dqm-ml-pytorch 2.0.0rc0

Python library designed to provide core dqml domain gap metrics, as well as common API shared by metrics

DQM-ML PyTorch

PyTorch-based metrics for DQM-ML V2. Provides advanced domain gap analysis for comparing dataset distributions.

Installation

pip install dqm-ml-pytorch

Note: dqm-ml-pytorch provides metric processors only — no CLI or job orchestration. Use directly via Python or with dqm-ml-job for YAML config execution.

Usage

Using Python Directly

import numpy as np
from dqm_ml_pytorch import DomainGapProcessor

# Create source and target embeddings (example data)
source_embeddings = np.random.randn(100, 2048).astype(np.float32)
target_embeddings = np.random.randn(100, 2048).astype(np.float32)

# Create and configure the processor
processor = DomainGapProcessor(
    name="domain_drift",
    config={
        "INPUT": {"embedding_col": "embedding"},
        "DELTA": {"metric": "mmd_linear"}
    }
)

# Compute statistics for both datasets
source_stats = processor.compute_batch_metric({"embedding": source_embeddings})
target_stats = processor.compute_batch_metric({"embedding": target_embeddings})

# Compute the domain gap delta
result = processor.compute_delta(source_stats, target_stats)
print(f"Domain Gap (MMD): {result['domain_gap_mmd_linear']}")

With dqm-ml-job

For running from a YAML config, install together with dqm-ml-job:

pip install dqm-ml-job dqm-ml-pytorch

Then use this config:

metrics_processor:
  domain_drift:
    type: domain_gap
    INPUT:
      embedding_col: "features"
    DELTA:
      metric: "mmd_linear"

Features

Metric	Full Name	Best For
FID	Fréchet Inception Distance	Image embeddings
MMD	Maximum Mean Discrepancy	General kernel-based comparison
Wasserstein	1D Earth Mover's Distance	1D distributions
KLMVN	KL-Divergence (Multivariate Normal)	Gaussian distributions

Output

Returns statistical distance values:

• domain_gap_fid
• domain_gap_mmd_linear
• domain_gap_wasserstein_1d
• domain_gap_klmvn_diag

Requirements

• torch
• torchvision
• scipy

Dependencies

DQM-ML is modular. For domain gap metrics:

# Minimal: use as library only
pip install dqm-ml-pytorch

# For YAML config execution
pip install dqm-ml-job dqm-ml-pytorch

# Full stack with all metrics
pip install dqm-ml-job dqm-ml-core dqm-ml-pytorch

See Also

• Domain Gap Documentation
• Configuration Guide