0.13.0

What's Changed

• Introduce finalize_dag method in Plan that caches DAG optimization by @tomwhite in #340
• Add VirtualInMemoryArray that keeps small arrays in memory by @tomwhite in #336
• Fix Modal test following #336 by @tomwhite in #341
• Change the DAG to have separate nodes for operations and arrays by @tomwhite in #337
• Allow more control over optimizations applied by @tomwhite in #342
• Handle the case where an array is created by calling map_blocks with no input arrays by @tomwhite in #343
• Add a test for visualizing the quad means computation by @tomwhite in #345
• Show nbytes for each array and total on visualization by @tomwhite in #344
• Fuse multiple inputs by @tomwhite in #346
• Upgrade to sphinx-notes/pages v3 by @tomwhite in #347
• Minor improvements to DAG visualization by @tomwhite in #348
• Remove deprecated code by @tomwhite in #349
• Add generated doc for primitive and core ops to design page by @tomwhite in #351
• Support arbitrary block index functions in blockwise by @tomwhite in #350
• Refactor pipeline to reduce coupling between runtime and primitive layers by @tomwhite in #352
• Implement stack using general_blockwise by @tomwhite in #353
• Implement reshape_chunks using general_blockwise by @tomwhite in #354
• More control over operation fusion by @tomwhite in #355
• Minor memory fixes by @tomwhite in #358
• Use Dask's 'broadcast trick' to save memory for single-valued arrays by @tomwhite in #359
• Add a show_hidden option to visualize by @tomwhite in #360
• Reinstate slow tests by @tomwhite in #361
• Add function_nargs to BlockwiseSpec by @tomwhite in #363
• Add missing api doc by @tomwhite in #364
• Implement partial_reduce and tree_reduce using generalized blockwise by @tomwhite in #365
• Quad means validation test by @tomwhite in #367
• Take allowed_mem into account when fusing primitive operations by @tomwhite in #366
• Fuse operations with different numbers of tasks by @tomwhite in #368
• Show number of tasks prominently on graph visualization by @tomwhite in #371
• Introduce event objects for all callback methods by @tomwhite in #373
• Minor Modal improvements by @tomwhite in #375
• Run tests using JAX as the backend array API (on CPU) by @tomwhite in #374
• Introduce max_total_num_input_blocks as a heuristic to control fusion by @tomwhite in #376
• Rename array_name to name in runtime following #337 by @tomwhite in #377
• Enable backup tasks by default for all executors that support them by @tomwhite in #378
• Remove unused array_names parameter in executors by @tomwhite in #380
• Introduce compute_id for each computation by @tomwhite in #382
• Add a Rich progress bar by @tomwhite in #383
• Change timeline visualization default format to SVG by @tomwhite in #384
• Preserve laziness in fuse_multiple to avoid materializing array blocks in _partial_reduce by @tomwhite in #386
• Fix Modal unit tests following change in #382 by @tomwhite in #387
• Add debug log statements to optimizer to see why nodes are fused by @tomwhite in #388
• Allow optional path in from_zarr and to_zarr by @tomwhite in #391
• Remove CUBED_MODAL_FORCE_BUILD env var … by @tomwhite in #393
• Add introductory slides to docs by @tomwhite in #394
• Use donfig to control default spec by @tomwhite in #392
• small changes to coiled/aws examples by @rsignell in #369
• Skip vecdot test due to zero-size array failures by @tomwhite in #399
• Show both operation ID and name in visualization by @tomwhite in #396
• Add support for minimum_workers in Coiled by @tomwhite in #398
• Create a single set of example scripts that can run on any executor by @tomwhite in #395

New Contributors

• @rsignell made their first contribution in #369

Full Changelog: 0.12.0...0.13.0