmanghelich: batmanghelich@gmail.com).

Q11. Can I use GONDOLA with MS WINDOWS?
A11: I used linux (Ubuntu) to develope the software and I have not tried it with WINDOWS. Theoretically you should be able to do that. You can install Python, Jython, Weka, ITK and CMAKE and it should work. I have not tried it but the building procedure should be similar.

Q12. I don't have Sun Grid Engine (SGE) can I still use GONDOLA?
A12. 

Q13. GONDOLA is running, how can I find out which iteration is it?
A13. GONDOLA saves a .log file inside of each fold that reports the latest update: values of the objective function, various terms, iteration number, etc.

Q14. What if servers crashed and it stoped GONDOLA in the middle of the optimization? Do I need to re-run it?
A14. As long as the .mat file is not corrupted, you can continue GONDOLA from the latest save. Notice that if GONDOLA is stopped crashed less than 3hrs before it finished, you need to re-run it from the begining because GONDOLA saves intermediate results every 3hrs (unless it finished that save final results) to avoid overwhelming network file system (NFS) in case of multiple jobs.
