jointview

A small marimo app for looking at two price or NAV series of a Polars DataFrame at once.

Every numeric column is a series you can pick, one on the left and one on the right. The middle holds both as two lines on one pair of axes; underneath each dropdown sits the summary of exactly the rows in that plot.

┌──────────┬───────────────────────────┬──────────┐
│ left     │      two-line chart       │ right    │
│ dropdown │   (indexed to 100)        │ dropdown │
│ ─────────│                           │──────────│
│ summary  │                           │ summary  │
│ table    │                           │ table    │
└──────────┴───────────────────────────┴──────────┘

Both lines share one y-axis: a second scale would invent a relationship that is not in the data. So the default is to index both series to 100 at the first date they have in common, which is how a fund priced at 1.02 and one priced at 1,450 end up comparable. The switch above the plot turns that off when the levels already share a scale.

Run it

There is nothing to install: the app ships with the package, and uvx fetches both for the length of the run.

uvx jointview                            # generated demo frame
uvx jointview navs.parquet               # your own data
uvx jointview navs.parquet --height 900  # taller plot for a taller screen

Until the package is on PyPI, point uvx at the repository — or at a checkout of it:

uvx --from git+https://github.com/jebel-quant/jointview jointview navs.parquet
uv run jointview navs.parquet            # from a clone
uv run jointview --edit                  # open the notebook itself

--edit only makes sense from a clone: it opens the notebook marimo is serving, and under uvx that is a copy in a throwaway environment.

Arguments after a bare -- belong to marimo rather than to the app, which is how the server itself is configured:

uvx jointview navs.parquet -- --port 8080 --headless

The plot takes the full width of the window and the tables sit either side of it. Its height is the one thing the page cannot work out for itself — 700px suits a laptop, and --height is there for a monitor that has more to give.

The file reads as .parquet, .csv, .tsv, .json, .ndjson, .arrow, .ipc and .feather. The first temporal column becomes the x-axis; without one the rows are numbered. Every numeric column is offered as a series. Given no file you get jointview.data.demo_frame(): daily NAVs for six made-up funds that share a market factor and start anywhere between 1 and 1,450.

The summary

Both tables are computed from the common sample — the dates where both series are present — so the numbers always describe the lines you are looking at.

A figure that cannot be formed — a Sharpe ratio for a flat series, a growth rate for a series that starts at zero — shows as — rather than blanking the table.

Use the pieces on their own

Neither the chart nor the statistics need marimo:

import polars as pl
from jointview import line_chart, summary, metrics

frame = pl.read_parquet("navs.parquet")
line_chart(frame, "tech_fund", "balanced").save("lines.html")

summary(frame["tech_fund"])              # a formatted two-column frame
metrics(frame["tech_fund"])["Sharpe ratio"]   # the raw number

line_chart is a plain Altair chart: two 2px lines, a legend and a label at the end of each line, and a crosshair that reads both series at the hovered date. Curves longer than max_points (4,000 by default) are thinned by a fixed stride — the last point always survives, so the endpoints and the summary agree.

Develop

uv sync
uv run pytest