plot-io 0.8.0__tar.gz → 0.12.0__tar.gz

plot_io-0.12.0/PKG-INFO

@@ -0,0 +1,215 @@
+Metadata-Version: 2.5
+Name: plot_io
+Version: 0.12.0
+Summary: Plots for PDF, EPS, SVG, PNG, JPEG, JPEG XL and AVIF, drawn with vec_io
+Project-URL: Repository, https://github.com/ajgouhier/plot_io
+Author-email: Arthur Gouhier <ajgouhier@gmail.com>
+License-Expression: BSD-3-Clause
+License-File: LICENSE
+Requires-Python: >=3.11
+Requires-Dist: vec-io<0.8,>=0.7
+Provides-Extra: dev
+Requires-Dist: mypy; extra == 'dev'
+Requires-Dist: ruff; extra == 'dev'
+Requires-Dist: ziglang==0.15.1; extra == 'dev'
+Description-Content-Type: text/markdown
+
+# plot_io
+
+Plotting library in Python. No dependency other than [vec_io](https://github.com/ajgouhier/vec_io).
+
+* **Familiar:** a figure of plots; `scatter`, `line`, `fill`, `box`, `heatmap`, `bar`,
+  `hist`, `kde` (1-D and 2-D), `ecdf`, `violin`, `strip`, `swarm`, `area`, `errorbar`, `stem`,
+  `regression`, `hist2d`, `hexbin`, `contour`, `pie`, `dots`; reference lines, spans, text and
+  annotations; `color`, `marker`, `opacity`, `limits`, `scale`, `legend`.
+* **Statistical helpers:** `qq`, `raincloud`, `volcano`, `ma`, `manhattan`, `dendrogram`
+  plot types; significance brackets; point labels placed without overlaps; hierarchical
+  clustering (`plt.linkage`, scipy's methods); `plt.joint`, `plt.pairs` and `plt.clustermap`
+  figures in one call.
+* **Units:** lengths in px, pt, in, mm, `em`, `%` of the axes or of another setting, data
+  units and durations (`width="1.5in"`, `tick_length="0.5em"`, `width="60%"`,
+  `cap_length="12h"`); `var()` and `calc()` relate settings (`"calc(var(--gap) * 2)"`).
+* **One way to set things:** keywords at creation or `set(...)` on any object; figure
+  settings (font, color maps, line width…) are inherited by plots and series.
+* **Layouts:** an explicit grid of ids with spans, or automatic rows and columns; plots
+  sized individually; axis lines (or plot borders) aligned across the figure; margins from
+  the real text metrics; shared axes; secondary axes; facets of a table.
+* **Titles, legends and colorbars on any side:** several titles per plot; titles,
+  legends and colorbar keys shared by any rectangle of plots, nested as headers;
+  `share_legend` for one legend per row, column or figure; markup (`<i>`, `<b>`,
+  `<sup>`, …) in any text.
+* **Data:** columns of any table by name (`data=`: dicts, pandas, polars, numpy), one
+  series per group (`group=`), numbers given as colors or sizes, point density (scipy's
+  `gaussian_kde`, in C) for colors, sizes, opacity and drawing order.
+* **Axes:** linear, log, log2, log10, sqrt, exp, symlog, logit, time and category scales;
+  limits such as `"0/max"` or `"q:0.01/q:0.99"`; ticks with shared decimals and ×10ᵏ
+  factors, calendar ticks for dates.
+* **Colors:** one system of color maps with kinds (qualitative, sequential, diverging,
+  cyclic; `okabe_ito` for color-vision deficiencies), a default map per kind, `center=`
+  for data about a value, interpolation in Oklab, colorbars;
+  marker sizes by area, equal for every shape.
+* **Styles:** `style="matplotlib"`, `"seaborn"`, `"ggplot2"`, `"altair"`, `"matlab"`,
+  `"tufte"` give the look of those libraries; styles are presets of settings, and your
+  own build on them (a variant: `{"base": "seaborn", …}`).
+* **Output:** PDF, EPS (transparency flattened), SVG, PNG, JPEG, JPEG XL and AVIF, by the
+  file's extension; CMYK, PDF/A and text as outlines are options of the save
+  (`color="cmyk"`, `pdf={"a": "2b"}`, `text="paths"`).
+* **Large data:** one million points mapped and drawn in C; `simplify="auto"` drops what
+  cannot be seen, `rasterize=` turns a series into an image inside the PDF.
+* **Genome tracks:** `bigwig` signals, `bed` intervals, `genes` and `gtf` gene models,
+  `hic` triangles and `links` over a region, and `site_profile` and `site_heatmap` around
+  sites, read from gwseq_io's readers (or any object with their methods) without
+  depending on gwseq_io.
+
+## Install
+
+```sh
+pip install plot_io
+```
+
+Python 3.11+. Text is shaped (kerned) and its fonts embedded by vec_io; it needs an
+installed outline font, the default family list (`Helvetica, Nimbus Sans, Nimbus Sans L,
+TeX Gyre Heros, FreeSans, Arial, Liberation Sans, Arimo`) taking its first installed family
+([docs/api.md](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md#1-conventions)). JPEG XL
+and AVIF output need vec_io's optional libjxl and libavif.
+
+Wheels hold plot_io's C library prebuilt for Linux (x86_64, aarch64; glibc or musl),
+Windows (x86_64, arm64) and macOS 13+ (arm64). Elsewhere (Intel Macs, other CPUs), pip
+builds it from the sdist with the local C compiler (gcc or clang; on Windows clang or MinGW,
+not MSVC); `pip install --no-binary plot_io plot_io` does so on any platform, and
+`pip install .` from a checkout. With the pinned zig in the environment (`pip install
+--require-hashes -r tools/zig.lock`, then `--no-build-isolation`), the sdist builds the
+release binary itself.
+
+```sh
+python -m plot_io._native verify    # the library loads, is the release binary, computes as it should
+```
+
+## Quick start
+
+```python
+import math
+import random
+
+import plot_io as plt
+
+rng = random.Random(1)
+x = [rng.uniform(0, 10) for _ in range(100)]
+y = [v + rng.gauss(0, 1) for v in x]
+s = plt.Scatter(x, y, name="data")
+s.plot.x_axis(label="Time (s)")
+s.plot.save("scatter.pdf")
+```
+
+```python
+t = [i / 10 for i in range(1, 101)]
+v = [math.exp(-k / 4) * (2 + math.sin(3 * k)) for k in t]
+upper, lower = [a * 1.3 for a in v], [a * 0.7 for a in v]
+va, vb = [rng.gauss(0, 1) for _ in range(50)], [rng.gauss(1, 1) for _ in range(50)]
+matrix = [[math.sin(i / 2) * math.cos(j / 3) for j in range(6)] for i in range(4)]
+columns, rows = list("ABCDEF"), ["w", "x", "y", "z"]
+z = [a - b for a, b in zip(x, y)]
+
+fig = plt.Figure(layout=[[1, 1, 2], [3, 4, 2]])
+p = fig.plot(1, title="Signal")
+p.line(t, v, name="raw")
+p.fill(t, upper, lower, opacity=0.3)
+p.y_axis(scale="log", label="Power")
+fig.plot(2).box({"a": va, "b": vb})
+h = fig.plot(3).heatmap(matrix, x=columns, y=rows, colorbar=True)
+fig.plot(4).scatter(x, y, c=z)
+fig.save("figure.pdf")
+fig.save("figure.png", dpi="2x")                         # 192 dpi; a number is dots per inch
+fig.save("figure.eps")                                   # PostScript: transparency flattened
+fig.save("figure.svg")
+data = fig.write("pdf", compression="small")             # bytes
+```
+
+```python
+fig = plt.Figure(style="ggplot2")                         # or a dict: {"base": "seaborn", "font_size": 9}
+```
+
+```python
+categories, counts = ["a", "b", "c"], [3, 5, 2]
+p = plt.Plot(width="3in", font_size="9pt", vars={"accent": "#c00"})
+p.x_axis(tick_length="0.5em", minor_tick_length="50%")
+p.bar(categories, counts, width="60%", color="var(--accent)")  # a custom variable
+p.x_axis.explain("tick_length")                          # where a value comes from
+```
+
+```python
+genes = {"umap_x": x, "umap_y": y}
+df = {"x": x, "y": y, "cluster": [int(a) % 3 for a in x],
+      "condition": ["ctrl" if a < 5 else "drug" for a in y],
+      "replicate": [i % 2 + 1 for i in range(len(x))]}
+p = plt.Plot()
+p.scatter("umap_x", "umap_y", data=genes, c="density", draw_order="density")   # columns
+p.scatter("x", "y", data=df, group="cluster", marker="auto")              # one series per group
+fig, panels = plt.facet(df, col="condition", row="replicate")               # small multiples
+panels.scatter("x", "y")
+```
+
+## Documentation
+
+* [Tutorial](https://github.com/ajgouhier/plot_io/blob/main/docs/tutorial.md): first steps, from a figure to its file.
+* [Guide](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md): the concepts by topic, with the settings tables.
+* [Examples](https://github.com/ajgouhier/plot_io/blob/main/docs/examples.md): every image of the docs, the scripts that draw them, recipes.
+* [Plot types](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md#plot-types): a page per plot type and figure helper, in
+  [docs/plots/](https://github.com/ajgouhier/plot_io/tree/main/docs/plots).
+* [Reference](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md#reference): a page per other class, in
+  [docs/reference/](https://github.com/ajgouhier/plot_io/tree/main/docs/reference).
+* [Coming from matplotlib](https://github.com/ajgouhier/plot_io/blob/main/docs/from_matplotlib.md): its calls and names in plot_io.
+
+## Development
+
+```sh
+python tools/check.py               # every check, one after the other (--list shows them)
+python tools/check.py --fast        # without the slow ones (~25 s)
+python tools/check.py --only suite,ruff,mypy
+python -m unittest discover -s tests -t tests
+python tools/bench.py --check       # timings: fails over 1.5 × a target
+python tools/gen_docs.py            # regenerate the generated sections and pages of docs/
+python tools/gen_docs.py --images   # ... and the images of the plot-type pages and the tutorial
+python tools/visual_check.py --save # render the examples as references
+python tools/visual_check.py        # ... and later compare with them (pixels, tolerances)
+PLOT_IO_GOLDEN=update python -m unittest discover -s tests -t tests -p test_golden.py
+python tools/record_stats.py > tests/stats_reference.json  # with numpy, scipy, statsmodels
+```
+
+The tests use only their own fonts (`tests/fonts`: VioTest, from vec_io's tests, through
+`VEC_IO_FONT_DIRS`; installed fonts are ignored), so their text is the same on every
+machine. The goldens (`tests/golden.json`) are keyed by meaning: the laid-out geometry of
+each scene (rounded to 1/64 px), its drawing calls and its 96 dpi pixels (the last two
+compared where the fonts match those recorded). Pixels are compared exactly, whatever
+vec_io's version, against `tests/golden/*.png` (in the repository, not in the sdist; else
+against a hash in `golden.json`); a changed scene writes its actual image and a diff to
+`$PLOT_IO_GOLDEN_OUT` (default `<tempdir>/plot_io-golden/`) and names them. A missing
+`golden.json`, or a recorded scene no longer drawn, fails. `PLOT_IO_GOLDEN=update` rewrites
+only the files that changed; `PLOT_IO_GOLDEN=dump:DIR` writes the geometry, the drawing
+calls and the PNG of each scene to `DIR` to see what changed.
+
+The tests that read real files with gwseq_io run when it is importable: installed, or a
+built checkout beside this one (`../gwseq_io/crates/gwseq-io-py/python`, added to the path
+by `tests/test_genome.py`; elsewhere, put that directory on `PYTHONPATH`). Otherwise they
+are skipped.
+
+`PLOT_IO_NATIVE_LIB=PATH` loads that build of plot_io's C library instead of the one in
+`plot_io/_native/bin/` (a sanitizer build, any other build; see
+`plot_io/_native/README.md`).
+
+The statistics kernels and color mapping are in plot_io's own C library, `plio`
+(`plot_io/_native`, built like vec_io's: zig, freestanding, five targets). Building and its
+checks:
+
+```sh
+pip install --require-hashes -r tools/zig.lock   # the pinned zig
+python tools/build_native.py        # bindings from src/plio_api.h, then all targets (not in git)
+python tools/build_native.py --check   # every target rebuilt byte for byte as manifest.json says
+python tools/crosscheck.py --qemu PATH/qemu-aarch64-static   # same results on aarch64
+python tools/dist.py                # the sdist and one wheel per platform built from it
+```
+
+The binaries are built, not committed: `manifest.json` records each target's hash. In a
+checkout, `pip install -e .` builds the host's binary the way pip does from the sdist.
+
+See [plot_io/_native/README.md](plot_io/_native/README.md).

plot_io-0.12.0/README.md

@@ -0,0 +1,199 @@
+# plot_io
+
+Plotting library in Python. No dependency other than [vec_io](https://github.com/ajgouhier/vec_io).
+
+* **Familiar:** a figure of plots; `scatter`, `line`, `fill`, `box`, `heatmap`, `bar`,
+  `hist`, `kde` (1-D and 2-D), `ecdf`, `violin`, `strip`, `swarm`, `area`, `errorbar`, `stem`,
+  `regression`, `hist2d`, `hexbin`, `contour`, `pie`, `dots`; reference lines, spans, text and
+  annotations; `color`, `marker`, `opacity`, `limits`, `scale`, `legend`.
+* **Statistical helpers:** `qq`, `raincloud`, `volcano`, `ma`, `manhattan`, `dendrogram`
+  plot types; significance brackets; point labels placed without overlaps; hierarchical
+  clustering (`plt.linkage`, scipy's methods); `plt.joint`, `plt.pairs` and `plt.clustermap`
+  figures in one call.
+* **Units:** lengths in px, pt, in, mm, `em`, `%` of the axes or of another setting, data
+  units and durations (`width="1.5in"`, `tick_length="0.5em"`, `width="60%"`,
+  `cap_length="12h"`); `var()` and `calc()` relate settings (`"calc(var(--gap) * 2)"`).
+* **One way to set things:** keywords at creation or `set(...)` on any object; figure
+  settings (font, color maps, line width…) are inherited by plots and series.
+* **Layouts:** an explicit grid of ids with spans, or automatic rows and columns; plots
+  sized individually; axis lines (or plot borders) aligned across the figure; margins from
+  the real text metrics; shared axes; secondary axes; facets of a table.
+* **Titles, legends and colorbars on any side:** several titles per plot; titles,
+  legends and colorbar keys shared by any rectangle of plots, nested as headers;
+  `share_legend` for one legend per row, column or figure; markup (`<i>`, `<b>`,
+  `<sup>`, …) in any text.
+* **Data:** columns of any table by name (`data=`: dicts, pandas, polars, numpy), one
+  series per group (`group=`), numbers given as colors or sizes, point density (scipy's
+  `gaussian_kde`, in C) for colors, sizes, opacity and drawing order.
+* **Axes:** linear, log, log2, log10, sqrt, exp, symlog, logit, time and category scales;
+  limits such as `"0/max"` or `"q:0.01/q:0.99"`; ticks with shared decimals and ×10ᵏ
+  factors, calendar ticks for dates.
+* **Colors:** one system of color maps with kinds (qualitative, sequential, diverging,
+  cyclic; `okabe_ito` for color-vision deficiencies), a default map per kind, `center=`
+  for data about a value, interpolation in Oklab, colorbars;
+  marker sizes by area, equal for every shape.
+* **Styles:** `style="matplotlib"`, `"seaborn"`, `"ggplot2"`, `"altair"`, `"matlab"`,
+  `"tufte"` give the look of those libraries; styles are presets of settings, and your
+  own build on them (a variant: `{"base": "seaborn", …}`).
+* **Output:** PDF, EPS (transparency flattened), SVG, PNG, JPEG, JPEG XL and AVIF, by the
+  file's extension; CMYK, PDF/A and text as outlines are options of the save
+  (`color="cmyk"`, `pdf={"a": "2b"}`, `text="paths"`).
+* **Large data:** one million points mapped and drawn in C; `simplify="auto"` drops what
+  cannot be seen, `rasterize=` turns a series into an image inside the PDF.
+* **Genome tracks:** `bigwig` signals, `bed` intervals, `genes` and `gtf` gene models,
+  `hic` triangles and `links` over a region, and `site_profile` and `site_heatmap` around
+  sites, read from gwseq_io's readers (or any object with their methods) without
+  depending on gwseq_io.
+
+## Install
+
+```sh
+pip install plot_io
+```
+
+Python 3.11+. Text is shaped (kerned) and its fonts embedded by vec_io; it needs an
+installed outline font, the default family list (`Helvetica, Nimbus Sans, Nimbus Sans L,
+TeX Gyre Heros, FreeSans, Arial, Liberation Sans, Arimo`) taking its first installed family
+([docs/api.md](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md#1-conventions)). JPEG XL
+and AVIF output need vec_io's optional libjxl and libavif.
+
+Wheels hold plot_io's C library prebuilt for Linux (x86_64, aarch64; glibc or musl),
+Windows (x86_64, arm64) and macOS 13+ (arm64). Elsewhere (Intel Macs, other CPUs), pip
+builds it from the sdist with the local C compiler (gcc or clang; on Windows clang or MinGW,
+not MSVC); `pip install --no-binary plot_io plot_io` does so on any platform, and
+`pip install .` from a checkout. With the pinned zig in the environment (`pip install
+--require-hashes -r tools/zig.lock`, then `--no-build-isolation`), the sdist builds the
+release binary itself.
+
+```sh
+python -m plot_io._native verify    # the library loads, is the release binary, computes as it should
+```
+
+## Quick start
+
+```python
+import math
+import random
+
+import plot_io as plt
+
+rng = random.Random(1)
+x = [rng.uniform(0, 10) for _ in range(100)]
+y = [v + rng.gauss(0, 1) for v in x]
+s = plt.Scatter(x, y, name="data")
+s.plot.x_axis(label="Time (s)")
+s.plot.save("scatter.pdf")
+```
+
+```python
+t = [i / 10 for i in range(1, 101)]
+v = [math.exp(-k / 4) * (2 + math.sin(3 * k)) for k in t]
+upper, lower = [a * 1.3 for a in v], [a * 0.7 for a in v]
+va, vb = [rng.gauss(0, 1) for _ in range(50)], [rng.gauss(1, 1) for _ in range(50)]
+matrix = [[math.sin(i / 2) * math.cos(j / 3) for j in range(6)] for i in range(4)]
+columns, rows = list("ABCDEF"), ["w", "x", "y", "z"]
+z = [a - b for a, b in zip(x, y)]
+
+fig = plt.Figure(layout=[[1, 1, 2], [3, 4, 2]])
+p = fig.plot(1, title="Signal")
+p.line(t, v, name="raw")
+p.fill(t, upper, lower, opacity=0.3)
+p.y_axis(scale="log", label="Power")
+fig.plot(2).box({"a": va, "b": vb})
+h = fig.plot(3).heatmap(matrix, x=columns, y=rows, colorbar=True)
+fig.plot(4).scatter(x, y, c=z)
+fig.save("figure.pdf")
+fig.save("figure.png", dpi="2x")                         # 192 dpi; a number is dots per inch
+fig.save("figure.eps")                                   # PostScript: transparency flattened
+fig.save("figure.svg")
+data = fig.write("pdf", compression="small")             # bytes
+```
+
+```python
+fig = plt.Figure(style="ggplot2")                         # or a dict: {"base": "seaborn", "font_size": 9}
+```
+
+```python
+categories, counts = ["a", "b", "c"], [3, 5, 2]
+p = plt.Plot(width="3in", font_size="9pt", vars={"accent": "#c00"})
+p.x_axis(tick_length="0.5em", minor_tick_length="50%")
+p.bar(categories, counts, width="60%", color="var(--accent)")  # a custom variable
+p.x_axis.explain("tick_length")                          # where a value comes from
+```
+
+```python
+genes = {"umap_x": x, "umap_y": y}
+df = {"x": x, "y": y, "cluster": [int(a) % 3 for a in x],
+      "condition": ["ctrl" if a < 5 else "drug" for a in y],
+      "replicate": [i % 2 + 1 for i in range(len(x))]}
+p = plt.Plot()
+p.scatter("umap_x", "umap_y", data=genes, c="density", draw_order="density")   # columns
+p.scatter("x", "y", data=df, group="cluster", marker="auto")              # one series per group
+fig, panels = plt.facet(df, col="condition", row="replicate")               # small multiples
+panels.scatter("x", "y")
+```
+
+## Documentation
+
+* [Tutorial](https://github.com/ajgouhier/plot_io/blob/main/docs/tutorial.md): first steps, from a figure to its file.
+* [Guide](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md): the concepts by topic, with the settings tables.
+* [Examples](https://github.com/ajgouhier/plot_io/blob/main/docs/examples.md): every image of the docs, the scripts that draw them, recipes.
+* [Plot types](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md#plot-types): a page per plot type and figure helper, in
+  [docs/plots/](https://github.com/ajgouhier/plot_io/tree/main/docs/plots).
+* [Reference](https://github.com/ajgouhier/plot_io/blob/main/docs/api.md#reference): a page per other class, in
+  [docs/reference/](https://github.com/ajgouhier/plot_io/tree/main/docs/reference).
+* [Coming from matplotlib](https://github.com/ajgouhier/plot_io/blob/main/docs/from_matplotlib.md): its calls and names in plot_io.
+
+## Development
+
+```sh
+python tools/check.py               # every check, one after the other (--list shows them)
+python tools/check.py --fast        # without the slow ones (~25 s)
+python tools/check.py --only suite,ruff,mypy
+python -m unittest discover -s tests -t tests
+python tools/bench.py --check       # timings: fails over 1.5 × a target
+python tools/gen_docs.py            # regenerate the generated sections and pages of docs/
+python tools/gen_docs.py --images   # ... and the images of the plot-type pages and the tutorial
+python tools/visual_check.py --save # render the examples as references
+python tools/visual_check.py        # ... and later compare with them (pixels, tolerances)
+PLOT_IO_GOLDEN=update python -m unittest discover -s tests -t tests -p test_golden.py
+python tools/record_stats.py > tests/stats_reference.json  # with numpy, scipy, statsmodels
+```
+
+The tests use only their own fonts (`tests/fonts`: VioTest, from vec_io's tests, through
+`VEC_IO_FONT_DIRS`; installed fonts are ignored), so their text is the same on every
+machine. The goldens (`tests/golden.json`) are keyed by meaning: the laid-out geometry of
+each scene (rounded to 1/64 px), its drawing calls and its 96 dpi pixels (the last two
+compared where the fonts match those recorded). Pixels are compared exactly, whatever
+vec_io's version, against `tests/golden/*.png` (in the repository, not in the sdist; else
+against a hash in `golden.json`); a changed scene writes its actual image and a diff to
+`$PLOT_IO_GOLDEN_OUT` (default `<tempdir>/plot_io-golden/`) and names them. A missing
+`golden.json`, or a recorded scene no longer drawn, fails. `PLOT_IO_GOLDEN=update` rewrites
+only the files that changed; `PLOT_IO_GOLDEN=dump:DIR` writes the geometry, the drawing
+calls and the PNG of each scene to `DIR` to see what changed.
+
+The tests that read real files with gwseq_io run when it is importable: installed, or a
+built checkout beside this one (`../gwseq_io/crates/gwseq-io-py/python`, added to the path
+by `tests/test_genome.py`; elsewhere, put that directory on `PYTHONPATH`). Otherwise they
+are skipped.
+
+`PLOT_IO_NATIVE_LIB=PATH` loads that build of plot_io's C library instead of the one in
+`plot_io/_native/bin/` (a sanitizer build, any other build; see
+`plot_io/_native/README.md`).
+
+The statistics kernels and color mapping are in plot_io's own C library, `plio`
+(`plot_io/_native`, built like vec_io's: zig, freestanding, five targets). Building and its
+checks:
+
+```sh
+pip install --require-hashes -r tools/zig.lock   # the pinned zig
+python tools/build_native.py        # bindings from src/plio_api.h, then all targets (not in git)
+python tools/build_native.py --check   # every target rebuilt byte for byte as manifest.json says
+python tools/crosscheck.py --qemu PATH/qemu-aarch64-static   # same results on aarch64
+python tools/dist.py                # the sdist and one wheel per platform built from it
+```
+
+The binaries are built, not committed: `manifest.json` records each target's hash. In a
+checkout, `pip install -e .` builds the host's binary the way pip does from the sdist.
+
+See [plot_io/_native/README.md](plot_io/_native/README.md).