Couchbase Lite's Predictive Query API allows applications to leverage pre-trained, Machine Learning(ML) models to run predictive queries against data in embedded Couchbase Lite database in a convenient, fast and always-available way. These predictions can be combined with predictions made against real-time data captured by your app to enable a range of compelling applications. The Predictive Query API in Couchbase Lite is the first of it's kind in an embedded database. We had announced the Developer Preview version of Predictive Query API with Couchbase Mobile 2.5 last year. With Couchbase Mobile 2.7, we are happy to announce the General Availability of this feature.

The project is written in python 2.7 and Pytorch 1.0.1. If CUDA is available, it will be used automatically. The models can also run on CPU as they are not excessively big. We carried out homoscedastic and heteroscedastic regression experiements on toy datasets, generated with (Gaussian Process ground truth), as well as on real data (six UCI datasets). The heteroscedastic notebooks contain both toy and UCI dataset experiments for a given (ModelName).

We keep the BERT encoder unfrozen so that all weights are updated with every iteration. Given the number of trainable parameters it's useful to train the model on multiple GPUs in parallel. I used 4 Tesla K80's for about 4500 training samples. Just remember that to access any model attribute, you can access it using modelName.module.attribute I used Stochastic Gradient Descent with momentum as the optimizer and found that cycling both the learning rates and momentum really helped to get the training and validation losses down.

The python package pastalog has a node.js Once you have a server running, you can start logging your progress. Go to localhost:8120 and view your logs updating in real time. Note: If you want to compare models across batch sizes, a good approach is to pass to step the fractional number of times the model has seen the data (number of epochs). In that case, you will have a fairer comparison between a model with batchsize 50 and another with batchsize 100, for example.