A list,

A deep learning framework is an interface, library, or tool that makes it easier and faster to build deep learning models without having to go into the details of the underlying algorithms. Deep learning frameworks use pre-built and optimized sets of components to define models, providing a clear and concise approach to model implementation. At present, the mainstream deep learning frameworks include TensorFlow, Keras, PyTorch and Caffe.

TensorFlow

TensorFlow, developed by researchers and engineers on the Google Brain team, is the most commonly used software library in the field of deep learning. This is an open source mathematical computing software using C++ language development, using Data Flow Graph (Data Flow Graph) form of interaction. TensorFlow has many components, the most prominent of which are: (1) Tensorboard: helps with effective data visualization using data flow diagrams (2) TensorFlow: Currently, there are many third-party libraries for encapsulation based on TensorFlow, among which Keras, Tflearn, TFSlim and TensorLayer are well-known.

Keras

Keras is written in Python and runs on TensorFlow (as well as CNTK and Theano). The TensorFlow interface is challenging because it is a low-level library, and some implementations may be difficult for new users to understand. Keras, on the other hand, is a high-level API developed for rapid experimentation. Therefore, if you want fast results, Keras automatically processes the core tasks and generates the output. Keras supports convolutional neural networks and recursive neural networks and can run seamlessly on both CPU and GPU. A common complaint of deep learning beginners is that they cannot understand complex models correctly. If you are one of those users, Keras is the right choice for you! Its goal is to minimize user action and make its model truly easy to understand.

Caffe

Caffe is another popular deep learning framework for image processing. It was developed by Yangqing Jia during his PhD at the University of California, Berkeley. Again, it’s open source! As the name suggests, it has exceptionally good support for convolutional networks and is also written in C++.

PyTorch

Developed by the Torch7 team, PyTorch differs from Torch in that, as the name suggests, it uses Python as the development language. PyTorch can be seen both as NUMPY with GPU support and as a powerful neural network with automatic derivation.

Second, PyTorch environment construction

1. Install Anaconda 3.5

Anaconda is a Python distribution for scientific computing. It supports Linux, Mac, and Windows systems. Anaconda provides package management and environment management functions to easily solve Python coexisting, switching, and third-party package installation problems.

1.1 download:

It can be downloaded directly from the official website of Anaconda, but because the server of Anaconda is in foreign countries, the download speed will be very slow. Here, it is recommended to use the image of Tsinghua University to download. Choose the right version for you to download. In my case, choose Anaconda3-5.1.0-Windows-x86_64.exe

1.2 installation

After downloading the installation, their choice of path is good, basically the default can.

After the installation is complete, configure environment variables for Anaconda. Open the control panel and choose Advanced System Settings > Environment Variables > System Variables to find Path, click Edit, and add the storage paths of the three folders (note that the three paths must be separated by semicolons), as shown in the figure:

At this point, Anaconda 3.5 for Windows is installed and set. Open the program and find Anaconda Navigator.

1.3 Orderly management environment

Here you need to check your own Python version, win+R enter CMD, enter Python to get the corresponding version information. (Path cannot be queried if the Python environment is not configured)

If you don’t know your Python version, win+R, go to the CMD console, type In Python and you’ll see your version.

Conda create -n pyTorch Python =3.8

After that, a prompt pops up, type Y, you can install. After the installation is successful, enter the following command:

conda info --envs

You can see that in conda, there is a new pyTorch environment (you can call it whatever you want, whatever you want), and the star on the right indicates which environment you are currently in.

Next, we will install PyTorch in the PyTorch environment, using the following command to activate the pyTorch environment (whatever name you choose to activate). Conda activate PyTorch You can see that the base on the left becomes PyTorch, indicating that you have successfully entered the PyTorch environment.

2. Install PyTorch & Torchvision

Go to the PyTorch website and choose according to your own needs. As shown in the figure:

After the installation is complete, check whether the installation is successful. Open the Jupyter of Anaconda, create a Python file, and run demo:

So far, the PyTorch environment has been successfully installed

PyCharm installation

Go to the official website, download the corresponding version, and click Install.

reference

1, Five deep learning framework analysis 2, WIN10 PyTorch environment configuration (installation of half a day of blood and tears history) 3, PyTorch environment construction