added missing artifacts; one test still failing

LucaMarconato · LucaMarconato · commit 3112d1578137 · 2024-12-26T17:53:03.000+01:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -12,166 +12,167 @@ and this project adheres to [Semantic Versioning][].
 
 ### Fixed
 
-- Transformations of Points and Shapes are now applied before rendering with datashader (#378)
+-   Transformations of Points and Shapes are now applied before rendering with datashader (#378)
+-   Fix bug due to `sc.get.obs_df()` returning a different type (#393)
 
 ## [0.2.8] - 2024-11-26
 
 ### Changed
 
-- Support for `xarray.DataTree` (which moved from `datatree.DataTree`) (#380)
+-   Support for `xarray.DataTree` (which moved from `datatree.DataTree`) (#380)
 
 ## [0.2.7] - 2024-10-24
 
 ### Added
 
-- The user can now specify `datashader_reduction` to control the rendering behaviour (#309)
-- Rendering outlines of shapes with datashader works now (#309)
+-   The user can now specify `datashader_reduction` to control the rendering behavior (#309)
+-   Rendering outlines of shapes with datashader works now (#309)
 
 ### Fixed
 
-- datashader now uses canvas size = image size which speeds up the rendering (#309)
-- datashader now uses the `linear` as interpolation method for colormaps instead of the default `eq_hist` to make it equivalent to matplotlib (#309)
-- point sizes of datashader now agree with matplotlib also when dpi != 100 (#309)
-- Giving a custom colormap when rendering a multiscale image now works (#586)
+-   datashader now uses canvas size = image size which speeds up the rendering (#309)
+-   datashader now uses the `linear` as interpolation method for colormaps instead of the default `eq_hist` to make it equivalent to matplotlib (#309)
+-   point sizes of datashader now agree with matplotlib also when dpi != 100 (#309)
+-   Giving a custom colormap when rendering a multiscale image now works (#586)
 
 ## [0.2.6] - 2024-09-04
 
 ### Changed
 
-- Lowered RMSE-threshold for plot-based tests from 45 to 15 (#344)
-- When subsetting to `groups`, `NA` isn't automatically added to legend (#344)
-- When rendering a single image channel, a colorbar is now shown (#346)
-- Removed `percentiles_for_norm` parameter (#346)
-- Changed `norm` to no longer accept bools, only `mpl.colors.Normalise` or `None` (#346)
+-   Lowered RMSE-threshold for plot-based tests from 45 to 15 (#344)
+-   When subsetting to `groups`, `NA` isn't automatically added to legend (#344)
+-   When rendering a single image channel, a colorbar is now shown (#346)
+-   Removed `percentiles_for_norm` parameter (#346)
+-   Changed `norm` to no longer accept bools, only `mpl.colors.Normalise` or `None` (#346)
 
 ### Fixed
 
-- Filtering with `groups` now preserves original cmap (#344)
-- Non-selected `groups` are now not shown in `na_color` (#344)
-- Several issues associated with `norm` and `colorbar` (#346)
+-   Filtering with `groups` now preserves original cmap (#344)
+-   Non-selected `groups` are now not shown in `na_color` (#344)
+-   Several issues associated with `norm` and `colorbar` (#346)
 
 ## [0.2.5] - 2024-08-23
 
 ### Changed
 
-- Replaced `outline` parameter in `render_labels` with alpha-based logic (#323)
-- Lowered RMSE-threshold for plot-based tests from 60 to 45 (#323)
-- Removed `preprocessing` (.pp) accessor (#329)
+-   Replaced `outline` parameter in `render_labels` with alpha-based logic (#323)
+-   Lowered RMSE-threshold for plot-based tests from 60 to 45 (#323)
+-   Removed `preprocessing` (.pp) accessor (#329)
 
 ### Fixed
 
-- Minor fixes for several tests as a result of the threshold change (#323)
+-   Minor fixes for several tests as a result of the threshold change (#323)
 
 ## [0.2.4] - 2024-08-07
 
 ### Added
 
-- Added utils function for 0-transparent cmaps (#302)
+-   Added utils function for 0-transparent cmaps (#302)
 
 ### Fixed
 
-- Took RNG out of categorical label test (#306)
-- Performance bug when plotting shapes (#298)
-- scale parameter was ignored for single-scale images (#301)
-- Changes to support for dask-expr (#283)
-- Added error handling for non-existent elements (#305)
-- Specifying vmin and vmax properly clips image data (#307)
-- import bug `get_cmap()` (8fd969c)
+-   Took RNG out of categorical label test (#306)
+-   Performance bug when plotting shapes (#298)
+-   scale parameter was ignored for single-scale images (#301)
+-   Changes to support for dask-expr (#283)
+-   Added error handling for non-existent elements (#305)
+-   Specifying vmin and vmax properly clips image data (#307)
+-   import bug `get_cmap()` (8fd969c)
 
 ## [0.2.3] - 2024-07-03
 
 ### Added
 
-- Datashader support for points and shapes (#244)
+-   Datashader support for points and shapes (#244)
 
 ### Changed
 
-- All parameters are now provided for a single element (#272)
+-   All parameters are now provided for a single element (#272)
 
 ### Fixed
 
-- Fix color assignment for NaN values (#257)
-- Zorder of rendering now strictly follows the order of the render_x calls (#244)
+-   Fix color assignment for NaN values (#257)
+-   Zorder of rendering now strictly follows the order of the render_x calls (#244)
 
 ## [0.2.2] - 2024-05-02
 
 ### Fixed
 
-- Fixed `fill_alpha` ignoring `alpha` channel from custom cmap (#236)
-- Fix channel str support (#221)
+-   Fixed `fill_alpha` ignoring `alpha` channel from custom cmap (#236)
+-   Fix channel str support (#221)
 
 ## [0.2.1] - 2024-03-26
 
 ### Minor
 
-- Adjusted GitHub worklows
+-   Adjusted GitHub worklows
 
 ## [0.2.0] - 2024-03-24
 
 ### Added
 
-- Support for plotting multiple tables @melonora
+-   Support for plotting multiple tables @melonora
 
 ### Fixed
 
-- Several bugfixes, especially for colors and palettes @melonora
+-   Several bugfixes, especially for colors and palettes @melonora
 
 ## [0.1.0] - 2024-01-17
 
 ### Added
 
-- Multiscale image handling: user can specify a scale, else the best scale is selected automatically given the figure size and dpi (#164)
-- Large images are automatically rasterized to speed up performance (#164)
-- Added better error message for mismatch in cs and ax number (#185)
-- Beter test coverage for correct plotting of elements after transformation (#198)
-- Can now stack render commands (#190, #192)
-- The `color` argument in render_shapes/points now accepts actual colors as well (#199)
-- Input arguments are now evaulated for their types in basic.py (#199)
+-   Multiscale image handling: user can specify a scale, else the best scale is selected automatically given the figure size and dpi (#164)
+-   Large images are automatically rasterized to speed up performance (#164)
+-   Added better error message for mismatch in cs and ax number (#185)
+-   Beter test coverage for correct plotting of elements after transformation (#198)
+-   Can now stack render commands (#190, #192)
+-   The `color` argument in render_shapes/points now accepts actual colors as well (#199)
+-   Input arguments are now evaulated for their types in basic.py (#199)
 
 ### Fixed
 
-- Now dropping index when plotting shapes after spatial query (#177)
-- Points are now being correctly rotated (#198)
-- User can now pass Colormap objects to the cmap argument in render_images. When only one cmap is given for 3 channels, it is now applied to each channel (#188, #194)
+-   Now dropping index when plotting shapes after spatial query (#177)
+-   Points are now being correctly rotated (#198)
+-   User can now pass Colormap objects to the cmap argument in render_images. When only one cmap is given for 3 channels, it is now applied to each channel (#188, #194)
 
 ## [0.0.6] - 2023-11-06
 
 ### Added
 
-- Pushed `get_extent` functionality upstream to `spatialdata` (#162)
+-   Pushed `get_extent` functionality upstream to `spatialdata` (#162)
 
 ## [0.0.5] - 2023-10-02
 
 ### Added
 
-- Can now scale shapes (#152)
-- Can now plot columns from GeoDataFrame (#149)
+-   Can now scale shapes (#152)
+-   Can now plot columns from GeoDataFrame (#149)
 
 ### Fixed
 
-- Multipolygons are now handled correctly (#93)
-- Legend order is now deterministic (#143)
-- Images no longer normalised by default (#150)
-- Filtering of shapes and points using the `groups` argument is now possible, coloring by palette and cmap arguments works for shapes and points (#153)
-- Colorbar no longer autoscales to [0, 1] (#155)
-- Plotting shapes after a spatial query is now possible (#163)
+-   Multipolygons are now handled correctly (#93)
+-   Legend order is now deterministic (#143)
+-   Images no longer normalised by default (#150)
+-   Filtering of shapes and points using the `groups` argument is now possible, coloring by palette and cmap arguments works for shapes and points (#153)
+-   Colorbar no longer autoscales to [0, 1] (#155)
+-   Plotting shapes after a spatial query is now possible (#163)
 
 ## [0.0.4] - 2023-08-11
 
 ### Fixed
 
-- Multi-scale images/labels are now correctly substituted and the action is logged (#131).
-- Empty geometries among the shapes can be handeled (#133).
-- `outline_width` parameter in render_shapes is now a float that actually determines the line width (#139).
+-   Multi-scale images/labels are now correctly substituted and the action is logged (#131).
+-   Empty geometries among the shapes can be handeled (#133).
+-   `outline_width` parameter in render_shapes is now a float that actually determines the line width (#139).
 
 ## [0.0.2] - 2023-06-25
 
 ### Fixed
 
-- Multiple bugfixes of which I didn't keep track of.
+-   Multiple bugfixes of which I didn't keep track of.
 
 ## [0.0.1] - 2023-04-04
 
 ### Added
 
-- Initial release of `spatialdata-plot` with support for `images`, `labels`, `points` and `shapes`.
+-   Initial release of `spatialdata-plot` with support for `images`, `labels`, `points` and `shapes`.
diff --git a/docs/contributing.md b/docs/contributing.md
@@ -78,6 +78,16 @@ out to the developers of the dependency before the package is released to a wide
 
 [scanpy-test-docs]: https://scanpy.readthedocs.io/en/latest/dev/testing.html#writing-tests
 
+### Testing the correctness of the plots
+
+Many tests will produce plots and check that they are correct by comparing them with a previously saved and serialized version of the same plots. The ground truth images are located in `tests/_images`. Different OS/versions may produce similar but not identical plots (for instance the ticks/padding could vary). To take into account for this please consider the following:
+
+- you should not use locally generated plots as ground truth images, but you should commit images that have been generated by a GitHub Action. The recommended workflow is to go to the ["actions" page for the repo](https://github.com/scverse/spatialdata-plot/actions/workflows/test.yaml), download the artifacts, and upload them as ground truth (after having reviewed them).
+- the ground truth images need to be updated when a new test is passing, or when a test starts producing a slightly different (but consistent) plot.
+- please never replace the ground truth images without having manually reviewed them.
+- if you run the tests locally in macOS or Windows they will likely fail because the ground truth images are generated using Ubuntu. To overcome this you can use `act`, which will generate a Docker reproducing the environment used in the GitHub Action. After the Docker container is generated you can use it within IDEs to run tests and debug code.
+- in the case of PyCharm, it is easier to create a container from a `Dockerfile` instead of using `act`. Please in such case use the `Dockerfile` made availabel in the repository. If you encountering problems with `act` or `docker`, please [get in touch with the developers via Zulip](https://scverse.zulipchat.com/#narrow/channel/443514-spatialdata-dev) and we will help troubleshoot the issue. See also additional details [here](https://github.com/scverse/spatialdata-plot/pull/397).
+
 ## Publishing a release
 
 ### Updating the version number
diff --git a/src/spatialdata_plot/pl/basic.py b/src/spatialdata_plot/pl/basic.py
@@ -950,7 +950,8 @@ def show(
 
                     if wanted_labels_on_this_cs:
                         if (table := params_copy.table_name) is not None:
-                            colors = sc.get.obs_df(sdata[table], params_copy.color)
+                            assert isinstance(params_copy.color, str)
+                            colors = sc.get.obs_df(sdata[table], [params_copy.color])
                             if isinstance(colors[params_copy.color].dtype, pd.CategoricalDtype):
                                 _maybe_set_colors(
                                     source=sdata[table],
diff --git a/src/spatialdata_plot/pl/render.py b/src/spatialdata_plot/pl/render.py
@@ -453,7 +453,8 @@ def _render_points(
     )
 
     if col_for_color is not None:
-        cols = sc.get.obs_df(adata, col_for_color)
+        assert isinstance(col_for_color, str)
+        cols = sc.get.obs_df(adata, [col_for_color])
         # maybe set color based on type
         if isinstance(cols[col_for_color].dtype, pd.CategoricalDtype):
             _maybe_set_colors(
diff --git a/test.Dockerfile b/test.Dockerfile
@@ -0,0 +1,48 @@
+# Dockerfile for Testing Plot Correctness
+#
+# This Dockerfile is used to create a consistent testing environment for
+# verifying the correctness of plot outputs against ground truth images.
+# The ground truth images are stored in `tests/_images` and are generated
+# using a GitHub Action on Ubuntu. This ensures consistency across different
+# environments, as local differences in OS or library versions can cause
+# slight variations in plot rendering (e.g., ticks or padding).
+
+# Define a build argument for the target platform.
+# Default is set to linux/amd64 for x86_64 machines.
+ARG TARGETPLATFORM=linux/amd64
+
+# Use the specified platform to pull the correct base image.
+# Override TARGETPLATFORM during build for different architectures, such as linux/arm64 for Apple Silicon.
+# For example, to build for ARM64 architecture (e.g., Apple Silicon),
+# use the following command on the command line:
+#
+#     docker build --build-arg TARGETPLATFORM=linux/arm64 -t my-arm-image .
+#
+# Similarly, to build for the default x86_64 architecture, you can use:
+#
+#     docker build --build-arg TARGETPLATFORM=linux/amd64 -t my-amd64-image .
+#
+FROM --platform=$TARGETPLATFORM ubuntu:latest
+LABEL authors="Luca Marconato"
+
+ENV PYTHONUNBUFFERED=1
+ENV MPLBACKEND=agg
+
+WORKDIR /spatialdata-plot
+COPY . /spatialdata-plot
+
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+    build-essential \
+    python3-venv \
+    python3-dev \
+    git \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN python3 -m venv /opt/venv
+ENV PATH="/opt/venv/bin:$PATH"
+RUN pip install --upgrade pip wheel
+
+RUN pip install -e ".[dev,test]"
+
+CMD ["pytest", "-v", "tests/pl/test_render.py"]
diff --git a/tests/_images/Points_points_categorical_color_column_matplotlib.png b/tests/_images/Points_points_categorical_color_column_matplotlib.png
diff --git a/tests/_images/Points_points_continuous_color_column_datashader.png b/tests/_images/Points_points_continuous_color_column_datashader.png
diff --git a/tests/_images/Points_points_continuous_color_column_matplotlib.png b/tests/_images/Points_points_continuous_color_column_matplotlib.png
diff --git a/tests/pl/test_render_points.py b/tests/pl/test_render_points.py
@@ -5,7 +5,6 @@
 import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
-import pytest
 import scanpy as sc
 from anndata import AnnData
 from spatialdata import SpatialData, deepcopy
@@ -112,13 +111,17 @@ def test_plot_datashader_continuous_color(self, sdata_blobs: SpatialData):
             element="blobs_points", size=40, color="instance_id", alpha=0.6, method="datashader"
         ).pl.show()
 
-    @pytest.mark.parametrize("method", ["matplotlib", "datashader"])
-    def test_plot_points_categorical_color_column(self, sdata_blobs: SpatialData, method: str):
-        sdata_blobs.pl.render_points("blobs_points", color="genes", method=method).pl.show()
+    def test_plot_points_categorical_color_column_matplotlib(self, sdata_blobs: SpatialData):
+        sdata_blobs.pl.render_points("blobs_points", color="genes", method="matplotlib").pl.show()
 
-    @pytest.mark.parametrize("method", ["matplotlib", "datashader"])
-    def test_plot_points_continuous_color_column(self, sdata_blobs: SpatialData, method: str):
-        sdata_blobs.pl.render_points("blobs_points", color="instance_id", method=method).pl.show()
+    def test_plot_points_categorical_color_column_datashader(self, sdata_blobs: SpatialData):
+        sdata_blobs.pl.render_points("blobs_points", color="genes", method="datashader").pl.show()
+
+    def test_plot_points_continuous_color_column_matplotlib(self, sdata_blobs: SpatialData):
+        sdata_blobs.pl.render_points("blobs_points", color="instance_id", method="matplotlib").pl.show()
+
+    def test_plot_points_continuous_color_column_datashader(self, sdata_blobs: SpatialData):
+        sdata_blobs.pl.render_points("blobs_points", color="instance_id", method="datashader").pl.show()
 
     def test_plot_datashader_matplotlib_stack(self, sdata_blobs: SpatialData):
         sdata_blobs.pl.render_points(

Original file line number	Diff line number	Diff line change
`@@ -453,7 +453,8 @@ def _render_points(`
`453`	`453`	`)`
`454`	`454`
`455`	`455`	`if col_for_color is not None:`
`456`		`- cols = sc.get.obs_df(adata, col_for_color)`
	`456`	`+ assert isinstance(col_for_color, str)`
	`457`	`+ cols = sc.get.obs_df(adata, [col_for_color])`
`457`	`458`	`# maybe set color based on type`
`458`	`459`	`if isinstance(cols[col_for_color].dtype, pd.CategoricalDtype):`
`459`	`460`	`_maybe_set_colors(`