Transcript:
Hello, I’’m Robert Crowe. Today I will talk about Tensorflow Extended. It is also called TFX and how it can help you. Put the magical machine learning model into production. This video is real-world machine learning. The first episode of the Five-part series Will help you quickly use TFX. Create your own production machine learning channel. In Today’’s video, we will ask this question. Where is this TFX sacred? Let’s reveal the secret of Xin. When we think of machine learning, we usually only consider The model we can create now. After all, this is all research papers. What to follow? But when we want to use this magical model, We need to consider when making it available to the world. All the conditions needed to produce the solution. This is why we use TFX to establish a production channel. We can provide our amazing models to the world. Google created TFX Because we need it. And there is no other project available that can meet our needs. Google has done a lot of machine learning projects. Not only Google, but also all subsidiaries of Alphabet. Almost all the work we do includes machine learning. In fact, Tfx was not created by Google. The first machine learning channel framework. It evolved from an early trial and now it is the most. The default framework for Google’s machine learning production solutions. Now Google has created an open source version of TFX. For everyone to use. Not just Google Tfx has had a profound impact on our partners, Including Twitter, Airbnb and Paypal. When machine learning developers put models into production. What conditions do we need to consider First when we started planning to develop machine learning applications, We need to consider all general aspects of machine learning. This includes obtaining labeled data when we are doing supervised learning And make sure that our data set can cover well. Possible input space. We also want to minimize the dimensionality of our feature set. At the same time, maximize the prediction information it contains. We need to consider fairness. And make sure that our app is not disturbed by unfair prejudice. We also need to consider rare cases, Especially in applications such as healthcare. We may be rare. But important situations make predictions Finally. We need to consider that this will be a living solution. With the inflow of new data and changes in the environment and our data Lifecycle management plan. These solutions will be over time. And evolve. In addition, we also need to remember. We are putting the software application into production. This means we still have to face. All requirements for production software applications, Including scalability, consistency, modularity And testability, as well as safety and protection. We are now more than just training models. In and of itself, these are Challenges faced by production software deployment. We cannot forget them because we are doing machine learning. How do we meet these requirements? And put our amazing new model into production. We will not pretend that we have all the answers. This is an evolving field within the machine learning community. We welcome your contributions. If you are facing the production environment Machine, learning challenges are interested in a deeper discussion. This is a good paper. It tells all the details about TFX. TFX allows you to create production machine learning channels, including Production software deployment and best practices. Many requirements. It will first extract your data and pass data verification. Characteristic engineering, training evaluation and service process. In addition to TensorFlow itself, we also contribute to the machine learning channel. Every major stage, TensorFlow data verification, TensorFlow conversion and TensorFlow model analysis. Create a library. Tfx uses a series of channel components. These components use the orange. These libraries allow you to create Own component. In order to combine all of this, we store for the channel. Some things like configuration and orchestration, create some horizontal layers. These levels are important for management And optimize the channel and on it. Very important to the running application. We will discuss more in subsequent episodes. So far, this information should give you an idea Discussion when we want to use TFX to implement production machine learning channels. In the next video, we will discuss how the TFX channel actually works. For more information about Tfx, please visit tensorfloworg/tfx. Don’’t, forget to leave a comment below and press like Thanks for watching.