Quick Start

This example shows the smallest useful HiveWatch setup: log a run with the local SSE emitter, persist the data, and serve the dashboard separately.

Minimal example

import hivewatch as hw
from hivewatch.emitters import SSEEmitter

hw.init(
    algorithm="FedAvg",
    emitters=[SSEEmitter(port=7070, serve_map=False)],
)

for round_num in range(5):
    hw.round_start(round_num)

    hw.log_client_update(
        client_id="client-1",
        round=round_num,
        local_accuracy=0.70 + round_num * 0.03,
        local_loss=0.90 - round_num * 0.08,
        num_samples=500,
        bytes_sent=8192,
        lat=41.88,
        lng=-87.63,
        country="US",
    )

    hw.log_round(
        round=round_num,
        global_accuracy=0.74 + round_num * 0.03,
        global_loss=0.85 - round_num * 0.07,
    )

hw.finish()

This creates two run artifacts under runs/:

• <run_id>.jsonl with the full event stream.

• <run_id>.map.json with map-ready metadata for replay and deferred viewing.

Serve the dashboard

hivewatch map run --runs-dir runs --port 7070

Open http://localhost:7070 to inspect the run.

Switch emitters

HiveWatch can attach more than one emitter to the same run:

import hivewatch as hw
from hivewatch.emitters import MLflowEmitter, WandbEmitter

hw.init(
    emitters=[
        WandbEmitter(project="my-fl-project"),
        MLflowEmitter(experiment="my-fl-project"),
    ],
)