Alejandro de la Vega

Hey @soichih, thanks for checking in. Yes, we still are, and that's great to hear. It just requires digging into some libraries I don't maintain to optimize them. If the...

@effigies I tried `gunzip`ing the nifti files used by neuroscout to test the memory mapping, but unfortunately memory consumption still peaked at 8GB. does that seem like the right way...

I've done some memory profiling on nistats. Memory usage on a ~398M compressed nifti fMRI run with a design matrix with two predictors peaks at 5.5 GB. There are two...

So to be more clear, it's just in `nb.load(img).get_data()` that we get up to 3GB. Any suggestions nibabel guru @effigies? It's a bit confusing to me how much smaller the...

It's wrapped up in the check_niimg function. I've tested it alone and it has the same memory footprint. On Thu, Nov 7, 2019, 9:43 AM Chris Markiewicz wrote: > Where's...

I think that gets called and is high in memory, but just calling nb.load_image().get_data() has the same footprint so I'm not sure it's avoidable. On Thu, Nov 7, 2019, 10:09...

I tried it both ways: ``` Line # Mem usage Increment Line Contents ================================================ 8 122.8 MiB 122.8 MiB @profile 9 def run(): 10 122.8 MiB 0.0 MiB data_dir =...

Header .nii.gz ``` object, endian='

Okay, looks like one issue was that the `run_img` was being cast to `float64` rather than `float32`. Force recasting to `float32` prior to nistats (thanks @effigies), gets us down to...

hey @PeerHerholz circling back to this. If we're going to include this, let's first include it into pliers. Any advice on "standard" vs tf version? I assume only the latter...