v2.3.0

What's Changed

• ENH: Remove deprecated --use-feature=in-tree-build
• ENH: Versions as strings, not numbers, so that 3.1 != 3.10, etc.
• COMP: Use numpy<1.24 so that np.bool* used by other packages does not give errors
• STYLE: Simplify study dictionary keys
• COMP: Use current versions of flake8, itk, pytest
• ENH: Bump version to 2.3.0
• ENH: Make predictions with wrapped model
• ENH: Implement CreateTorchDataloader
• ENH: Spin off read_large_image functionality from CreateTensorFlowDataset
• ENH: Remove Tensorflow from framework-agnostic functionality
• ENH: Spin off ChunkLocations functionality from CreateTensorFlowDataset
• ENH: Add tensorflow (cpu & gpu) as dependency
• STYLE: Format Python with black -C
• BUG: Upper bound of mask_threshold is 1.0 not 1.1
• ENH: Demonstrate predictions in pytorch_stream.ipynb
• STYLE: Simplify study dictionary keys
• STYLE: Shorten variable names
• ENH: Support both deprecated and current keys in study_description
• ENH: Concatenate tensorflow.Datasets more performantly
• DOC: Simplify study dictionary creation
• ENH: Record overlap_width and overlap_height in study dictionary
• ENH: Add get_tiles method to get just tiles from a study dictionary
• PERF: Make _py_read_chunk be @staticmethod
• PERF: Support num_workers and worker_index in CreateTensorFlowDataset
• PERF: More efficiently create input to tf.data.Dataset.from_tensor_slices
• PERF: Use large_image_source_tiff to read a chunk from disk
• PERF: Use N log N approach for splitting tiles into chunks
• STYLE: Remove unused code for tensorflow.data.Dataset.from_generator
• PERF: Avoid large-memory intermediates when building one_chunk_per_slice
• PERF: Have tensorflow.py_function implicitly convert response to a tensor
• PERF: Create frequently used tensorflow.constant values only once
• PERF: Convert some class methods to @staticmethod
• PERF: Replace some for loops with Python comprehensions
• BUG: Limit private_threadpool_size to 1 to keep Tensorflow from freezing on GPU
• ENH: update pooch links - pytorch
• BUG: Fix confusion between mask width and height
• PERF: Use numpy/scipy/itertools instead of loops/comprehensions for masks
• COMP: Placate Python 3.6, which cannot handle dtype=(np.int64, (2,))
• COMP: Install Python package large_image[tiff]

New Contributors

• @cooperlab made their first contribution in #114

Full Changelog: v2.2.4...v2.3.0