fedops-dataset 0.3.3

Local-first dataset toolkit for multimodal federated learning artifacts (partition/feature/simulation)

fedops-dataset

fedops-dataset is a local-first dataset toolkit for multimodal FL experiments (FedMS2-v8 style).

It supports:

• raw data bootstrap (fetch-raw)
• partition/feature/simulation generation (create-v8)
• Python API for runtime-driven usage (FedOpsLocalDataset)

Python requirement: >=3.8

Install

pip install fedops-dataset

Local-First Quickstart

1) Fetch or setup raw data

# CREMA-D
fedops-dataset fetch-raw --dataset crema_d --data-root /path/to/fed_multimodal/data

# PTB-XL
fedops-dataset fetch-raw --dataset ptb-xl --data-root /path/to/fed_multimodal/data

# Hateful Memes: provide a prepared source folder, then symlink/copy into data-root
fedops-dataset fetch-raw \
  --dataset hateful_memes \
  --data-root /path/to/fed_multimodal/data \
  --hateful-memes-source-dir /path/to/hateful_memes_source \
  --hateful-memes-mode symlink

2) Validate raw roots

fedops-dataset check-raw-datasets \
  --data-root /path/to/fed_multimodal/data \
  --hateful-memes-root /path/to/fed_multimodal/data/hateful_memes

3) Generate v8 artifacts (alpha, ps, pm)

# Dry run first
fedops-dataset create-v8 \
  --dataset hateful_memes \
  --alpha 0.1 \
  --sample-missing-rate 0.2 \
  --modality-missing-rate 0.8 \
  --repo-root /path/to/fed-multimodal \
  --data-root /path/to/fed_multimodal/data \
  --hateful-memes-root /path/to/fed_multimodal/data/hateful_memes \
  --dry-run

# Real run
fedops-dataset create-v8 \
  --dataset hateful_memes \
  --alpha 0.1 \
  --sample-missing-rate 0.2 \
  --modality-missing-rate 0.8 \
  --repo-root /path/to/fed-multimodal \
  --data-root /path/to/fed_multimodal/data \
  --hateful-memes-root /path/to/fed_multimodal/data/hateful_memes

Python API (Runtime-Driven)

Direct local usage

from fedops_dataset import FedOpsLocalDataset

ds = FedOpsLocalDataset(
    dataset="hateful_memes",
    alpha=0.1,
    sample_missing_rate=0.2,   # ps
    modality_missing_rate=0.8, # pm
    repo_root="/path/to/fed-multimodal",
    data_root="/path/to/fed_multimodal/data",
    hateful_memes_root="/path/to/fed_multimodal/data/hateful_memes",
)

ds.prepare(dry_run=False)
partition = ds.load_partition()
simulation = ds.load_simulation()
client0_records = ds.client_records(0, use_simulation=True)

Flower-style runtime config usage

from fedops_dataset import FedOpsLocalDataset

run_config = {
    "repo-root": "/path/to/fed-multimodal",
    "data-root": "/path/to/fed_multimodal/data",
    "hateful-memes-root": "/path/to/fed_multimodal/data/hateful_memes",
}

# Simulation mode example (Flower simulation engine)
node_config = {"partition-id": 0, "num-partitions": 10}

ds = FedOpsLocalDataset.from_runtime_config(
    dataset="crema_d",
    alpha=0.1,
    sample_missing_rate=0.2,
    modality_missing_rate=0.2,
    run_config=run_config,
    node_config=node_config,
)

mode = ds.node_mode(node_config)  # "simulation"
records = ds.client_records_from_node_config(node_config, use_simulation=True)

Path Semantics

• Simulation mode:
  • detected when node_config has partition-id and num-partitions
  • client records can be resolved from partition-id
• Deployment mode:
  • if node_config has data-path, it is used as runtime data root
  • each node can point to its own local data path
• No hardcoded path is required:
  • pass run_config/node_config, CLI args, or env vars

Environment Variables

export FEDOPS_REPO_ROOT=/path/to/fed-multimodal
export FEDOPS_OUTPUT_DIR=/path/to/fed-multimodal/fed_multimodal/output
export FEDOPS_DATA_ROOT=/path/to/fed_multimodal/data
export HATEFUL_MEMES_ROOT=/path/to/fed_multimodal/data/hateful_memes

Optional HF Artifact Client

FedOpsDatasetClient remains available if you also host artifacts in a Hugging Face dataset repo. It is optional for local/original-data mode.

Maintainer Release

cd fedops_dataset
export TWINE_USERNAME=__token__
export TWINE_PASSWORD=<pypi-token>
./scripts/publish_pypi.sh