nvflare.app_opt.xgboost.histogram_based_v2.sec.data_converter module

class AggregationContext(features: List[FeatureContext], sample_groups: Dict[int, List[int]])

Bases: object

class DataConverter

Bases: object

decode_aggregation_context(buffer: bytes, fl_ctx: FLContext) → AggregationContext

Decode the buffer to extract aggregation context info

Parameters:

• buffer – buffer to be decoded

• fl_ctx – FLContext info

Returns: if the buffer contains aggregation context, return an AggregationContext object;

otherwise, return None

decode_gh_pairs(buffer: bytes, fl_ctx: FLContext) → List[Tuple[int, int]]

Decode the buffer to extract (g, h) pairs.

Parameters:

• buffer – the buffer to be decoded

• fl_ctx – FLContext info

Returns: if the buffer contains (g, h) pairs, return a tuple of (g_numbers, h_numbers);

otherwise, return None

decode_histograms(buffer: bytes, fl_ctx: FLContext) → List[float]

Decode the buffer to extract flattened histograms

Parameters:

• buffer – buffer to be decoded

• fl_ctx – FLContext info

Returns: if the buffer contains histograms, return the flattened histograms

otherwise, return None

encode_aggregation_result(aggr_results: Dict[int, List[FeatureAggregationResult]], fl_ctx: FLContext) → bytes

Encode an individual rank’s aggr result to a buffer based on XGB data structure

Parameters:

• aggr_results – aggregation result for all features and all groups from all clients group_id => list of feature aggr results

• fl_ctx – FLContext info

Returns: a buffer of bytes

encode_histograms_result(histograms: List[float], fl_ctx: FLContext) → bytes

Encode flattened histograms to be sent back to XGBoost

Parameters:

• histograms – The flattened histograms for all features

• fl_ctx – FLContext info

Returns: a buffer of bytes

class FeatureAggregationResult(feature_id: int, aggregated_hist: List[Tuple[int, int]])

Bases: object

class FeatureContext(feature_id, sample_bin_assignment, num_bins: int)

Bases: object