MPL Scatter Plot

MPL Scatter Plot is the node for the question "is there a relationship between these two columns?" It takes a DataFrame, an x column, and a y column, and returns a Matplotlib scatter plot as a ComfyUI IMAGE - every row becomes a dot, and patterns you'd never spot in the table jump out immediately. Cluster, correlation, or total noise: one run and you know.

It's part of HowToSD's ComfyUI-Data-Analysis pack, sharing the exact input set and render pipeline with MPL Bar and MPL Line - the only difference is the Matplotlib call underneath. Same author, same philosophy: do the Pandas work in the graph, then see the result.

How it works

The wrapper resolves your typed column names against the DataFrame, then calls ax.scatter(df[x_column], df[y_column]) - one dot per row. Then the shared post-step takes over: title and axis labels applied, figure saved to a PNG buffer, converted to RGB, normalized to a [0,1] image tensor with a batch dimension. Standard ComfyUI IMAGE output. The x_tick_as_int toggle is there too, fixing the float-year-ticks problem (2002.5) via MaxNLocator(integer=True).

Because it's a scatter, the node expects many rows - the resolution of the chart depends on how much data you feed it. Ten rows give you a lonely little chart; a thousand rows show real structure.

Inputs and outputs

• dataframe (required, DATAFRAME) - your data, loaded or transformed.
• x_column_name / y_column_name (required, STRING) - the two columns to compare.
• title, x_axis_label, y_axis_label (required, STRING, default empty).
• x_tick_as_int (required, BOOLEAN, default false) - integer x ticks.
• IMAGE output - into Preview Image or Save Image.

When you'd reach for it

Exploration. Before you build a fancy aggregation or model, scatter the candidate columns against each other and check whether the relationship is even there. In the pack's terms: load the stats CSV, pick two numeric columns, plot, and immediately see if hits track with something else. It pairs naturally with Pandas Corr - scatter shows you the shape, corr gives you the number.

It's also the honest check before you trust a correlation coefficient. A single outlier can pump up a Pearson value while the scatter shows you it's one weird point. See it first.

Installing it

Pack install, once. ComfyUI Manager: search "Data analysis", install ComfyUI-Data-Analysis, restart, reload. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/HowToSD/ComfyUI-Data-Analysis
mv ComfyUI-Data-Analysis data-analysis   # README: examples expect this folder name
pip install -r data-analysis/requirements.txt

Dependencies: pandas, matplotlib (the rendering engine), seaborn, scipy, scikit-learn, openpyxl, lxml. No GPU - CPU Matplotlib. PyTorch nodes moved to ComfyUI-Pt-Wrapper in March 2025.

Gotchas

• Exact column names or a KeyError. Confirm names via Pandas Columns → a show node if you didn't author the file.
• NA values get dropped from the plot silently. If the chart looks thin, that's why - run Pandas Drop NA or Fill NA first if you want to control it.
• Column dtypes matter. Scatter on two text columns errors; both should be numeric. Cast with Pandas As Float if needed.