Python API catalog

Search classes and functions by name or module. Documentation is generated from the Python source without importing training dependencies.

93 of 93 APIs

Search Python API

Name	Description
AugmentationConfig	Configurable augmentation and abstention regimes applied in training.
AugmentationConfig	Tier-1 PPG augmentation settings for cross-domain robustness.
BandMetricsConfig	Band-limited evaluation configuration.
BandMetricsConfig	Band-limited evaluation configuration.
bandpass_filter_batch	Apply a physiokit bandpass filter across a batch of signals.
BetaAnnealConfig	Schedule for the RVQ commitment-loss weight (beta).
build_augmenter	Create augmentation pipeline.
build_decoder_2d	Build a configurable decoder that mirrors the encoder stages.
build_decoder_2d_spatial	Build a 2-D spatial decoder that mirrors the spatial encoder.
build_decoder_2d_ssm	Decoder that mixes time with diagonal SSM blocks at every spatial scale.
build_encoder_2d	Build a configurable encoder with ``2**num_stages`` downsampling.
build_encoder_2d_invres	Encoder using inverted-residual blocks with optional causal padding.
build_encoder_2d_spatial	Build a 2-D spatial encoder for spectrogram inputs.
build_hierarchical_adaptor_decoder_2d	Build a summed-latent decoder with per-level residual adaptors.
build_hierarchical_decoder_2d	Build a coarse + residual/detail decoder for per-level RVQ tensors.
build_ppg_tfrecord_cache	Build a subject-split TFRecord cache for PPG windows.
build_preprocessor	Create preprocessing pipeline: random crop + layer normalization.
build_rvq_autoencoder	Build encoder, bottleneck, decoder, and composite VQAutoencoder.
build_rvq_autoencoder_2d_spatial	Build 2-D spatial RVQ autoencoder for STFT spectrograms.
CacheConfig	Config for TFRecord cache dataset mode.
CacheConfig	Config for TFRecord cache dataset mode.
Codec	Uniform runtime interface for compressionKIT codecs.
collect_random_samples	Grab a subset of (input, target) tensors from a tf.data pipeline.
CompressionResult	`dataclass`
compute_compression_stats	Compute compression ratio and related statistics.
compute_ecg_hr_hrv	Detect R-peaks and compute HR/HRV from a single ECG signal.
compute_ppg_physiokit_metrics	Compute HR/HRV metrics for one PPG signal using physiokit.
compute_signal_metrics	Compute scalar reconstruction metrics on two aligned signals.
DataConfig	Data loading and preprocessing configuration.
DataConfig	Data loading and preprocessing configuration.
DerivativeLossConfig	First-derivative (smoothness) penalty on reconstruction.
DerivativeLossConfig	First-derivative (smoothness) penalty on reconstruction.
DwtLossConfig	Frequency-weighted MSE via Haar DWT subbands.
EcgRvqConfig	Top-level configuration for ECG RVQ training pipeline.
EncodedFrame	`dataclass`
EntropyPrior	Lightweight entropy prior using a causal CNN TFLite model.
EvaluationConfig	Post-training evaluation configuration.
EvaluationConfig	Post-training evaluation configuration.
export_decoder_tflite	Export decoder to INT8 TFLite and C header.
export_denoiser_tflite	Export a wavelet-gain denoiser to INT8 TFLite and C header.
export_encoder_tflite	Export encoder to INT8 TFLite and C header using helia_edge.
FilterConfig	Config for bandpass filtering of input or target signals.
FilterConfig	Config for bandpass filtering of input or target signals.
FilteredLossConfig	Lowpass-filtered MSE to focus on frequencies the model can represent.
generate_synthetic_ppg_batch	Generate synthetic PPG segments via physiokit.
HierarchicalRVQAutoencoder	RVQ autoencoder that routes per-level RVQ tensors to the decoder.
HybridSpihtCodec	Learned-denoiser + SPIHT runtime loaded from a deploy directory.
LabelTrustConfig	Reference-free label-trust loss weighting (``n0``-aware strong/weak labels).
load_codec	Hydrate a codec from a local deploy directory or HF repo id.
load_ppg_dataset	Load multiple EDF files into a stacked numpy array ``[N, num_samples]``.
load_ppg_file_splits	Return train/val/test EDF file splits for subject-level separation.
load_ppg_signal	Load a single PPG signal from an EDF file, resampling to *target_rate*.
load_ppg_splits	Load train/val/test numpy arrays of resampled PPG signals.
LongRecordingEvalConfig	Long-recording overlap-add evaluation for clinically meaningful HR/HRV.
LrScheduleConfig	Learning rate schedule configuration.
LrScheduleConfig	Learning rate schedule configuration.
make_ppg_inmemory_dataset	Build a tf.data pipeline from in-memory numpy arrays.
make_ppg_stream_dataset	Build a streaming tf.data dataset that samples random windows per subject.
make_ppg_tfrecord_dataset	Build dataset from TFRecord windows compatible with RVQ trainer.
ModelConfig	RVQ autoencoder model architecture configuration.
ModelConfig	RVQ autoencoder model architecture configuration.
OutputConfig	Output directory and logging configuration.
OutputConfig	Output directory and logging configuration.
PhysiokitMetricsConfig	PhysioKit HR/HRV evaluation configuration.
PpgRvqConfig	Top-level configuration for PPG RVQ training pipeline.
PRD	Percent RMS difference metric with optional energy normalization.
PrefixSupervisedVQAutoencoder	VQ autoencoder with auxiliary coarse-to-fine RVQ prefix losses.
resolve_deploy_dir	Return a local deploy directory, downloading from HF if needed.
RVQCodec	Lightweight RVQ autoencoder codec using LiteRT for inference.
RvqPrefixLossConfig	Auxiliary coarse-to-fine supervision for multi-level RVQ prefixes.
save_sample_artifacts	Save per-sample CSV, plot, and compute metrics for one evaluation sample.
SpectralLossConfig	Multi-scale spectral loss for frequency-domain reconstruction quality.
SpectralLossConfig	Multi-scale spectral loss for frequency-domain reconstruction quality.
SpihtCodec	Thin runtime wrapper around :class:`SpihtAcCodec`.
SpihtCodecConfig	Typed codec parameters for SPIHT.
StitchingEvalConfig	Long-recording stitching evaluation.
StreamingConfig	Config for subject-level streaming dataset mode.
StreamingConfig	Config for subject-level streaming dataset mode.
summarize_ecg_alignment	Compare ECG HR/HRV/peak-timing between paired original and reconstructed signals.
summarize_physiokit_alignment	Compare physiokit HR/HRV metrics between original and reconstructed signals.
summarize_ppg_peak_alignment	Compare PPG pulse peak timing between paired original/reconstructed signals.
SyntheticMixConfig	Config for synthetic ECG mixing during training.
SyntheticMixConfig	Config for synthetic PPG mixing during in-memory training.
TrainingConfig	Training hyperparameters and schedule configuration.
TrainingConfig	Training hyperparameters and schedule configuration.
TransformConfig	Signal transform domain for compression input.
TransformConfig	Signal transform domain for compression input.
TruePRD	Normalized PRD metric (convenience alias).
TwoStageCodec	Two-stage compression: RVQ codec + entropy-coded prior.
UnifiedCacheConfig	Config for unified per-source TFRecord cache dataset mode.
UnifiedSourceConfig	One source in a unified multi-source training mix.
WandbConfig	Weights & Biases logging configuration.
WandbConfig	Weights & Biases logging configuration.