Perform machine learning on JSON APIs with TensorFlow.js

With REFORM and TensorFlow.js you can start training machine learning models on JSON APIs in minutes.

JSON APIs contain huge amounts of valuable data. You can find everything from call transcripts to meteorological data and from social media posts to banking data in JSON APIs. This way of accessing data is more commonly referred to as "REST APIs" and even just "APIs" as JSON is the modern standard for data. JSON owes this esteem to its numerous advantages such as how easily it maps to application and device data structures, compound documents and sparse fieldsets.

JSON is not a normal tabular data format. Unlike tabular data the structure of each piece of JSON is tailored to a specific purpose. For example a piece of JSON for a form about you and your pets would have a very different structure to a piece of JSON for a manufacturing dashboard. Machine learning models expect standard tabular data just as spreadsheets and charts do. In order to use JSON to train our models and make predictions we need to transform it into meaningful tables.

A screenshot of REFORM. The structure of a JSON dataset is presented as a filesystem which is being browsed and from which relational columns are being picked

REFORM lets you access as tables in TensorFlow.js for training and prediction. Simply provide the details of the API endpoints, browse even the most complex data as if it were a file browser and pick what you're interested in. Then simply provide the access link to tf.data.csv with the names of the columns you want to predict.

tf.data.csv("https://reform.example.com/api/tables/3a4f239c-984f-43f5-ab84-86cead94f0c8/live/dataset", { columnConfigs: { price: { isLabel: true } }});

REFORM magically transforms the latest data into a mathematically correct analytic ready table and feeds this into your models for training and prediction.

REFORM also supports use cases where data needs additional JOINs or GROUP BYs before being used for machine learning. REFORM will transform your data into tables in AWS Redshift, Google BigQuery, Snowflake, AWS Athena and MS SQL Server which support JOINs and GROUP BYs and can be used as data sources by TensorFlow.js.