– Is your first-phase, second-phase, or global ranking performing well?
– Is combining #BM25 with #embeddings work?
– Does your fieldset lacks some important fields?
– Is the linguistic stemming working?
– Which embedding is used? With exact or approximate near neighbours?
Fortunately, Vespa provides a tracing mode with seven levels of details https://lnkd.in/dU9J-Err:
– Basic tracing in container
– Basic tracing, more details
– Basic tracing, even more details
– Include timing info from backend
– More timing info from backend
– Include query blueprint
– Include query execution tree
But it is also convenient to get those traces within your application, in real time, without too much trouble digging in the documentation.
For its WooCommerce‘s integration to Vespa.ai, WPSOLR has integrated the trace levels inside the dashboard, An admin can use the search, and spy on the Vespa results, the Vespa query, and the Vespa query traces.
You can see one example in the attached screenshot of Vespa.ai demo with ANN search https://lnkd.in/dHtMX3Ss
WooCommerce + Vespa.ai and query tracing: wpsolr.com
