Would you mind describing in more details the part of the error message you find incomprehensible? It seems that you have provided threshold=3 and it is telling you you 0. < threshold <= 2.

I will go into more detailed version of this input type, what happens when user gave a list of 4D imgs. But as far as I see it should be a single 4D Nifti like object not a list of objects. If that's the case, I agree its incomprehensible error message. I will check it anyway.

Do you also have problem with single 4D image ?

All this makes sense, but will be suprising to many people who have done neuroimaging with other tools: they are used to equating 'thresholds' and 'statistical thresholds', while the concept of threshold used here is different, something like a 'selection threshold'. Maybe we should give a more explicit naming.

'ratio', 'selection_ratio' ?

Just to share some information about the usage of threshold and its strategies:

regions extraction based on statistical thresholds can be used based upon selecting thresholding_strategy='img_value' which will act according to given threshold in statistical value provided that the value should not exceed the max.

regions extraction based on selection threshold can be done upon selecting thresholding_strategy='ratio_n_voxels' or thresholding_strategy='percentile' where threshold should be given upon the n_maps provided should not exceed more than n_maps as stated in the error message.

In this particular error case, selecting thresholding_strategy='img_value' should work if one is trying to threshold upon their statistical values or otherwise leaving to defaults should also work.

We had naming problem before. Since, we have two different strategies, fixing one name from current threshold to ratio or selection_ratio will again be a conflict to statistical thresholding strategy. Let me know if I am wrong.

'cutoff'?

I feel 'cutoff' than 'threshold' is even more difficult to reach to users.

I am thinking of fixing this issue by rewriting error message as below:

threshold=3.0 provided to select for number of nonzero voxels across total number of maps (n_maps) is not valid. Please provide valid number between 0. and n_maps=2

What do you think ? @dohmatob @lesteve @GaelVaroquaux @AlexandreAbraham @banilo

'number of nonzero voxels across total number of maps (n_maps)' is still unclear to me, because when you say "number of nonzero voxels" people will think "maybe several hundreds (or thousands)".
IMHO the semantics of this parameters would be clearer if it were between 0 and 1, as it would represent a fraction of the population, which is easier to explain. I have to say, it is really hard to figure out what is actually done on the fit().

I have to say, it is really hard to figure out what is actually done on the fit().

Is it not clear in terms of documentation or implementation ? I am not sure which part is hard to figure ?

IMHO the semantics of this parameters would be clearer if it were between 0 and 1, as it would represent a fraction of the population, which is easier to explain.

It is easier to explain but it defeats the purpose of the method presented here. The main advantage of this thresholding approach is that you can chose a value corresponding to a certain level of sparsity in the brain (1 if you want to avoid overlap, up to 3 if overlaps are OK for you) and then, no matter how many maps you provide, your atlas will always "look" the same, no adjustment needed. When dealing with a lot of atlases, I find it very convenient.

This makes sense, but then we need a serious refactoring of the documentation to clarify this, and ideally an example dedicated to region extraction that illustrates the impact of these parameters.

On 13/01/2016 10:06, KamalakerDadi wrote:

I have to say, it is really hard to figure out what is actually
done on the fit().

Is it not clear in terms of documentation or implementation ? I am not
sure which part is hard to figure ?

—
Reply to this email directly or view it on GitHub
#907 (comment).

This means 3 things:

• The code does not run a clear operation on the data, like
  thresholding, mean computation or extraction of connected components or
  even a watershed analysis.
• Handling it with a multi-variate point of view (n_maps > 1) is also
  non-classical. I'm not aware of any such procedure apart from that one.
• I am currently trying to run it on other data to build an intuition,
  and do not get it.
  Of course I could take time and read the code, but I believe that
  user-level functions should be better explained.
  HTH

• Handling it with a multi-variate point of view (n_maps > 1) is also
  non-classical. I'm not aware of any such procedure apart from that one.

Right, but that's exactly what the object is there for.

It should probably be named different, with a name that stresses that it
is multi-maps.

ping @GaelVaroquaux @AlexandreAbraham @lesteve

I would like to know your opinion and ideas to bring code to more user friendly and understandable. I am working on to propose some idea. Meanwhile, any ideas or suggestions would be really helpful and great.