Pandas At Float

Somewhere in that DataFrame is exactly one number you need - a threshold, a best score, a single metric - and the whole rest of the graph wants it as a plain float. Pandas At Float is the node that reaches into the table by row and column labels, pulls out that one cell, and hands it to ComfyUI as a FLOAT that can plug into anything. It's the "get one value out" tool in a pack that mostly moves whole DataFrames around.

How it works

It's dataframe.at[row_index, column_label] returned as a float. The name is the mechanism: pandas' .at does fast, label-based single-cell access. You specify where with two text fields and two type switches, because your index and column labels can be strings or integers and the node needs to know which before it casts your typed input.

Inputs and outputs

• dataframe - the frame to read (DATAFRAME)
• row_index - the row's label (STRING)
• row_index_type - enum: string or int
• column_label - the column's label (STRING)
• column_label_type - enum: string or int
• Output: FLOAT

The type switches are the only gotcha on input. Set them to int if your labels are integers; leave them string for text labels. Pick wrong and you get a KeyError because "3" and 3 are different labels to pandas.

Why you'd bother

A single scalar unlocks the rest of the graph. Pull a computed best-value out of a stats table and feed it into a comparison, a prompt template, or a scalar input elsewhere; extract one number from a CSV and use it to drive a downstream parameter. This pack is built so tables can talk to the non-table parts of ComfyUI, and the Pandas At family is the bridge. If the cell isn't naturally a float, run the column through Pandas As Float first so the read doesn't choke.

Installing it

Ships with ComfyUI-Data-Analysis (Hide Inada / HowToSD.com). Manager: search Data Analysis → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/HowToSD/ComfyUI-Data-Analysis.git
mv ComfyUI-Data-Analysis data-analysis
pip install -r data-analysis/requirements.txt

No GPU, no models; pandas comes from requirements.txt.

Where people get burned

Label mismatch (string vs int) is the number-one footgun - double-check both type enums if you see a KeyError. And remember this reads by label, not position; for "third row, second column" you want Pandas Iat Float instead. Missing cells raise rather than returning NaN here, which is actually a favor - a loud failure beats a silent NaN poisoning your downstream math.