Direct logging

Push results to Braven from your own scripts — wandb-style init / config / summary / upload / finish.

Use this when you already have analysis code and want to log results directly — no folder watcher, no pipeline.

import braven

braven.init(name="My Experiment", project="TCS3448")
braven.config("learning_rate", 0.001)
braven.summary("accuracy", 0.94)
braven.upload("plot.png")
braven.finish()

braven.init()

Creates a new Experiment and sets it as the active run.

name · str | None

Experiment name shown in the UI.

notes · str | None

Optional free-text description.

project · str | None

Project name as shown in the app. Required if you belong to more than one project.

company · str | None

Company name — only needed to disambiguate when the same project name exists under two companies.

Returns a Run. Raises RuntimeError if neither company nor project is given, or if the name doesn’t match anything you have access to.

braven.init() is dual-purpose: called with name=/project=/company= (or no pipeline-specific kwargs) it does direct logging. Called with experiment_id=/api_url=/etc. (the shape the worker uses) it instead does pipeline-mode setup — see Pipeline scripts. You won’t hit this ambiguity in your own scripts; it only matters if you’re reading the SDK source.

Run methods

Everything below works identically as a module-level call (braven.config(...)) against the currently active run, or as a method on the Run object init() returns.

run.config(key, value) · → None

Log an input/configuration parameter, experiment-level.

run.summary(key, value) · → None

Log an output metric, experiment-level.

run.device(key, type=None) · → Device

Return a device-scoped handle — see Devices. Memoised per run: repeated calls with the same key return the same handle, and the first call’s type wins.

run.plot_series(name, y, x=None, x_label=None, y_label=None, mode='line') · → None

Log a labeled data series as an interactive plot. Call again with the same name and a different y_label to group multiple traces onto one chart.

run.upload(path_or_fig, name=None) · → None

Upload a file, or a live matplotlib Figure. For a Figure: saves the PNG (same as fig.savefig()) and best-effort extracts an interactive companion from its line/scatter data — see Series vs. Figures.

run.finish() · → None

Flush metadata. Idempotent — safe to call once at the end via braven.finish().

Device methods

Obtained via run.device(key) or module-level braven.device(key).

dev = braven.device("SENSOR-4471", type="Photodiode")
dev.config("bias_voltage", 3.3)          # alias for log_config
dev.summary("SNR", 14.2)                 # alias for log_summary
dev.log_series("response", values)
dev.plot_series("response_curve", y=signal, x=wavelength)

dev.* calls are tagged with that device’s key; the same metric name ("SNR") stays identical across every device — the device is a separate coordinate, never a name suffix like SNR_dev1.

Multiple devices in one experiment

for sensor_id, snr in results.items():
    braven.device(sensor_id).log_summary("SNR", snr)

braven.summary("max_device_mismatch", spread)   # experiment-level, not tied to one device

Full cheat sheet

braven.login(url, key)                       # once (or: python -m braven login)
braven.init(name=..., project=..., company=...)
braven.config(key, value)                    # input parameter
braven.summary(key, value)                   # output metric
braven.device(key, type=None)                # device-scoped handle
braven.plot_series(name, y, x=..., y_label=...)  # interactive chart
braven.upload(path_or_fig, name=...)         # file or matplotlib Figure
braven.finish()