High-performance ML with Cloud Bigtable, TensorFlow, and Cloud TPUs!
Misha Brukman
[ bcc: google-cloud-bigtable-announce ]
Dear Cloud Bigtable users,
We are excited to announce that TensorFlow 1.10 shipped with support for Cloud Bigtable as a data source and sink! Customers can now access data in Cloud Bigtable directly from open-source TensorFlow using the new Cloud Bigtable C++ client. As of TensorFlow 1.11, this integration is compatible with Cloud TPUs in addition to CPUs and GPUs. We are excited about new workloads this integration unlocks, such as reinforcement learning and more flexible data pipelines.
Potential application domains for this integration include:
fraud analytics in finance
personalization/recommendations in retail, media, adtech/martech, etc.
time series prediction and anomaly detection in monitoring, IoT, etc.
reinforcement learning applications such as:
robotics
industrial control systems
game playing
and more!
To get started, take a look at the code samples on GitHub, follow our tutorial for how to train ResNet-50 on Cloud TPU with data streamed from Cloud Bigtable, and let your friends and colleagues know to try this out! If you are building interesting applications with Cloud Bigtable, TensorFlow, and GPUs or Cloud TPUs, please let us know — we would love to learn more about your use cases! |
We will be developing more sample code and demos in the near future and we look forward to what you will build with this integration!
Best,
Misha
on behalf of the Cloud Bigtable, TensorFlow, and Cloud TPU teams
