Add tooltip option to Altair chart

[FoFFolo]

I added in the altair.py experimental script the option to see a tooltip on each agent in the canvas when hovering on them. The logic is to see every option described by the agent_portrayal function, except for the x, y, color and size properties.
When defining the type of each tooltip, I didn't manage to find an optimal algorithm to check if a string can be considered a date object, in order to change its type from "nominal" to "temporal", without using an external library called dateutil, which contains the function parser.parse() that returns a boolean if the string is a right date format.

Here is an example:

import mesa
import solara
from matplotlib.figure import Figure
from mesa.experimental import JupyterViz
from money_model.agent import MoneyAgent
from money_model.model import MoneyModel


def agent_portrayal(MoneyAgent):
    size = 10
    color = "tab:red"
    if MoneyAgent.wealth > 0:
        size = 50
        color = "tab:blue"
    if MoneyAgent.unique_id == 0:
        size = 60
        color = "yellow"
    return {"size": size, "color": color, "Unique_id": MoneyAgent.unique_id}


def make_histogram(model):
    # ...

model_params = {
    # ...
}

page = JupyterViz(
    model_class=MoneyModel,
    model_params=model_params,
    measures=[make_histogram],
    name="Money Model",
    agent_portrayal=agent_portrayal,
    space_drawer="altair"
)
# This is required to render the visualization in the Jupyter notebook
page

In the agent_portrayal function, I set that every agent will have a tooltip to see their unique_id.
If we run this script using solara:

If I click on "View Source" in the altair chart, the content is:

{
  "config": {"view": {"continuousWidth": 300, "continuousHeight": 300}},
  "data": {
    "values": [
      {"size": 50, "color": "tab:blue", "Unique_id": 3, "x": 0, "y": 1},
      {"size": 50, "color": "tab:blue", "Unique_id": 48, "x": 0, "y": 3},
      {"size": 50, "color": "tab:blue", "Unique_id": 26, "x": 0, "y": 5},
      {"size": 50, "color": "tab:blue", "Unique_id": 12, "x": 1, "y": 0},
     // ...
      {"size": 50, "color": "tab:blue", "Unique_id": 27, "x": 9, "y": 8}
    ]
  },
  "mark": {"type": "point", "filled": true},
  "encoding": {
    "color": {"field": "color", "type": "nominal"},
    "size": {"field": "size", "type": "quantitative"},
    "tooltip": [{"field": "Unique_id", "type": "quantitative"}],
    "x": {"axis": null, "field": "x", "type": "ordinal"},
    "y": {"axis": null, "field": "y", "type": "ordinal"}
  },
  "height": 280,
  "width": 280,
  "$schema": "https://vega.github.io/schema/vega-lite/v5.16.3.json"
}

What's new is the tooltip array element and the unique_id value on each agent.

[github-actions]

Performance benchmarks:

Model	Size	Init time [95% CI]	Run time [95% CI]
Schelling	small	🟢 -4.3% [-4.6%, -4.1%]	🔵 -0.5% [-0.7%, -0.3%]
Schelling	large	🟢 -5.5% [-5.9%, -5.1%]	🔵 -1.1% [-1.7%, -0.5%]
WolfSheep	small	🟢 -6.9% [-7.1%, -6.7%]	🔵 -1.1% [-1.2%, -1.0%]
WolfSheep	large	🟢 -6.8% [-7.1%, -6.6%]	🔵 -0.5% [-2.5%, +1.4%]
BoidFlockers	small	🟢 -4.4% [-4.9%, -3.9%]	🔵 -1.2% [-1.9%, -0.6%]
BoidFlockers	large	🔵 -2.5% [-3.0%, -2.0%]	🔵 -0.5% [-1.1%, +0.0%]

[rht]

For checking type, it is recommended with isinstance, see https://stackoverflow.com/questions/152580/whats-the-canonical-way-to-check-for-type-in-python:

first_by_key = all_agent_data[0][key]
if isinstance(first_by_key, int):
    ...
elif isinstance(first_by_key, ...):
    ...

[rht]

The naming is clearer with detect_tooltip_type.

[rht]

The [key for key in all_agent_data[0].keys() if key not in invalid_tooltips] here is a repetition from L57-59. They can precomputed once so that there is no repetition.

[rht]

There are Ruff errors that can't be autofixed in the pre-commit.ci.

[FoFFolo]

I started by changing type checks using isinstance, but then I discovered that there is a function inside the altair library that
does the job.
I replaced my implementations with an integrated function on Altair, called
infer_vegalite_type() which detect automatically the type for the tooltip,
which uses the infer_dtype() pandas function.
The altair function can be found in the altair library folder, in the path "altair/utils/core.py".

def infer_vegalite_type(
    data: object,
) -> Union[InferredVegaLiteType, Tuple[InferredVegaLiteType, list]]:
    """
    From an array-like input, infer the correct vega typecode
    ('ordinal', 'nominal', 'quantitative', or 'temporal')

    Parameters
    ----------
    data: object
    """
    typ = infer_dtype(data, skipna=False)

    if typ in [
        "floating",
        "mixed-integer-float",
        "integer",
        "mixed-integer",
        "complex",
    ]:
        return "quantitative"
    elif typ == "categorical" and hasattr(data, "cat") and data.cat.ordered:
        return ("ordinal", data.cat.categories.tolist())
    elif typ in ["string", "bytes", "categorical", "boolean", "mixed", "unicode"]:
        return "nominal"
    elif typ in [
        "datetime",
        "datetime64",
        "timedelta",
        "timedelta64",
        "date",
        "time",
        "period",
    ]:
        return "temporal"
    else:
        warnings.warn(
            "I don't know how to infer vegalite type from '{}'.  "
            "Defaulting to nominal.".format(typ),
            stacklevel=1,
        )
        return "nominal"

The altair proper function doesn't implement date object detection on strings, so the user has to specify the date object in the agent_portrayal function. E.g:

from datetime import datetime

def agent_portrayal(MoneyAgent):
    # ....
    return {"Date": datetime.now()}

[rht]

One last simplification, you can do

[
    alt.Tooltip(key, type=alst.utils.infer_vegalite_type([value])
    for key, value in all_agent_data[0].items()
    if key not in invalid_tooltips
]

[FoFFolo]

Okay done. Yeah I agree with you.

One last simplification, you can do

[rht]

I started by changing type checks using isinstance, but then I discovered that there is a function inside the altair library that
does the job.

I see, that simplifies the implementation a lot.

[rht]

Merged, thank you @FoFFolo.