Core concepts

Everything you see in the portal maps to a small set of concepts arranged in a hierarchy. Understanding them makes the rest of the portal straightforward.

The hierarchy

Workspace                      team / account-level container
└── Project                    a collection of related runs
    └── Run                    one training session (one execution of your script)
        ├── Attributes         metadata: hyperparameters, config, system info
        ├── Metrics (series)   the time-series your code logs (loss, accuracy, reward, …)
        └── Environments       reinforcement learning only: an instance of the world an agent acts in
            └── Episodes       one run-through within an environment
                └── Steps      one transition (state → action → reward); per-step metrics

Definitions

Workspace : The top-level container, usually one per team or account. It holds projects, members, and API keys. Manage it under Workspaces & Members.

Project : A named grouping of runs — typically one model, dataset, or research effort. See Projects.

Run : A single training session — one execution of your code. A run is the central object you analyze: it has attributes (metadata) and metrics (time-series). Runs are listed in the runs table.

Attribute : A piece of run metadata stored as a typed key-value pair (string, number, boolean, date, or set of strings). Hyperparameters from your config, plus system fields, are attributes. You browse them on a run’s Overview and can show them as columns or filters in the runs table.

Metric / series : A named time-series of numeric values that your code logs over the course of a run (for example loss). The portal renders metrics as charts. The full definition — how a series is identified by its name, scale, and labels — is covered in Metrics, series & attributes.

Environment (reinforcement learning) : A single instance of the world an agent interacts with during an RL run, identified by an env_id; a run can have many running in parallel. Each environment produces episodes and per-step data on their own scales. See Environments and Reinforcement learning concepts.

Episode (reinforcement learning) : One run-through within an environment, indexed by an episode number — a sequence of environment steps from reset until termination. Episode-level metrics (such as episode_reward) are summarized per episode.

Environment step (reinforcement learning) : One transition in the environment — a single state → action → reward step. Per-step metrics are logged on the environment-step scale (high-frequency). See Reinforcement learning concepts.

How the portal views map to concepts

• The runs table lists runs and shows attributes (and metric summaries) as columns.
• A run’s Overview shows its attributes.
• Charts and Dashboard visualize a run’s metrics.
• Compare and Side-by-side overlay metrics across multiple runs.
• Environments views show episode and per-step data for RL runs, broken down per environment.

Next: Metrics, series & attributes.