Skip to content

Venra provides a simple, high-level api for vespa.ai

License

Notifications You must be signed in to change notification settings

codycollier/venra

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

81 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

venra

Project Status: WIP Tests Release PyPI version

Venra provides a simple, high-level api for vespa.ai.

Venra targets subsets of Vespa's query, document, and system apis. It aims to encapsulate the complexity of dealing with the Vespa http interfaces, response behaviors, and json responses for common client tasks.

Venra is well suited for web backends, command line tools, and enrichment programs which need to retrieve, process, and update documents.

import venra

qdata = {}
qdata["yql"] = "select * from sources awesome_docs;"
response = venra.query.search(qdata)

docs = venra.query.extract_docs(response)
for r, doc in enumerate(docs):
    print(f"rank: {r} >> {doc.some_id} title: {doc.title}")

Note: This library is under active development and the api is currently unstable.

Installation

$ pip install venra

Usage

Basic Query:

import venra

# Build query
qdata = {}
qdata["yql"] = "select * from sources baz;"

# Run query
response = venra.query.search(qdata)

# Extract results via helpers
metrics = venra.query.extract_metrics(response)
docs = venra.query.extract_docs(response)

User Query and Grouping:

from pprint import pprint

from venra import config as vconfig
from venra import query as vquery


# Configure
user_query = "machine learning"
vconfig.vespa_host_app = "http://localhost:8080"

# Build query including a grouping
qdata = {}
qdata["yql"] = "select post_id, post_date from sources baz where userQuery()"
qdata["yql"] += f" | all(group(time.date(post_date)) order(-max(post_date)) max(32) each(output(count())) as(day_counts) );"
qdata["hits"] = 10
qdata["timeout"] = "3300ms"
qdata["model.queryString"] = user_query
qdata["model.type"] = "weakAnd"
qdata["presentation.summary"] = "full"
qdata["presentation.timing"] = "true"

# Run query
response = vquery.search(qdata)

# Extract results via helpers
metrics = vquery.extract_metrics(response)
groups = vquery.extract_groups(response)
myfacet = vquery.extract_group_pairs(groups, "day_counts", "count()")
docs = vquery.extract_docs(response)

# Query results ready for use in app
pprint(metrics)
pprint(myfacet)
pprint(docs)