当前位置：网站首页>Deep learning - multidimensional data and tensor

Deep learning - multidimensional data and tensor

2022-07-19 00:54:00 【InfoQ】

stay tensorflow in , Model input and convolution kernel are generally used 4-D Of tensor To express , For the process of building the model , It's all in multidimensional （ exceed 3 dimension ） Of tensor perhaps numpy  Array （ Weight value weights And deviation bias） Operate accordingly .

about 1 dimension ,2 dimension ,3 Dimension data , Can be understood visually , There is a corresponding reality reference . exceed 3 dimension , Data representation and manipulation , It's hard to understand . Nolan's movie 【 Interstellar through 】 The fifth dimensional space of , Rich imagination . Here, let's take a look at the representation of multidimensional data .

null
Express
Let's take a look first 1 dimension ,2 dimension ,3 Dimensional data representation
null
See , The rule of multidimensional data representation is 2 Each unit of dimensional data is 1 D data ,3 Each unit of dimensional data is 2 D data .  from shape Look at , The last one is 1 Number of units of dimensional data ,  The next to last is 2 In dimension 1 Number of dimensional units , And so on .

According to this Law ,4 Dimensional data is a collection of multiple 3D data ,5 Dimension data is multiple 4 A collection of dimensional data .

The data access
null
Data access of a dimension unit , Or reference 【 from shape Look at , The last one is 1 Number of units of dimensional data ,  The next to last is 2 In dimension 1 Number of dimensional units 】 This is to understand

tensorflow  Of concat  Connect and understand
  • concat It means merging ,tf.concat([v1,v2] , axis=3)  Merge  axis Indicates which dimension to merge and splice .
  •   tf.concat([v1,v2],axis=3)  # 1 Dimensional merging , Before the merger shape by (2,2,2,3)  After merger: (2,2,2,6)
null
  •   tf.concat([v1,v2],axis=2)  # 2 Dimensional merging , Before the merger shape by (2,2,2,3)  After merger: (2,2,4,3)

This article looks at tensorflow and numpy The construction and brief operation of multidimensional data in , Hopefully that helped .


null
原网站

版权声明
本文为[InfoQ]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/200/202207162321468357.html

边栏推荐

猜你喜欢

随机推荐