- Hands-On Deep Learning for Games
- Micheal Lanham
- 369字
- 2021-06-24 15:47:57
Monitoring training with TensorBoard
TensorBoard is essentially a mathematical graph or calculation engine that performs very well at crunching numbers, hence our use of it in deep learning. The tool itself is still quite immature, but there are some very useful features for monitoring training exercises.
Follow these steps to start monitoring training on our sample:
- You can monitor the training session by entering the following command into a new Anaconda or command window from the same directory/folder that you are running the sample from:
//first change directory to sample working folder
tensorboard --logdir=/tmp/autoencoder
- This will launch a TensorBoard server, and you can view the output by navigating your browser to the URL in italics, as shown in the window you are running TensorBoard from. It will typically look something like the following:
TensorBoard 1.10.0 at http://DESKTOP-V2J9HRG:6006 (Press CTRL+C to quit)
or use
http://0.0.0.0:6000
- Note, the URL should use your machine name, but if that doesn't work, try the second form. Be sure to allow ports 6000, and 6006 and/or the TensorBoard application through your firewall if prompted.
- When the sample is done running, you should see the following:

Auto-encoding digits using convolution
- Go back and compare the results from this example and the last example from Chapter 1, Deep Learning for Games. Note the improvement in performance.
Your immediate thought may be, "Is the increased training time we experienced worth the effort?" After all, the decoded images look quite similar in the previous example, and it trained much faster, except, remember we are training the network weights slowly by adjusting each weight over each iteration, which we can then save as a model. That model or brain can then be used to perform the same task again later, without training. Works scarily enough! Keep this concept in mind as we work through this chapter. In Chapter 3, GAN for Games, we will start saving and moving our brain models around.
In the next section, we take a more in-depth look at how convolution works. Convolution can be tricky to understand when you first encounter it, so take your time. It is important to understand how it works, as we will use it extensively later.
- 數(shù)據(jù)分析實戰(zhàn):基于EXCEL和SPSS系列工具的實踐
- 復雜性思考:復雜性科學和計算模型(原書第2版)
- 3D計算機視覺:原理、算法及應用
- 數(shù)字媒體交互設(shè)計(初級):Web產(chǎn)品交互設(shè)計方法與案例
- SQL優(yōu)化最佳實踐:構(gòu)建高效率Oracle數(shù)據(jù)庫的方法與技巧
- Oracle PL/SQL實例精解(原書第5版)
- 圖數(shù)據(jù)實戰(zhàn):用圖思維和圖技術(shù)解決復雜問題
- 高維數(shù)據(jù)分析預處理技術(shù)
- 達夢數(shù)據(jù)庫運維實戰(zhàn)
- Google Cloud Platform for Developers
- MySQL技術(shù)內(nèi)幕:SQL編程
- 區(qū)塊鏈+:落地場景與應用實戰(zhàn)
- Artificial Intelligence for Big Data
- Machine Learning for Mobile
- SQL進階教程(第2版)