nibabel 2.0.0

This release had large contributions from Eric Larson, Brendan Moloney, Nolan Nichols, Basile Pinsard, Chris Johnson and Nikolaas N. Oosterhof.

• New feature, bugfix release with minor API breakage;
• Minor API breakage: default write of NIfTI / Analyze image data offset value. The data offset is the number of bytes from the beginning of file to skip before reading the image data. Nibabel behavior changed from keeping the value as read from file, to setting the offset to zero on read, and setting the offset when writing the header. The value of the offset will now be the minimum value necessary to make room for the header and any extensions when writing the file. You can override the default offset by setting value explicitly to some value other than zero. To read the original data offset as read from the header, use the offset property of the image dataobj attribute;
• Minor API breakage: data scaling in NIfTI / Analyze now set to NaN when reading images. Data scaling refers to the data intercept and slope values in the NIfTI / Analyze header. To read the original data scaling you need to look at the slope and inter properties of the image dataobj attribute. You can set scaling explicitly by setting the slope and intercept values in the header to values other than NaN;
• New API for managing image caching; images have an in_memory property that is true if the image data has been loaded into cache, or is already an array in memory; get_data has new keyword argument caching to specify whether the cache should be filled by get_data;
• Images now have properties dataobj, affine, header. We will slowly phase out the get_affine and get_header image methods;
• The image dataobj can be sliced using an efficient algorithm to avoid reading unnecessary data from disk. This makes it possible to do very efficient reads of single volumes from a time series;
• NIfTI2 read / write support;
• Read support for MINC2;
• Much extended read support for PAR / REC, largely due to work from Eric Larson and Gregory R. Lee on new code, advice and code review. Thanks also to Jeff Stevenson and Bennett Landman for helpful discussion;
• parrec2nii script outputs images in LAS voxel orientation, which appears to be necessary for compatibility with FSL dtifit / fslview diffusion analysis pipeline;
• Preliminary support for Philips multiframe DICOM images (thanks to Nolan Nichols, Ly Nguyen and Brendan Moloney);
• New function to save Freesurfer annotation files (by Github user ohinds);
• Method to return MGH format vox2ras_tkr affine (Eric Larson);
• A new API for reading unscaled data from NIfTI and other images, using img.dataobj.get_unscaled(). Deprecate previous way of doing this, which was to read data with the read_img_data function;
• Fix for bug when replacing NaN values with zero when writing floating point data as integers. If the input floating point data range did not include zero, then NaN would not get written to a value corresponding to zero in the output;
• Improvements and bug fixes to image orientation calculation and DICOM wrappers by Brendan Moloney;
• Bug fixes writing GIfTI files. We were using a base64 encoding that didn't match the spec, and the wrong field name for the endian code. Thanks to Basile Pinsard and Russ Poldrack for diagnosis and fixes;
• Bug fix in freesurfer.read_annot with orig_ids=False when annot contains vertices with no label (Alexandre Gramfort);
• More tutorials in the documentation, including introductory tutorial on DICOM, and on coordinate systems;
• Lots of code refactoring, including moving to common code-base for Python 2 and Python 3;
• New mechanism to add images for tests via git submodules.