18 04 Adopting production ML systems. There's more to ML than implementing an ML algorithm. A production ML system involves a significant number of components, such as data collection, data verification, machine resource management, analysis tools, feature extraction, process management tools, configuration, monitoring, and serving infrastructure. 10 These components make up a large ecosystem; the model is only a single part . Firms need to fully adopt tools such as Kubeflow, CMLE, and TensorFlow, not because they build better models, but because the ecosystems they offer include model monitoring, prediction scaling, model rebuilding, error diagnosis, reporting, logging, and other functionality. To support its trading operations, Grasshopper, a proprietary trading firm based in Singapore, has been using the Tensorflow open-source ML framework (originally developed within the Google AI organization), and the Cloud Bigtable NoSQL database service (optimized for large analytical and operational workloads). "The result for us is a distilled dataset that we push into our own servers, where another machine learning bot or process listens to live market data and makes decisions while we trade," explains Tan T-Kiang, chief investment officer at Grasshopper11. The Nimble Buy-Side Case: Mana Partners • Overview: High-performance trading, asset management, and technology firm Mana Partners hosts many of its capabilities on the Google Cloud Platform. • The driver: Rather than build anything on-premises, the equities and statistical arbitrage-focused firm chose to build nearly everything it required on the cloud, and benefit from the developer and data analytics tools offered by Google Cloud. • The result: The cloud environment has been beneficial for the firm’s flexible working environment. Data sharing across the team is easy and permissioning is maintained to ensure that the right people are receiving the right data. “We’ve been on the cloud since we started, as we adopted the frame of mind early. We focus on our own specialization rather than trying to do things that have already been done by Google—they have the scale, why would we try and duplicate it? Plus, data encryption in the cloud environment means our data is secure.”- Vitaly Dukhon, Senior Vice President of Equities Strategy, Mana Partners 10https://developers.google.com/machine-learning/crash-course/production-ml-systems 11https://cloud.google.com/customers/grasshopper/