Hash sharding validates the end-to-end correctness of Proxima CE by comparing recall rates between Proxima CE and the Proxima CE recall tool across multiple search methods and data types.
Test conclusion
Across all tested data types and search methods, Proxima CE recall rates match the recall tool results closely. The correctness test meets expectations.
Test procedure
Design the test method
Data preparation
Prepare random datasets of three data types: FLOAT, BINARY, and INT8.
For Proxima CE: convert datasets into MaxCompute tables.
For the C++ baseline: process data using the bench performance test tool provided in the Proxima kernel.
Algorithm comparison
Run graph search, hierarchical clustering search, Satellite System Graph (SSG), graph clustering (GC), quantized clustering (QC), and linear search against each dataset to collect recall rates for both Proxima CE and the C++ baseline. Set Top K to 100. The recall tool of Proxima CE calculates the recall rate from 100 sample records in the query table. The primary comparison between Proxima CE and the recall tool is based on the linear search method.
Prepare data and parameters
Datasets
Each query table extracts 100 data records from the doc table.
| Data type | Dimensions | Records | Value range |
|---|---|---|---|
| FLOAT | 128 | 100,000 | (0,1) |
| INT8 | 128 | 100,000 | (-128,127) |
| BINARY | 512 | 100,000 | 0/1 |
Search parameters
| Search method | Parameters |
|---|---|
| Graph search |
|
| Hierarchical clustering search |
|
| Satellite System Graph (SSG) |
|
| Graph clustering (GC) |
|
| Quantized clustering (QC) |
|
| Linear search | — |
View test results
FLOAT, SquaredEuclidean distance
| Search method | Proxima CE recall rate | Recall tool recall rate |
|---|---|---|
| Graph search | 89.03% | 88.62% |
| Hierarchical clustering search | 98.91% | 98.14% |
| Satellite System Graph (SSG) | 96.00% | 95.76% |
| Graph clustering (GC) | 97.87% | 97.64% |
| Quantized clustering (QC) | 97.70% | 97.77% |
| Linear search | 100% | 100% |
INT8, SquaredEuclidean distance
| Search method | Proxima CE recall rate | Recall tool recall rate |
|---|---|---|
| Graph search | 89.89% | 89.93% |
| Hierarchical clustering search | 98.27% | 97.69% |
| Satellite System Graph (SSG) | 95.58% | 95.75% |
| Graph clustering (GC) | 97.72% | 97.36% |
| Quantized clustering (QC) | 97.68% | 97.71% |
| Linear search | 100% | 100% |
BINARY, Hamming distance
| Search method | Proxima CE recall rate | Recall tool recall rate |
|---|---|---|
| Graph search | 85.33% | 88.09% |
| Hierarchical clustering search | 91.45% | 95.27% |
| Satellite System Graph (SSG) | 75.89% | 77.83% |
| Graph clustering (GC) | 90.01% | 93.99% |
| Quantized clustering (QC) | 90.51% | 93.78% |
| Linear search | 100% | 100% |
Analyze the results
For FLOAT and INT8 data types, Proxima CE recall rates are consistent with recall tool results across all search methods — differences are within 1 percentage point. For BINARY data, differences are slightly larger (up to approximately 4 percentage points for some methods). The recall result for each search method and data type of Proxima CE is similar to the recall result obtained by using the recall tool.