wandb_buddy 0.1.0

Utilities for W&B experiment analysis

wandb_buddy

Utilities for W&B experiment analysis.

Installation

pip install wandb_buddy

Or install from source:

pip install -e .

Usage

import wandb_buddy as wb

# Tag-based filtering (simple)
df = wb.load_runs("your-entity", "your-project", tags=["experiment-v1"])

# Exclude certain tags
df = wb.load_runs("your-entity", "your-project", tags=["experiment-v1"], exclude_tags=["debug"])

# Raw filters (flexible)
df = wb.load_runs("your-entity", "your-project", filters={"config.lr": 0.001})

# Combined tag and raw filters
df = wb.load_runs("your-entity", "your-project", tags=["experiment-v1"], filters={"config.lr": 0.001})

# Include all run states (not just finished)
df = wb.load_runs("your-entity", "your-project", tags=["experiment-v1"], state="all")

# Custom timeout
df = wb.load_runs("your-entity", "your-project", tags=["experiment-v1"], timeout=10000)

The returned DataFrame contains: name, id, state, created_at, heartbeat_at, tags, sweep_id, plus all config and summary values from each run.

Development

pip install -e ".[dev]"