fabricatio-plot 0.1.13.dev0

An extension of fabricatio, which brings up the capability to plot dataframe with matplotlib

fabricatio-plot

Data visualization, synthetic data generation, and dataframe manipulation for the Fabricatio LLM agent framework.

---

Installation

pip install fabricatio[plot]
# or
uv pip install fabricatio[plot]

For optional file-format support:

pip install fabricatio[plot,excel]    # Excel I/O
pip install fabricatio[plot,parquet]  # Parquet I/O

Install with all Fabricatio components:

pip install fabricatio[full]

Overview

fabricatio-plot equips LLM agents with tools for matplotlib-based charting, programmatic synthetic data generation, dataframe CRUD operations, and file I/O. Toolboxes expose fine-grained functions agents can call; capabilities wrap them into higher-level LLM-driven workflows.

Key Components

Toolboxes

Low-level tools grouped by domain, exposed to agents via fabricatio-tool's ToolBox.

PlottingToolBox — matplotlib chart construction

Function	Description
create_figure(figsize, dpi)	Create a new matplotlib Figure
create_subplots(nrows, ncols, figsize)	Create a figure with a grid of subplots
plot_line(ax, x, y, ...)	Draw a line chart on given axes
plot_bar(ax, categories, values, ...)	Draw a vertical bar chart
plot_scatter(ax, x, y, ...)	Draw a scatter plot
set_labels(ax, title, xlabel, ylabel)	Set title and axis labels
set_legend(ax, location, fontsize)	Add a legend to the chart
configure_grid(ax, visible, linestyle, alpha)	Configure grid lines
save_plot(fig, save_path, dpi, transparent)	Save figure to file

DataSynToolbox — synthetic column generation

Function	Description
numeric_column(n_rows, low, high)	Uniform-distribution numeric column
normal_column(n_rows, mean, std)	Normally distributed numeric column
categorical_column(n_rows, categories)	Random categorical column
datetime_column(n_rows, start, end)	Random datetime column in a range
text_column(n_rows, prefix)	Sequential text column (item_0, item_1, …)
correlated_column(base_series, correlation)	Column correlated with an existing series
inject_missing(series, rate)	Inject NaN values at a given rate

DataCrudToolbox — dataframe create/read/update/delete

Function	Description
create_empty_dataframe(columns, dtypes)	Create an empty DataFrame with typed columns
add_computed_column(df, new_column, expression)	Add a column via pandas expression
get_row_by_index(df, index_value)	Retrieve a single row by index
fill_missing_values(df, column, strategy)	Fill NaN with mean, median, mode, or constant
transform_column(df, column, transformation)	Apply log, sqrt, square, or normalize
drop_columns(df, columns)	Remove specified columns
drop_rows_by_condition(df, condition)	Remove rows matching a query condition
rename_column(df, old_name, new_name)	Rename a single column
set_index_from_column(df, column, drop)	Set the DataFrame index from a column

DataIoToolBox — file-based I/O

Function	Description
load_csv(file_path)	Read CSV into DataFrame
load_excel(file_path, sheet_name)	Read Excel into DataFrame (requires excel extra)
load_parquet(file_path)	Read Parquet into DataFrame (requires parquet extra)
save_data(df, file_path, fmt)	Save DataFrame as CSV, Excel, or Parquet

Capabilities

Higher-level abstractions that orchestrate toolboxes with LLM reasoning.

• Plot — aggregates PlottingToolBox, DataCrudToolbox, and DataIoToolBox. The plot(requirement, data, output_spec) method accepts a natural-language requirement and delegates to its toolboxes to produce charts.
• SynthesizeData — uses LLM to generate structured data from a natural-language description. generate_header() produces column names; generate_csv_data() returns a DataFrame; synthesize_data() handles large datasets by batching parallel generation.

Actions

Concrete fabricatio-core Action implementations ready to wire into task pipelines.

• MakeCharts — takes a plot requirement (or the task's assembled prompt) and produces a chart, optionally saving to a specified path.
• MakeSynthesizedData — synthesizes a DataFrame from the task prompt and stores it under the key synthesized_data.
• SaveDataCSV — saves a DataFrame to a path (given explicitly or determined by the LLM from the task context).

Usage

Toolboxes are stateless and can be called directly:

from fabricatio_plot.toolboxes.plot import create_figure, plot_line, set_labels, save_plot

fig = create_figure(figsize=(10, 4))
ax = fig.subplots()
plot_line(ax, x=[1, 2, 3], y=[4, 5, 6], label="Series A")
set_labels(ax, title="Sample Chart", xlabel="X", ylabel="Y")
save_plot(fig, "output.png")

Capabilities drive LLM-orchestrated workflows:

from fabricatio_plot.capabilities.plot import Plot

handler = Plot()
result = await handler.plot("Create a bar chart of monthly sales from the CSV at data/sales.csv")

Actions plug into Fabricatio task graphs:

from fabricatio_plot.actions.plot import MakeCharts

action = MakeCharts(plot_requirement="Scatter plot of x vs y", chart_save_path="scatter.png")
await action.execute(task)

Dependencies

• fabricatio-core — core interfaces, configuration, and task model
• fabricatio-tool — ToolBox and Handle abstractions
• matplotlib — chart rendering
• numpy — array operations
• pandas — DataFrame manipulation
• Optional: openpyxl (Excel I/O), pyarrow (Parquet I/O)

License

MIT — see LICENSE