Designing and constructing the data pipeline

One of the most important requirements when it comes to training machine learning (ML) models and deep neural networks (DNNs) is having large training datasets with distributions (mostly unknown, which we learn about during ML or DNN training) from a given sample space so that ML models and DNNs can learn from this given training data and generalize well to unseen future or separated out test data. Also, a validation dataset, which often comes from the same source as the training set distribution, is critical to fine-tuning model hyperparameters. In many cases, developers start with whatever data is available—either a little or a lot—to train machine learning models, including high capacity deep ...

Get What's New in TensorFlow 2.0 now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.