feathub
feathub copied to clipboard
FeatHub - A stream-batch unified feature store for real-time machine learning
SlidingWindowKeyedProcessFunction.java: ```java if (timestamps.isEmpty() || timestamps.get(timestamps.size() - 1) < timestamp) { timestampList.add(timestamp); } else { // TODO: Use skip list to optimize the performance of inserting late data. for (int...
Feathub should allow materializating multiple `TableDescriptor`s, which may have different schema and keys, into the same redis or memory online store. Unless the feature tables have entries with the same...
FlinkProcessor's RedisSink should support writing features with timestamp to Redis, where only features with larger keys preserve in Redis. This function has not been supported so far because in a...
When a Feathub job is writing a map-typed or list-typed feature into Redis with FlinkProcessor, if the feature is an empty map or empty list, there will not be a...
Add Java implementation to evaluate Feathub expression so that Redis lookup source can directly process key_expr, and generify the `lookup_join_redis_source` method.