Episode Length Plot Calculator

Explore episode duration trends, compare training runs, detect outliers, apply smoothing, calculate statistics, and export clear reinforcement learning performance reports instantly online with confidence.

Analysis Results

Episode Length Plot

Distribution and Run Comparison

Training Trend Analysis

Interactive Results Table

Episode Length Smoothed Rolling Mean Deviation Reward Success Truncated Outlier Run

Input Data

Enter CSV rows using: episode,length,reward,success,truncated,run. Only episode and length are required.

Experiment Settings

Plot Configuration

Smoothing and Outlier Detection

Formula Used

Mean episode length: L̄ = (1 / n) × ΣLᵢ
Moving average: MAₜ = (1 / w) × ΣLᵢ for the selected window.
Sample standard deviation: s = √[Σ(Lᵢ − L̄)² / (n − 1)]
Percentage change: ((late mean − early mean) / early mean) × 100
Total environment steps: S = ΣLᵢ

How to Use

  1. Paste episode data or upload a CSV file.
  2. Choose experiment, smoothing, plotting, and outlier settings.
  3. Select comparison options for multiple runs or seeds.
  4. Press Calculate and Plot to generate results.
  5. Review statistics, trends, outliers, and run comparisons.
  6. Export the chart, processed data, or printable report.

Example Data Format

episodelengthrewardsuccesstruncatedrun
1423800Seed 1
2575100Seed 1
3847710Seed 1

Interpretation Guide

Episode length counts environment steps before termination or truncation. Improvement depends on the task objective. Longer episodes help survival tasks, while shorter episodes may help goal-reaching tasks.

Strong trends should be checked beside reward and success rate. Time limits can create artificial plateaus. Compare multiple seeds before declaring convergence.

Smoothing reveals broad direction but can hide instability. Keep the raw series visible for diagnosis. Outliers may represent rare failures, successes, or logging problems.

Frequently Asked Questions

What is episode length?

It is the number of environment steps completed during one episode.

Is a shorter episode always better?

No. The preferred direction depends on the environment and learning objective.

What does truncation mean?

Truncation ends an episode because of an external limit, usually maximum steps.

Why use smoothing?

Smoothing reduces short-term noise and makes the training direction easier to inspect.

Which smoothing method should I choose?

Moving averages are simple, while EMA responds faster to recent changes.

How are outliers detected?

The calculator supports Z-score, IQR, modified Z-score, and custom thresholds.

Can I compare multiple runs?

Yes. Add a run name in the sixth column for each observation.

What does confidence band mean?

It estimates variation around the cross-run mean using aligned episode values.

Can I analyze rewards too?

Yes. Include reward values and select the reward relationship chart.

Does the calculator store my data?

Settings can be stored locally in your browser. Uploaded data stays client-side.

Related Calculators

Episode Reward CurveAverage Reward CurveExploration Rate CurveValue Function HeatmapPolicy HeatmapQ-Value PlotState-Transition DiagramRegret CurveSuccess Rate Curve

Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.