Cellxgene

Added
* Switch between multiple embedding visualizations in the UI (eg, tsne, umap, etc). (776, 766)
* To add guardrails around very large datasets, CLI will generate a warning if a large sparse non-CSC dataset loaded. (723)
* If a specific port is chosen on the CLI (--port), the app will use it and fail if not able to do so. If no port is specified, the application will scan for an unused port number (698, 761)
* Add logo to UI (782)

Changed
* Python anndata version dependency updated to >= 0.6.20, resolving the incompatibility with the most recent SciPy release (see anndata issue 146 for details). Also removed the temporary work-around on scipy version requirements introduced in 0.9.3.
* Updated most javascript package dependencies (765)
* Use anndata to read h5ad files in `launch`, rather than the ScanPy API wrapper around AnnData. Improves startup speed. (763)
* Layout of embedding visualization improved. All points will display upon load, and scaling behavior is improved when the window size changes. (655, 768, 776, 789)
* Pan/zoom scaling is now 1:1 (722)
* Various testing & CI framework improvements (777, 772, 757)
* CLI --debug option will generate an error if used with --port option. If you just want verbose error logging, try --verbose.

Fixed
* Fixed several bugs relating to categorical metadata with large (>100) number of categories. (764, 781, 721, 737)
* Undo/redo bug fix (787)
* Correctly handle datasets which use a column name "name" in `var` or `obs` metadata. (785)

Fixed
* Fixed `ImportError: cannot import name 'IndexMixin'` by excluding conflicting scipy version (770)

Fixed
* Correctly handle column slicing of csr data (754)

Fixed
* __init__.py file missing from new module (748)

Added
* “Clip” to percentile range for continuous data - (metadata and expression) outside of range will be excluded from histograms and grayed out on layout. This improves handling of zero-heavy distributions, or other cases where outliers need to be suppressed (570, 672)
* Prepare calculates and adds QC metrics as to continuous metadata (uses new ScanPy 'calculate_qc_metrics') (697)

Changed
* Documentation updates (691, 706, 649, 702)
* Color-by buttons toggle if clicked twice (693)
* ScanPy requirements updated to 1.3.7(688)
* Server-side refactoring in prep for PyQt & notebook support (711)
* Performance improvements accessing large sparse expression data (719)

Fixed
* Undo/redo improvements - histogram and graph selection now correctly interacts with undo/redo (679, 663)
* Selection of continuous data dimensions containing NaN values would select incorrect cells (715)
* Graph pan/zoom and selection now stay in sync (622)
* Graph selection paints correctly when window resized (203)
* Gene name type-ahead no longer clears text on blur (573)
* Would fail if simplejson was installed in python environment (700)

Added
* Python 3.7 support
* Redo/Undo
* (falexwolf) Support additional graph layouts
* Menu with FAQ, etc

Changed
* When selected embedding has more than two components cellxgene UI truncates to the first two instead of failing

Fixed
* Various UI bug fixes
* Histograms no longer allow tiny/invisible selections