diff --git a/hail/python/hail/docs/tutorials/10-ggplot2.ipynb b/hail/python/hail/docs/tutorials/10-ggplot2.ipynb
new file mode 100644
index 00000000000..9bebabe170d
--- /dev/null
+++ b/hail/python/hail/docs/tutorials/10-ggplot2.ipynb
@@ -0,0 +1,2188 @@
+{
+ "cells": [
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "id": "247fd83d-9599-4379-9549-e7863594dc7a",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ " \n",
+ " \n",
+ "
\n",
+ "
Loading BokehJS ...\n",
+ "
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"BokehJS does not appear to have successfully loaded. If loading BokehJS from CDN, this \\n\"+\n",
+ " \"may be due to a slow or bad network connection. Possible fixes:\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"- re-rerun `output_notebook()` to attempt to load from CDN again, or
\\n\"+\n",
+ " \"- use INLINE resources instead, as so:
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"from bokeh.resources import INLINE\\n\"+\n",
+ " \"output_notebook(resources=INLINE)\\n\"+\n",
+ " \"\\n\"+\n",
+ " \"
\\n\"+\n \"
\\n\"+\n \"BokehJS does not appear to have successfully loaded. If loading BokehJS from CDN, this \\n\"+\n \"may be due to a slow or bad network connection. Possible fixes:\\n\"+\n \"
\\n\"+\n \"
\\n\"+\n \"- re-rerun `output_notebook()` to attempt to load from CDN again, or
\\n\"+\n \"- use INLINE resources instead, as so:
\\n\"+\n \"
\\n\"+\n \"
\\n\"+\n \"from bokeh.resources import INLINE\\n\"+\n \"output_notebook(resources=INLINE)\\n\"+\n \"\\n\"+\n \"
| | |
idx | idx_2 | idx_3 |
| int32 | float64 | float64 |
\n",
+ "| 0 | 4.68e+01 | 3.27e+01 |
\n",
+ "| 1 | 8.24e+01 | 3.61e+01 |
\n",
+ "| 2 | 7.33e+01 | 7.62e+01 |
\n",
+ "| 3 | 8.99e+01 | 1.82e+01 |
\n",
+ "| 4 | 4.03e+01 | 2.52e+01 |
\n",
+ "| 5 | 5.45e+01 | 1.44e+01 |
\n",
+ "| 6 | 6.47e+01 | 9.74e+01 |
\n",
+ "| 7 | 6.72e+01 | 5.70e+01 |
\n",
+ "| 8 | 5.44e+01 | 3.35e+00 |
\n",
+ "| 9 | 4.80e+01 | 2.50e+01 |
\n",
+ "
showing top 10 rows
\n"
+ ],
+ "text/plain": [
+ "+-------+----------+----------+\n",
+ "| idx | idx_2 | idx_3 |\n",
+ "+-------+----------+----------+\n",
+ "| int32 | float64 | float64 |\n",
+ "+-------+----------+----------+\n",
+ "| 0 | 4.68e+01 | 3.27e+01 |\n",
+ "| 1 | 8.24e+01 | 3.61e+01 |\n",
+ "| 2 | 7.33e+01 | 7.62e+01 |\n",
+ "| 3 | 8.99e+01 | 1.82e+01 |\n",
+ "| 4 | 4.03e+01 | 2.52e+01 |\n",
+ "| 5 | 5.45e+01 | 1.44e+01 |\n",
+ "| 6 | 6.47e+01 | 9.74e+01 |\n",
+ "| 7 | 6.72e+01 | 5.70e+01 |\n",
+ "| 8 | 5.44e+01 | 3.35e+00 |\n",
+ "| 9 | 4.80e+01 | 2.50e+01 |\n",
+ "+-------+----------+----------+\n",
+ "showing top 10 rows"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ }
+ ],
+ "source": [
+ "from hail import rand_unif\n",
+ "from hail.utils import range_table\n",
+ "\n",
+ "data = range_table(100)\n",
+ "data = data.annotate(idx_2=rand_unif(0, 100), idx_3=rand_unif(0, 100))\n",
+ "data.show()"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "id": "98c77839-1706-4f9c-abb2-daccba645304",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.Chart(...)"
+ ]
+ },
+ "execution_count": 2,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import ChartWrapper\n",
+ "\n",
+ "wrapper = ChartWrapper(data)\n",
+ "wrapper.chart.mark_point().encode(\n",
+ " x=\"idx\",\n",
+ " y=\"idx_2\"\n",
+ ")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 5,
+ "id": "614c6573-effd-4bb4-a1ad-11844ffde8d0",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "
\n",
+ " \n",
+ " \n",
+ " | \n",
+ " x | \n",
+ " x2 | \n",
+ " y | \n",
+ "
\n",
+ " \n",
+ " \n",
+ " \n",
+ " | 0 | \n",
+ " 1.000005 | \n",
+ " 4.263316 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 1 | \n",
+ " 4.263316 | \n",
+ " 7.526628 | \n",
+ " 0 | \n",
+ "
\n",
+ " \n",
+ " | 2 | \n",
+ " 7.526628 | \n",
+ " 10.789939 | \n",
+ " 2 | \n",
+ "
\n",
+ " \n",
+ " | 3 | \n",
+ " 10.789939 | \n",
+ " 14.053250 | \n",
+ " 2 | \n",
+ "
\n",
+ " \n",
+ " | 4 | \n",
+ " 14.053250 | \n",
+ " 17.316561 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 5 | \n",
+ " 17.316561 | \n",
+ " 20.579872 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 6 | \n",
+ " 20.579872 | \n",
+ " 23.843183 | \n",
+ " 2 | \n",
+ "
\n",
+ " \n",
+ " | 7 | \n",
+ " 23.843183 | \n",
+ " 27.106494 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 8 | \n",
+ " 27.106494 | \n",
+ " 30.369805 | \n",
+ " 2 | \n",
+ "
\n",
+ " \n",
+ " | 9 | \n",
+ " 30.369805 | \n",
+ " 33.633116 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 10 | \n",
+ " 33.633116 | \n",
+ " 36.896427 | \n",
+ " 2 | \n",
+ "
\n",
+ " \n",
+ " | 11 | \n",
+ " 36.896427 | \n",
+ " 40.159738 | \n",
+ " 4 | \n",
+ "
\n",
+ " \n",
+ " | 12 | \n",
+ " 40.159738 | \n",
+ " 43.423049 | \n",
+ " 5 | \n",
+ "
\n",
+ " \n",
+ " | 13 | \n",
+ " 43.423049 | \n",
+ " 46.686360 | \n",
+ " 4 | \n",
+ "
\n",
+ " \n",
+ " | 14 | \n",
+ " 46.686360 | \n",
+ " 49.949671 | \n",
+ " 6 | \n",
+ "
\n",
+ " \n",
+ " | 15 | \n",
+ " 49.949671 | \n",
+ " 53.212982 | \n",
+ " 1 | \n",
+ "
\n",
+ " \n",
+ " | 16 | \n",
+ " 53.212982 | \n",
+ " 56.476293 | \n",
+ " 5 | \n",
+ "
\n",
+ " \n",
+ " | 17 | \n",
+ " 56.476293 | \n",
+ " 59.739605 | \n",
+ " 2 | \n",
+ "
\n",
+ " \n",
+ " | 18 | \n",
+ " 59.739605 | \n",
+ " 63.002916 | \n",
+ " 4 | \n",
+ "
\n",
+ " \n",
+ " | 19 | \n",
+ " 63.002916 | \n",
+ " 66.266227 | \n",
+ " 4 | \n",
+ "
\n",
+ " \n",
+ " | 20 | \n",
+ " 66.266227 | \n",
+ " 69.529538 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 21 | \n",
+ " 69.529538 | \n",
+ " 72.792849 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 22 | \n",
+ " 72.792849 | \n",
+ " 76.056160 | \n",
+ " 4 | \n",
+ "
\n",
+ " \n",
+ " | 23 | \n",
+ " 76.056160 | \n",
+ " 79.319471 | \n",
+ " 5 | \n",
+ "
\n",
+ " \n",
+ " | 24 | \n",
+ " 79.319471 | \n",
+ " 82.582782 | \n",
+ " 6 | \n",
+ "
\n",
+ " \n",
+ " | 25 | \n",
+ " 82.582782 | \n",
+ " 85.846093 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ " | 26 | \n",
+ " 85.846093 | \n",
+ " 89.109404 | \n",
+ " 4 | \n",
+ "
\n",
+ " \n",
+ " | 27 | \n",
+ " 89.109404 | \n",
+ " 92.372715 | \n",
+ " 2 | \n",
+ "
\n",
+ " \n",
+ " | 28 | \n",
+ " 92.372715 | \n",
+ " 95.636026 | \n",
+ " 7 | \n",
+ "
\n",
+ " \n",
+ " | 29 | \n",
+ " 95.636026 | \n",
+ " 98.899337 | \n",
+ " 3 | \n",
+ "
\n",
+ " \n",
+ "
\n",
+ "
\n",
+ "
\n",
+ "
Loading BokehJS ...\n",
+ "
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"BokehJS does not appear to have successfully loaded. If loading BokehJS from CDN, this \\n\"+\n",
+ " \"may be due to a slow or bad network connection. Possible fixes:\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"- re-rerun `output_notebook()` to attempt to load from CDN again, or
\\n\"+\n",
+ " \"- use INLINE resources instead, as so:
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"
\\n\"+\n",
+ " \"from bokeh.resources import INLINE\\n\"+\n",
+ " \"output_notebook(resources=INLINE)\\n\"+\n",
+ " \"\\n\"+\n",
+ " \"
\\n\"+\n \"
\\n\"+\n \"BokehJS does not appear to have successfully loaded. If loading BokehJS from CDN, this \\n\"+\n \"may be due to a slow or bad network connection. Possible fixes:\\n\"+\n \"
\\n\"+\n \"
\\n\"+\n \"- re-rerun `output_notebook()` to attempt to load from CDN again, or
\\n\"+\n \"- use INLINE resources instead, as so:
\\n\"+\n \"
\\n\"+\n \"
\\n\"+\n \"from bokeh.resources import INLINE\\n\"+\n \"output_notebook(resources=INLINE)\\n\"+\n \"\\n\"+\n \"
| | |
idx | idx_2 | idx_3 |
| int32 | float64 | float64 |
\n",
+ "| 0 | 4.68e+01 | 3.27e+01 |
\n",
+ "| 1 | 8.24e+01 | 3.61e+01 |
\n",
+ "| 2 | 7.33e+01 | 7.62e+01 |
\n",
+ "| 3 | 8.99e+01 | 1.82e+01 |
\n",
+ "| 4 | 4.03e+01 | 2.52e+01 |
\n",
+ "| 5 | 5.45e+01 | 1.44e+01 |
\n",
+ "| 6 | 6.47e+01 | 9.74e+01 |
\n",
+ "| 7 | 6.72e+01 | 5.70e+01 |
\n",
+ "| 8 | 5.44e+01 | 3.35e+00 |
\n",
+ "| 9 | 4.80e+01 | 2.50e+01 |
\n",
+ "
showing top 10 rows
\n"
+ ],
+ "text/plain": [
+ "+-------+----------+----------+\n",
+ "| idx | idx_2 | idx_3 |\n",
+ "+-------+----------+----------+\n",
+ "| int32 | float64 | float64 |\n",
+ "+-------+----------+----------+\n",
+ "| 0 | 4.68e+01 | 3.27e+01 |\n",
+ "| 1 | 8.24e+01 | 3.61e+01 |\n",
+ "| 2 | 7.33e+01 | 7.62e+01 |\n",
+ "| 3 | 8.99e+01 | 1.82e+01 |\n",
+ "| 4 | 4.03e+01 | 2.52e+01 |\n",
+ "| 5 | 5.45e+01 | 1.44e+01 |\n",
+ "| 6 | 6.47e+01 | 9.74e+01 |\n",
+ "| 7 | 6.72e+01 | 5.70e+01 |\n",
+ "| 8 | 5.44e+01 | 3.35e+00 |\n",
+ "| 9 | 4.80e+01 | 2.50e+01 |\n",
+ "+-------+----------+----------+\n",
+ "showing top 10 rows"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ }
+ ],
+ "source": [
+ "from hail import rand_unif\n",
+ "from hail.utils import range_table\n",
+ "\n",
+ "data = range_table(100)\n",
+ "data = data.annotate(idx_2=rand_unif(0, 100), idx_3=rand_unif(0, 100))\n",
+ "data.show()"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "860d0a3a",
+ "metadata": {},
+ "source": [
+ "## Plot Objects\n",
+ "\n",
+ "When creating a plot, we start out with a basic **plot object**, which wraps our data. We can create such an object using the [`ggplot`](https://hail.is/docs/0.2/ggplot/index.html#hail.ggplot.ggplot) function, and take a closer look using Python's builtin [`print`](https://docs.python.org/3.9/library/functions.html#print) function:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "id": "45ac0248",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Plot(\n",
+ " data = ,\n",
+ " mapping = Mapping(\n",
+ " x = None,\n",
+ " y = None,\n",
+ " ),\n",
+ " layers = [],\n",
+ ")\n"
+ ]
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import ggplot\n",
+ "\n",
+ "plot = ggplot(data)\n",
+ "print(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "b584643d",
+ "metadata": {},
+ "source": [
+ "## Aesthetic Mappings\n",
+ "\n",
+ "Next, we'll need to specify which values to plot along our x- and y-axes.\n",
+ "\n",
+ "We'll use the [`aes`](https://hail.is/docs/0.2/ggplot2/index.html#hail.ggplot2.aes) function to create an **aesthetic mapping**, which links data to visual components of our plot. `aes` assumes its first two arguments are `x` and `y`, and can be passed any other arguments via keyword.\n",
+ "\n",
+ "Let's map the `idx` column of our data to the x-axis, and the `idx_2` column to the y-axis:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 3,
+ "id": "4bd3ce19",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Plot(\n",
+ " data = ,\n",
+ " mapping = Mapping(\n",
+ " x = idx,\n",
+ " y = idx_2,\n",
+ " ),\n",
+ " layers = [],\n",
+ ")\n"
+ ]
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import aes\n",
+ "\n",
+ "plot += aes(\"idx\", \"idx_2\")\n",
+ "print(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "28d8b3c5-f656-4e16-8eb9-f04b38962a96",
+ "metadata": {},
+ "source": [
+ "Aesthetic mappings can also be used to specify visual properties of a plot, such as the color of a line, as we'll see in the next section."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "d3edd5a4",
+ "metadata": {},
+ "source": [
+ "## Layers\n",
+ "\n",
+ "Like R's `ggplot2`, Hail `ggplot2` bases its approach to plotting on a [layered grammar of graphics](https://ggplot2-book.org/introduction.html#what-is-the-grammar-of-graphics).\n",
+ "\n",
+ "A **layer** is a self-contained collection of visual characteristics, data, and/or aggregations over that data. We'll build up our plot from the data by adding different layers to it.\n",
+ "\n",
+ "### Geoms\n",
+ "\n",
+ "A **geom** is a type of layer that specifies which kind of plot we're making. For example, `geom_point` indicates that we're making a scatterplot:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 4,
+ "id": "257821d6",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Plot(\n",
+ " data = ,\n",
+ " mapping = Mapping(\n",
+ " x = idx,\n",
+ " y = idx_2,\n",
+ " ),\n",
+ " layers = [Layer(mapping=Mapping(x=None, y=None, color=None), data=None, geom='circle', stat='identity', params={})],\n",
+ ")\n"
+ ]
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import geom_point\n",
+ "\n",
+ "plot += geom_point()\n",
+ "print(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "6a6ac0ec",
+ "metadata": {},
+ "source": [
+ "Since we've now added information about what the plot will look like, we can display it using the `show` function:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 5,
+ "id": "71d0e4a2",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.Chart(...)"
+ ]
+ },
+ "execution_count": 5,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import show\n",
+ "\n",
+ "show(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "4fa9d8b9",
+ "metadata": {},
+ "source": [
+ "We can also add multiple geoms to a single plot.\n",
+ "\n",
+ "For example, we can plot a line on top of our scatterplot using `geom_line`. This line will still plot the values of the `idx` column along the x-axis, but use the `idx_3` column for the y-axis. We can override the `y` mapping for this geom by passing in its own `aes`:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 6,
+ "id": "0c266d8f",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.LayerChart(...)"
+ ]
+ },
+ "execution_count": 6,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import geom_line\n",
+ "\n",
+ "plot += geom_line(aes(y=\"idx_3\"))\n",
+ "show(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "bd233399-39a9-46e0-a125-bfbc044e866e",
+ "metadata": {},
+ "source": [
+ "It's a little tough to visually distinguish this line from the scatterplot's points, so let's remove it and re-add it with a different color."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "d138d5c4",
+ "metadata": {},
+ "source": [
+ "### Removing Layers\n",
+ "\n",
+ "With R's `ggplot2`, if you add a layer to a plot object, [it can be tough to remove it](https://stackoverflow.com/questions/50434608/remove-geoms-from-an-existing-ggplot-chart).\n",
+ "\n",
+ "However, Hail GGPlot's plot objects keep track of each addition made to them, allowing us to use the `undo` method to roll back a single addition:"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 7,
+ "id": "5ffeb79f",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.Chart(...)"
+ ]
+ },
+ "execution_count": 7,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import undo\n",
+ "\n",
+ "plot = undo(plot)\n",
+ "show(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "045ebf6c-713b-4f2c-bd81-e2722daeb917",
+ "metadata": {},
+ "source": [
+ "Now we can specify the color of our line by passing its [hexadecimal color code](https://www.w3schools.com/colors/colors_picker.asp) to the [`color`]() keyword of [`aes`](https://hail.is/docs/0.2/ggplot/index.html#hail.ggplot.aes):"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 8,
+ "id": "349211fd-6d5c-4c2d-8f0d-4450a91ddd24",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.LayerChart(...)"
+ ]
+ },
+ "execution_count": 8,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "plot += geom_line(aes(y=\"idx_3\", color=\"#e3b5ce\"))\n",
+ "show(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "c3d781bb",
+ "metadata": {},
+ "source": [
+ "### Stats\n",
+ "\n",
+ "Let's try making another type of plot. First, we can clear out the layers that we've added so far all at once by passing the [`depth`]() keyword argument to [`undo`]():"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 9,
+ "id": "e99dc4c1-c9c1-4138-80c7-90fab37ac121",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Plot(\n",
+ " data = ,\n",
+ " mapping = Mapping(\n",
+ " x = None,\n",
+ " y = None,\n",
+ " ),\n",
+ " layers = [],\n",
+ ")\n"
+ ]
+ }
+ ],
+ "source": [
+ "plot = undo(plot, depth=3)\n",
+ "print(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "153f0fbb-1ce2-41c1-a196-e0088baa5b15",
+ "metadata": {},
+ "source": [
+ "Now, let's make a histogram using [`geom_histogram`]():"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
+ "id": "8a4c6ec3-62a3-463c-8500-3fc449746d32",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.Chart(...)"
+ ]
+ },
+ "execution_count": 10,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "from hail.ggplot2 import geom_histogram\n",
+ "\n",
+ "plot += geom_histogram(aes(\"idx_2\"))\n",
+ "show(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "46b31847-534b-4caa-bf4a-565cab1b9fb1",
+ "metadata": {},
+ "source": [
+ "[`geom_histogram`]() places the data into 30 bins by default, but what if we want to bin the data differently? We can use the [`bins`]() argument to [`geom_histogram`]():"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 11,
+ "id": "8c20df8d-092b-48f5-85cb-45e486bbbb5b",
+ "metadata": {},
+ "outputs": [
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ "\n",
+ "\n",
+ ""
+ ],
+ "text/plain": [
+ "alt.Chart(...)"
+ ]
+ },
+ "execution_count": 11,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "plot = undo(plot)\n",
+ "plot += geom_histogram(aes(\"idx_2\"), bins=3)\n",
+ "show(plot)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "86193b45-31a4-428e-9e1c-222c632ad7a6",
+ "metadata": {},
+ "source": [
+ "TODO: you may have noticed that some geoms, like histogram, implicitly compute some statistics about the data before rendering it. but what if you need to transform your data independently of a geom?\n",
+ "\n",
+ "TODO: stats are cached, so when you recompute them, the plot object will attempt to reuse the cached values of previously applied aggregations"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "825baee3",
+ "metadata": {
+ "jp-MarkdownHeadingCollapsed": true
+ },
+ "source": [
+ "## Titles\n",
+ "\n",
+ "TODO"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "8fdbe273",
+ "metadata": {
+ "jp-MarkdownHeadingCollapsed": true
+ },
+ "source": [
+ "## Axis Labels\n",
+ "\n",
+ "TODO"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "9f3b6784",
+ "metadata": {
+ "jp-MarkdownHeadingCollapsed": true
+ },
+ "source": [
+ "## Scales\n",
+ "\n",
+ "TODO: what are scales"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "74613547",
+ "metadata": {
+ "jp-MarkdownHeadingCollapsed": true
+ },
+ "source": [
+ "## Facets\n",
+ "\n",
+ "TODO"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "4c37a13e",
+ "metadata": {},
+ "source": [
+ "## Examples"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "bc860479",
+ "metadata": {},
+ "source": [
+ "### Cumulative Histogram\n",
+ "\n",
+ "TODO"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "ac2baeee",
+ "metadata": {},
+ "source": [
+ "### 2D Histogram\n",
+ "\n",
+ "TODO"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "b60781bf",
+ "metadata": {},
+ "source": [
+ "### Scatter Plot\n",
+ "\n",
+ "TODO"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "5b802442",
+ "metadata": {},
+ "source": [
+ "### QQ Plot\n",
+ "\n",
+ "TODO: To create a quantile-quantile (QQ) plot, ..."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "633d723b",
+ "metadata": {},
+ "source": [
+ "### Manhattan Plot\n",
+ "\n",
+ "TODO (use actual genetics data)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "73fdbf30",
+ "metadata": {},
+ "source": [
+ "## Common Mistakes\n",
+ "\n",
+ "TODO: don't use the class constructors, call the special functions"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "id": "c08df783",
+ "metadata": {},
+ "source": [
+ "TODO: example specifically demonstrating downsampling"
+ ]
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "Python 3 (ipykernel)",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.9.19"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}
diff --git a/hail/python/hail/ggplot2/.ipynb_checkpoints/ggplot2-checkpoint.py b/hail/python/hail/ggplot2/.ipynb_checkpoints/ggplot2-checkpoint.py
new file mode 100644
index 00000000000..df390de4d89
--- /dev/null
+++ b/hail/python/hail/ggplot2/.ipynb_checkpoints/ggplot2-checkpoint.py
@@ -0,0 +1,190 @@
+from dataclasses import asdict, replace
+from textwrap import dedent, indent
+from typing import Any, Dict, List, Literal, Optional, Tuple, Union
+
+from altair import X2, Chart, LayerChart, X, Y
+from pandas import DataFrame
+
+import hail as hl
+from hail import MatrixTable, Table
+from hail.ggplot2.utils import typeguard_dataclass
+
+
+### types ###
+Data = Union[Table, MatrixTable]
+
+
+@typeguard_dataclass
+class Mapping:
+ x: Optional[str]
+ y: Optional[str]
+ # TODO add the rest of the supported aesthetic names
+ color: Optional[str]
+
+
+Geom = Literal["bar", "line", "circle"]
+Stat = Literal["identity", "bin"]
+
+
+@typeguard_dataclass
+class Layer:
+ mapping: Mapping
+ data: Optional[Data]
+ geom: Optional[Geom]
+ stat: Stat
+ # FIXME if there's only one type per param name we can make this a typeddict
+ params: Dict[str, Any]
+
+
+@typeguard_dataclass
+class Plot:
+ data: Optional[Data]
+ mapping: Mapping
+ layers: list[Layer]
+
+
+### module-level variables ###
+_plot_cache: Dict[int, List[Plot]] = {}
+_stat_cache: Dict[Tuple[int, ...], Data] = {}
+
+
+### constructor functions ###
+def aes(x: Optional[str] = None, y: Optional[str] = None, color: Optional[str] = None) -> Mapping:
+ return Mapping(x, y, color)
+
+
+def geom_histogram(mapping: Mapping = aes(), data: Optional[Data] = None, bins: int = 30) -> Layer:
+ return Layer(mapping, data, "bar", "bin", {"bins": bins})
+
+
+def geom_line(mapping: Mapping = aes(), data: Optional[Data] = None) -> Layer:
+ return Layer(mapping, data, "line", "identity", {})
+
+
+def geom_point(mapping: Mapping = aes(), data: Optional[Data] = None) -> Layer:
+ return Layer(mapping, data, "circle", "identity", {})
+
+
+def ggplot(data: Optional[Data] = None, mapping: Mapping = aes()) -> Plot:
+ global _plot_cache
+ new_plot = Plot(data, mapping, [])
+ _plot_cache |= {id(new_plot): []}
+ return new_plot
+
+
+### functionality ###
+def extend(plot: Plot, other: Any) -> Plot:
+ global _plot_cache
+ kwargs: Optional[Dict[str, Any]] = None
+ if isinstance(other, Mapping):
+ kwargs = {
+ "mapping": replace(
+ plot.mapping,
+ **{k: v for k, v in {"x": other.x, "y": other.y, "color": other.color}.items() if v is not None},
+ )
+ }
+ elif isinstance(other, Layer):
+ kwargs = {"layers": [*plot.layers, other]}
+
+ if kwargs is None:
+ raise ValueError("unsupported addition to plot")
+
+ new_plot = replace(plot, **kwargs)
+ _plot_cache |= {id(new_plot): _plot_cache[id(plot)] + [plot]}
+ _plot_cache = {k: v for k, v in _plot_cache.items() if k != id(plot)}
+ return new_plot
+
+
+setattr(Plot, "__add__", extend)
+
+
+_altair_configure_mark_keys = {"color"}
+_altair_encode_keys = {"x": X, "x2": X2, "y": Y}
+
+
+def show(plot: Plot) -> Union[Chart, LayerChart]:
+ global _stat_cache
+ base_chart = None
+ for layer in plot.layers:
+ mapping_dict = {}
+ for mapping in [plot.mapping, layer.mapping]:
+ mapping_dict = {**mapping_dict, **{k: v for k, v in asdict(mapping).items() if v is not None}}
+ # TODO should we break the stat stuff out to its own function?
+ kwargs = {"x": {}, "x2": {}, "y": {}}
+ cached = _stat_cache.get((id(plot.data), layer.stat), None)
+ if cached is not None:
+ data, df = cached
+ elif layer.stat == "identity":
+ data = plot.data
+ df = data.to_pandas()
+ elif layer.stat == "bin":
+ # TODO add caching
+ x = mapping_dict.get("x", None)
+ if x is None:
+ raise ValueError("x must be supplied for stat bin")
+ data = plot.data.aggregate(
+ hl.agg.hist(
+ plot.data[x],
+ plot.data.aggregate(hl.agg.min(plot.data[x])),
+ plot.data.aggregate(hl.agg.max(plot.data[x])),
+ layer.params["bins"],
+ )
+ )
+ df = DataFrame([
+ {x: data["bin_edges"][i], "x2": data["bin_edges"][i + 1], "y": data["bin_freq"][i]}
+ for i in range(len(data["bin_freq"]))
+ ])
+ kwargs["x"] = {"bin": "binned"}
+ mapping_dict["x2"] = "x2"
+ mapping_dict["y"] = "y"
+ else:
+ raise ValueError("unknown stat")
+ _stat_cache |= {(id(plot.data), layer.stat): (data, df)}
+ chart = Chart(df)
+ if layer.geom is not None:
+ chart = getattr(chart, f"mark_{layer.geom}")(**{
+ k: v for k, v in mapping_dict.items() if k in _altair_configure_mark_keys
+ })
+ chart = chart.encode(**{
+ k: _altair_encode_keys[k](v, **kwargs[k]) for k, v in mapping_dict.items() if k in _altair_encode_keys
+ })
+ base_chart = chart if base_chart is None else base_chart + chart
+ return base_chart
+
+
+def undo(plot: Plot, *, depth: int = 1) -> Plot:
+ global _plot_cache
+ old_plot = _plot_cache[id(plot)][0 - depth]
+ _plot_cache |= {id(old_plot): _plot_cache[id(plot)][: 0 - depth]}
+ _plot_cache = {k: v for k, v in _plot_cache.items() if k != id(plot)}
+ return old_plot
+
+
+## introspection ##
+def plot_to_string(plot: Plot) -> str:
+ return dedent(f"""\
+ Plot(
+ data = {plot.data},
+ mapping = {indent_tail(str(plot.mapping), 3)},
+ layers = {indent_tail(str(plot.layers), 3)},
+ )""")
+
+
+def indent_tail(string: str, indent_level: int = 1) -> str:
+ return "".join([
+ indent(part, " " * indent_level) if index == 2 else part for index, part in enumerate(string.partition("\n"))
+ ])
+
+
+setattr(Plot, "__str__", plot_to_string)
+
+
+def mapping_to_string(mapping: Mapping) -> str:
+ return dedent(f"""\
+ Mapping(
+ x = {mapping.x},
+ y = {mapping.y},
+ )""")
+
+
+setattr(Mapping, "__str__", mapping_to_string)
diff --git a/hail/python/hail/ggplot2/__init__.py b/hail/python/hail/ggplot2/__init__.py
new file mode 100644
index 00000000000..a41e7235a23
--- /dev/null
+++ b/hail/python/hail/ggplot2/__init__.py
@@ -0,0 +1,29 @@
+from typeguard import install_import_hook
+
+install_import_hook("hail.ggplot2")
+
+# These imports need to be placed after the import hook in order for typechecking to work.
+# https://typeguard.readthedocs.io/en/stable/userguide.html#using-the-import-hook
+from .altair_wrapper import ChartWrapper # noqa: E402
+from .ggplot2 import ( # noqa: E402
+ aes,
+ extend,
+ geom_histogram,
+ geom_line,
+ geom_point,
+ ggplot,
+ show,
+ undo,
+)
+
+__all__ = [
+ "ChartWrapper",
+ "aes",
+ "extend",
+ "geom_point",
+ "geom_line",
+ "geom_histogram",
+ "ggplot",
+ "undo",
+ "show",
+]
diff --git a/hail/python/hail/ggplot2/altair_wrapper.py b/hail/python/hail/ggplot2/altair_wrapper.py
new file mode 100644
index 00000000000..dfe1a55f393
--- /dev/null
+++ b/hail/python/hail/ggplot2/altair_wrapper.py
@@ -0,0 +1,46 @@
+from typing import Any, Union
+
+from altair import Chart
+from pandas import DataFrame
+
+import hail
+from hail import MatrixTable, Table
+
+Data = Union[Table, MatrixTable]
+
+
+class ChartWrapper:
+ def __init__(self, data: Data, *args, **kwargs) -> None:
+ self.chart_args = args
+ self.chart_kwargs = kwargs
+ self.data = data
+
+ def __setattr__(self, name: str, value: Any) -> None:
+ super().__setattr__(name, value)
+ if name == "data":
+ self.update_data()
+
+ def update_data(self) -> None:
+ self.cache = {}
+ self.chart = Chart(self.data.to_pandas(), *self.chart_args, **self.chart_kwargs)
+
+ def histogram(self, x: str, bins: int = 30) -> None:
+ if (aggregated := self.cache.get("histogram", None)) is None:
+ self.cache["histogram"] = (
+ aggregated := self.data.aggregate(
+ hail.agg.hist(
+ self.data[x],
+ self.data.aggregate(hail.agg.min(self.data[x])),
+ self.data.aggregate(hail.agg.max(self.data[x])),
+ bins,
+ )
+ )
+ )
+ self.chart = Chart(
+ DataFrame([
+ {"x": aggregated["bin_edges"][i], "x2": aggregated["bin_edges"][i + 1], "y": aggregated["bin_freq"][i]}
+ for i in range(len(aggregated["bin_freq"]))
+ ]),
+ *self.chart_args,
+ **self.chart_kwargs,
+ )
diff --git a/hail/python/hail/ggplot2/ggplot2.py b/hail/python/hail/ggplot2/ggplot2.py
new file mode 100644
index 00000000000..0a435aa5d00
--- /dev/null
+++ b/hail/python/hail/ggplot2/ggplot2.py
@@ -0,0 +1,197 @@
+from dataclasses import asdict, replace
+from textwrap import dedent, indent
+from typing import Any, Dict, List, Literal, Optional, Tuple, Union
+
+from altair import X2, Chart, LayerChart, X, Y
+from pandas import DataFrame
+
+import hail as hl
+from hail import MatrixTable, Table
+from hail.ggplot2.utils import typeguard_dataclass
+
+# TODO before review
+# expose the underlying altair object
+# write unit tests asserting against that
+# check whether there's stuff you can put in the ggplot interface that doesnt work for histogram
+# apply stat caching to histogram
+# finish histogram section in notebook
+# TODO later
+# extract core wrapper that does hail table -> vega, expose ggplot interface and altair style interface over that
+
+### types ###
+Data = Union[Table, MatrixTable]
+
+
+@typeguard_dataclass
+class Mapping:
+ x: Optional[str]
+ y: Optional[str]
+ # TODO add the rest of the supported aesthetic names
+ color: Optional[str]
+
+
+Geom = Literal["bar", "line", "circle"]
+Stat = Literal["identity", "bin"]
+
+
+@typeguard_dataclass
+class Layer:
+ mapping: Mapping
+ data: Optional[Data]
+ geom: Optional[Geom]
+ stat: Stat
+ # FIXME if there's only one type per param name we can make this a typeddict
+ params: Dict[str, Any]
+
+
+@typeguard_dataclass
+class Plot:
+ data: Optional[Data]
+ mapping: Mapping
+ layers: list[Layer]
+
+
+### module-level variables ###
+# TODO eviction policy for cache & hl init level config for it
+_plot_cache: Dict[int, List[Plot]] = {}
+_stat_cache: Dict[Tuple[int, ...], Data] = {}
+
+
+### constructor functions ###
+def aes(x: Optional[str] = None, y: Optional[str] = None, color: Optional[str] = None) -> Mapping:
+ return Mapping(x, y, color)
+
+
+def geom_histogram(mapping: Mapping = aes(), data: Optional[Data] = None, bins: int = 30) -> Layer:
+ return Layer(mapping, data, "bar", "bin", {"bins": bins})
+
+
+def geom_line(mapping: Mapping = aes(), data: Optional[Data] = None) -> Layer:
+ return Layer(mapping, data, "line", "identity", {})
+
+
+def geom_point(mapping: Mapping = aes(), data: Optional[Data] = None) -> Layer:
+ return Layer(mapping, data, "circle", "identity", {})
+
+
+def ggplot(data: Optional[Data] = None, mapping: Mapping = aes()) -> Plot:
+ global _plot_cache
+ new_plot = Plot(data, mapping, [])
+ _plot_cache |= {id(new_plot): []}
+ return new_plot
+
+
+### functionality ###
+def extend(plot: Plot, other: Any) -> Plot:
+ global _plot_cache
+ kwargs: Optional[Dict[str, Any]] = None
+ if isinstance(other, Mapping):
+ kwargs = {
+ "mapping": replace(
+ plot.mapping,
+ **{k: v for k, v in {"x": other.x, "y": other.y, "color": other.color}.items() if v is not None},
+ )
+ }
+ elif isinstance(other, Layer):
+ kwargs = {"layers": [*plot.layers, other]}
+ if kwargs is None:
+ raise ValueError("unsupported addition to plot")
+ new_plot = replace(plot, **kwargs)
+ _plot_cache |= {id(new_plot): _plot_cache[id(plot)] + [plot]}
+ _plot_cache = {k: v for k, v in _plot_cache.items() if k != id(plot)}
+ return new_plot
+
+
+setattr(Plot, "__add__", extend)
+
+
+_altair_configure_mark_keys = {"color"}
+_altair_encode_keys = {"x": X, "x2": X2, "y": Y}
+
+
+def show(plot: Plot) -> Union[Chart, LayerChart]:
+ global _stat_cache
+ base_chart = None
+ for layer in plot.layers:
+ mapping_dict = {}
+ for mapping in [plot.mapping, layer.mapping]:
+ mapping_dict = {**mapping_dict, **{k: v for k, v in asdict(mapping).items() if v is not None}}
+ # TODO should we break the stat stuff out to its own function?
+ kwargs = {"x": {}, "x2": {}, "y": {}}
+ cached = _stat_cache.get((id(plot.data), layer.stat), None)
+ if cached is not None:
+ data, df = cached
+ elif layer.stat == "identity":
+ data = plot.data
+ df = data.to_pandas()
+ elif layer.stat == "bin":
+ x = mapping_dict.get("x", None)
+ if x is None:
+ raise ValueError("x must be supplied for stat bin")
+ data = plot.data.aggregate(
+ hl.agg.hist(
+ plot.data[x],
+ plot.data.aggregate(hl.agg.min(plot.data[x])),
+ plot.data.aggregate(hl.agg.max(plot.data[x])),
+ layer.params["bins"],
+ )
+ )
+ df = DataFrame([
+ {x: data["bin_edges"][i], "x2": data["bin_edges"][i + 1], "y": data["bin_freq"][i]}
+ for i in range(len(data["bin_freq"]))
+ ])
+ kwargs["x"] = {"bin": "binned"}
+ mapping_dict["x2"] = "x2"
+ mapping_dict["y"] = "y"
+ else:
+ raise ValueError("unknown stat")
+ # TODO remove the data from the cache, we only need the df
+ _stat_cache |= {(id(plot.data), layer.stat): (data, df)}
+ chart = Chart(df)
+ if layer.geom is not None:
+ chart = getattr(chart, f"mark_{layer.geom}")(**{
+ k: v for k, v in mapping_dict.items() if k in _altair_configure_mark_keys
+ })
+ chart = chart.encode(**{
+ k: _altair_encode_keys[k](v, **kwargs[k]) for k, v in mapping_dict.items() if k in _altair_encode_keys
+ })
+ base_chart = chart if base_chart is None else base_chart + chart
+ return base_chart
+
+
+def undo(plot: Plot, *, depth: int = 1) -> Plot:
+ global _plot_cache
+ old_plot = _plot_cache[id(plot)][0 - depth]
+ _plot_cache |= {id(old_plot): _plot_cache[id(plot)][: 0 - depth]}
+ _plot_cache = {k: v for k, v in _plot_cache.items() if k != id(plot)}
+ return old_plot
+
+
+## introspection ##
+def plot_to_string(plot: Plot) -> str:
+ return dedent(f"""\
+ Plot(
+ data = {plot.data},
+ mapping = {indent_tail(str(plot.mapping), 3)},
+ layers = {indent_tail(str(plot.layers), 3)},
+ )""")
+
+
+def indent_tail(string: str, indent_level: int = 1) -> str:
+ return "".join([
+ indent(part, " " * indent_level) if index == 2 else part for index, part in enumerate(string.partition("\n"))
+ ])
+
+
+setattr(Plot, "__str__", plot_to_string)
+
+
+def mapping_to_string(mapping: Mapping) -> str:
+ return dedent(f"""\
+ Mapping(
+ x = {mapping.x},
+ y = {mapping.y},
+ )""")
+
+
+setattr(Mapping, "__str__", mapping_to_string)
diff --git a/hail/python/hail/ggplot2/utils.py b/hail/python/hail/ggplot2/utils.py
new file mode 100644
index 00000000000..cb84814e44b
--- /dev/null
+++ b/hail/python/hail/ggplot2/utils.py
@@ -0,0 +1,38 @@
+from dataclasses import dataclass, fields
+from functools import wraps
+from typing import Any, Callable, TypeVar, Union
+
+from typeguard import check_type
+
+ReturnType = TypeVar("ReturnType")
+WrappedDecorator = Callable[[ReturnType], ReturnType]
+
+
+def typeguard_dataclass(cls: ReturnType = None, /, **kwargs: Any) -> Union[ReturnType, WrappedDecorator]:
+ """
+ Creates a `dataclass` that is `frozen` by default and has runtime typechecking for its fields.
+ """
+
+ @wraps(dataclass)
+ def wrapper(cls: ReturnType) -> ReturnType:
+ def __setattr__(obj: ReturnType, name: str, value: Any) -> None:
+ if len(types := [_field.type for _field in fields(obj) if _field.name == name]) == 0:
+ raise TypeError(f"'{getattr(cls, '__name__', str(cls))}' has no field '{name}'.")
+ super().__setattr__(name, check_type(value, types[0]))
+
+ def __post_init__(obj: ReturnType) -> None:
+ for _field in fields(obj):
+ check_type(getattr(obj, _field.name), _field.type)
+
+ setattr(
+ cls,
+ *(
+ ["__post_init__", __post_init__]
+ if (frozen := kwargs.get("frozen", True))
+ else ["__setattr__", __setattr__]
+ ),
+ )
+ dataclass(cls, frozen=frozen, **{k: v for k, v in kwargs.items() if k != "frozen"})
+ return cls
+
+ return wrapper if cls is None else wrapper(cls)
diff --git a/hail/python/pinned-requirements.txt b/hail/python/pinned-requirements.txt
index 394193216e0..de398cd6413 100644
--- a/hail/python/pinned-requirements.txt
+++ b/hail/python/pinned-requirements.txt
@@ -1,5 +1,7 @@
# This file was autogenerated by uv via the following command:
# uv pip compile --python-version 3.9 --python-platform linux hail/python/requirements.txt --output-file=hail/python/pinned-requirements.txt
+altair==5
+typeguard==4
aiodns==2.0.0
# via
# -c hail/python/hailtop/pinned-requirements.txt
diff --git a/hail/python/requirements.txt b/hail/python/requirements.txt
index ba04650d8d6..c15da77639f 100644
--- a/hail/python/requirements.txt
+++ b/hail/python/requirements.txt
@@ -2,6 +2,8 @@
-c dataproc-pre-installed-requirements.txt
-r hailtop/requirements.txt
+altair==5
+typeguard==4
avro>=1.10,<1.12
bokeh>=3,<3.4
decorator<5