数据库新演进:能边存数据边自我学习
来源标题:A new evolution of databases: Self learning Databases
一种新型数据库将数据存储与机器学习整合在一起,每次写入新数据都会同步更新学习状态,无需单独的训练流程。行数据与学习更新在同一事务中提交,重启后可从事先保存的状态恢复。存储、智能查询与解释共用同一套数据库 API。
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
Store your data, learn from every row, and ask intelligent questions. One database brings your data and machine learning together.
Every row is a learning step. New data becomes useful without a separate training pipeline.
Predict, find similar rows, flag unusual ones, discover groups, and anticipate what comes next.
Your records stay stored. Learning doesn’t replace the values you wrote.
Find the true nearest neighbors under the current distance metric. Approximate indexes can miss them.
Use explain to inspect contributing rows, distances, or scores, depending on the ability.
Rows and learning updates commit in the same transaction. A restart resumes from saved state.
Define your columns and choose what your database can do. Write records, then ask about a stored row or describe a new one.
Storage, learning, intelligent queries, and explanations share one database API.
customers.write([ {"id": "c-4", "plan": "pro", "visits": 32, "bought": 1} ])
Learning belongs to the database.Choose the columns that contribute.
columns={ "id": key, "plan": learn(category), "visits": learn(number), "bought": learn(number), "email": store(text), }
A write updates the rows and learned state in the same transaction. No separate model deployment.
customers.predict([ {"plan": "pro", "visits": 32} ])
Illustrative response[0.89]
Here, bought is a learned numeric target. The returned value is an estimate.
customers.explain.predict([ {"plan": "pro", "visits": 32} ])
Illustrative response excerpt"based_on": [ {"key": "c-1", "weight": 0.62}, {"key": "c-3", "weight": 0.38} ]
Data + learning + answersOne database.
Five ways to ask.
Choose the questions your application needs to answer. HeatherDB keeps the enabled abilities up to date as you write.
Give what you know about a row. Estimate a value you don’t, using what the database has learned.
payments.predict([ {"merchant": "transfer", "amount": 9200} ])
Illustrative response[{"value": "yes", "chance": 0.83}]
A predicted category and its estimated chance. Numeric targets return a number.
On matched public datasets, HeatherDB reached parity or better against the strongest tested specialist on all six primary quality metrics.
Aphorion Labs report, November 12, 2026. Complete learning mode. Parity means within 2% on the primary metric. Freshness compares 23 ms from new-user writes to recommendations with a 2.7 s ALS refit; it is not general query latency.
Methods, infrastructure estimates, and scoring-speed tradeoffs are included in the report.
A database that learns with your data. Be among the first to build with it.
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。