matplotlib Component Library

The expops-matplotlib package provides pre-built chart components that generate plots directly from pipeline metrics — without writing any Python chart code. Charts are defined as pipeline processes and read metrics via probe paths.

Installation

Install into the same environment as the expops CLI:

pip install expops-matplotlib

Note: The matplotlib runner executes in-process with expops, not in a separate reporting virtualenv. Both expops and expops-matplotlib must be installed in the same environment where you invoke expops run.

Available Components

matplotlib.BarChart.render

Generates a bar chart comparing a single metric across multiple probe paths (e.g. comparing train vs. eval accuracy for several models).

Required config keys:

• probe_paths — XPath selectors mapping display labels to pipeline processes (see Probe paths)
• parameters.probe_bar_metric — the metric name to read from each probe path

Optional parameters:

Parameter	Type	Description
title	string	Chart title
grid	bool	Show grid lines (default: false)
ylim	[min, max]	Y-axis limits
probe_bar_metric	string	Required. Metric name to plot per probe path

- name: "plot_metrics"
  component: "matplotlib.BarChart.render"
  probe_paths:
    train: "//*[@name='train_model']"
    eval: "//*[@name='predict_model']"
    mlp_train: "//*[@name='train_mlp']"
    mlp_eval: "//*[@name='predict_mlp']"
  parameters:
    title: "Accuracy (LR vs MLP)"
    grid: true
    ylim: [0, 1]
    probe_bar_metric: accuracy

---

matplotlib.Histogram.render

Generates a histogram from a metric logged across multiple steps (e.g. per-epoch training loss from iterative_logging).

Required config keys:

• probe_paths — XPath selectors for the process(es) to read metrics from
• parameters.probe_metric — the metric name whose values form the histogram

Optional parameters:

Parameter	Type	Description
title	string	Chart title
grid	bool	Show grid lines (default: false)
bins	int	Number of histogram bins (default: 10)
probe_metric	string	Required. Metric name to read from probe paths

- name: "mlp_loss_histogram"
  component: "matplotlib.Histogram.render"
  probe_paths:
    mlp_train: "//*[@name='train_mlp']"
  parameters:
    title: "MLP Train Loss Histogram"
    grid: true
    bins: 20
    probe_metric: train_loss

This is most useful in combination with iterative_logging on sklearn components that log a metric at each training step.

---

Probe Paths

Matplotlib components use the same XPath-based probe path system as custom @chart() functions. Each key in probe_paths becomes a display label; the XPath value selects which pipeline process to read metrics from.

Common patterns:

probe_paths:
  train: "//*[@name='train_model']"          # by process name
  eval: "//*[@name='predict_model']"
  p1_train: "//*[@partition='p1']/*[@name='train_model']"  # by partition

See Probe paths for the full XPath syntax.

---

Placing Chart Components in the Pipeline

Chart components are declared as regular processes in experiment.pipeline.processes. Add them to the end of your process_adjlist so they run after the processes they depend on:

experiment:
  pipeline:
    process_adjlist: |
      data_source split_data
      split_data train_model
      split_data predict_model
      train_model predict_model
      predict_model plot_metrics    # chart runs after evaluation
    processes:
      - name: "train_model"
        component: "sklearn.LogisticRegression.fit_predict"
        # ...

      - name: "predict_model"
        component: "sklearn.LogisticRegression.predict"
        # ...

      - name: "plot_metrics"
        component: "matplotlib.BarChart.render"
        probe_paths:
          train: "//*[@name='train_model']"
          eval: "//*[@name='predict_model']"
        parameters:
          title: "Train vs Eval Accuracy"
          grid: true
          ylim: [0, 1]
          probe_bar_metric: accuracy

---

matplotlib vs. Custom @chart() Functions

	expops-matplotlib component	Custom @chart() function
Setup	pip install expops-matplotlib	Write Python in src/charts/
Config	component: "matplotlib.BarChart.render"	reporting.static_entrypoint
Customisation	Parameters only	Full matplotlib API
Best for	Standard bar charts, histograms	Custom layouts, multi-panel figures

For full control over chart appearance, see Reporting — Static Charts.