Run away from takov customs to dig

1. In this game, all game currency can't be cashed, but you can buy gold. But the official ban on this kind of thing may be called

2. Now, there are many kinds of boxes that can be used for many purposes. The chance of finding a normal map is very slim. I went to the Xie teaching room for tens of thousands of times, and proced two gun boxes and one bullet box. Even if you encounter them, you may not be able to put your backpack down. Please find a merchant to exchange your goods

3. In short, the dead, even the things in the safety box can not be hung in the flea market

4. It can be registered with two different mobile phones and can be reused.

5. Pytorch is essentially a substitute for numpy, and supports GPU with advanced functions, which can be used to build and train deep neural networks. If you are familiar with numpy, Python and common deep learning concepts (convolution layer, loop layer, SGD, etc.), you will be very easy to use pytorch

tensorflow can be regarded as a programming language embedded in Python. The tensorflow code you write will be compiled into a graph by Python and run by tensorflow execution engine. I've seen a lot of novices who are bothered by the increased indirect layer. For the same reason, tensorflow has some additional concepts to learn, such as session, graph, variable scoping, and placeholder

in addition, more template code is needed to make a basic model run. Therefore, tensorflow's starting time must be longer than pytorch<

graph creation and debugging

winner: pytorch

Creating and running computational graphs may be the most different part of the two frameworks. In pytorch, the graph structure is dynamic, which means that the graph is built at run time. In tensorflow, the graph structure is static, which means that the graph is first "compiled" and then run

as a simple example, in pytorch, you can write a for loop structure with standard Python syntax

6. What are the advantages of pytorch dynamic computing graph; Torch calls himself numpy in the field of neural networks, because he can put the tensor generated by torch in the GPU to speed up the operation (if you have the right GPU), just like numpy will put array in the CPU to speed up the operation. There is a free switch between torch and numpy: import torch import numpy as NP NP_ data = np.arange(6).reshape((2, 3)) torch_ data = torch.from_ numpy(np_ data) tensor2array = torch_ data.numpy() print( ;; nnumpy array:;, np_ data, [[0 1 2], [3 4 5]] ;; ntorch tensor:;, torch_ data, 0 1 2 ; n 3 4 5 [torch.LongTensor of size 2x3] ;; ntensor to array:;, Mathematical calculation in tensor2array, [[0.12], [3.45]]) Python: 160; Many mathematical calculation functions in Python are the same as those in numpy. ABS absolute value calculation data = [- 1, - 2, 1, 2] tensor = torch. Floattensor (data) is converted into 32-bit floating-point tensor print (;; nabs;, ;; nnumpy: ;, np.abs(data), [1 2 1 2] ;; ntorch: ;, Torch. ABS (tensor) [1 2] sin trigonometric function 2. Variable variable 160; The variable of Python is equivalent to a wrapper. If you want to transfer data to the graph built by python, you need to package the data with variable first. The packaged variable has three attributes: data, creator and grad. Why is pytorch so popular

7. The difference between the two codes is that they have different effects on the device at run time
the data is run when the code is programmed, and the running speed of the latter is faster than that of the former.