# Setup code-server @ PrimeHub Jupyter Session We love VSCode, for its convenient IDE extensions. But how can we use that in our PrimeHub Jupyter session? With our precious GPUs? Here is the guide to host a VSCode server that can be accessed by web browsers!!! # Installation ## A. Install Code server Fist we nee to install [code-server](https://github.com/coder/code-server), an open source tool that host vscode in anywhere~ Install curl for fetching the installation code: ``` sudo apt-get update -y sudo apt-get install curl -y ``` You will have it after this installation step: ``` Bash curl -fsSL https://code-server.dev/install.sh | sh ``` Then you can host the code-server by running ```code-server```, but hold on for a second~ ## B. Get your Extensions from web You might find some extensions not available in Extension marketplace in the code-server IDE... So get your extensions in https://marketplace.visualstudio.com/ : > suggestions: EasyCodeAI, MS-vsliveshare, Tabine 1. Goto their webpages > e.g. https://marketplace.visualstudio.com/items?itemName=EasyCodeAI.chatgpt-gpt4-gpt3-vscode 2. Click 'Download Extension' to download .vsix file 3. Drag and drop the .vsix file to the directories in the Jupyer environment > e.g. EasyCodeAI.chatgpt-gpt4-gpt3-vscode-0.6.7.vsix Install Extensions by: ``` code-server --install-extension EXTENSION_FILE_NAME.vsix ``` e.g. Install GPT Extension ``` code-server --install-extension EasyCodeAI.chatgpt-gpt4-gpt3-vscode-0.6.7.vsix ``` ## C. Install ngrok [NGROK](https://dashboard.ngrok.com) help you forward the service from local to public web. e.g. form ```http://127.0.0.1:8080``` to ```http://my.website.com:80``` 1. Get NGROK zipfile ``` wget https://bin.equinox.io/c/bNyj1mQVY4c/ngrok-v3-stable-linux-amd64.tgz --no-check-certificate ``` 2. Unzip it to the place for execution and remove zip file ``` sudo tar xvzf ngrok-v3-stable-linux-amd64.tgz -C /usr/local/bin rm -f ngrok-v3-stable-linux-amd64.tgz ``` 3. Get your own NGROK token a. open https://dashboard.ngrok.com and register for member b. open https://dashboard.ngrok.com/get-started/your-authtoken c. copy that NGROK_TOKEN # Run Host and Client ## A. Start code-server - You cand starat it in the foreground ``` code-server ``` - or in the background ``` code-server & ``` Then it will return an address with port number. > e.g. ```127.0.0.1:8080``` You can access the hosted session with that address if you host the service but we are not in this case here... ### B. Run ngrok 1. You have to config your ngrok service with the token you got: ``` ngrok config add-authtoken NGROK_TOKEN ``` 2. Now start the ngrok service and for ward your port to the public website ``` ngrok http 8080 ``` ## C. Start the client from browser Go to the site that ngrok provide, and... > e.g. https://c302-180-109-17-41.ngrok-free.app/ Voila!!! Now you can have them both: 1. GPU resource from AIA PrimeHub server 2. Your favorite IDE extensions for coding 3. Spare your resources for local computer Note that your password for login will be at ```~/.config/code-server/config.yaml```. You can show that with ```cat ~/.config/code-server/config.yaml```
Last changed by  
 
ShuYuHuang
70

Read more

Few shot learning for Small Data ML

何謂小資料?為何要小資料AI? 從2005 年Roger Mougalas首次提出Big data這個詞彙之後，相關的題目一直都很夯。Big data是指企業為了自動化做機器學習所需要的大量資料。 其中包含了5個V：資料量(Volume)大、資料種類(Vriety)多、資料生產速度(Velocity)快、資料可信度(Veracity)要高、產生的價值(Value)要能回本。在這個5V增長之下，雖然每筆資料花費變少了，但收集、訓練資料的總花費節節攀升。 因此在近年小公司訓練資料量不足、訓練資料不夠多元、資料標註不足的情況下，小資料的機器學習開始興起。其中有很多key words，包含Small data、few shot learning、meta learning等等，都是解決小資料問題的一些技術方向。 先來看看small data有什麼應用場景。第一是使用少量資料學習更接近人類學習的方式、第二是用在改善稀有資料的建模、第三是節省收集資料、標註的人力成本。 更接近人類學習的方式

2/24/2021
GAN Papers with 1k+ citations

Updated on Feb. 2021 (Vanilla GAN)Goodfellow, I. J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., ... & Bengio, Y. (2014). Generative adversarial networks. arXiv preprint arXiv:1406.2661. Cited by 27497, 1st GAN paper (DCGAN)Radford, A., Metz, L., & Chintala, S. (2015). Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434. Cited by 8447, improve CNN in GAN using striding (Pix2Pix)Isola, P., Zhu, J. Y., Zhou, T., & Efros, A. A. (2017). Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1125-1134). Cited by 7899, Nvidia change noise-to-picture to picture-to-picture (cycleGAN)Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision (pp. 2223-2232). Cited by 7325, most well-known cycle loss with pix2pix discrimination (WGAN)Arjovsky, M., Chintala, S., & Bottou, L. (2017, July). Wasserstein generative adversarial networks. In International conference on machine learning (pp. 214-223). PMLR. Cited by 6202, replace JS divergence with Wasserstein distance (SRGAN)Ledig, C., Theis, L., Huszár, F., Caballero, J., Cunningham, A., Acosta, A., ... & Shi, W. (2017). Photo-realistic single image super-resolution using a generative adversarial network. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 4681-4690). Cited by 5026, the most well-known super-resolution paper (conditional GAN)Mirza, M., & Osindero, S. (2014). Conditional generative adversarial nets. arXiv preprint arXiv:1411.1784. Cited by 4729, 1st conditional GAN (WGAN-GP)Gulrajani, I., Ahmed, F., Arjovsky, M., Dumoulin, V., & Courville, A. (2017). Improved training of wasserstein gans. arXiv preprint arXiv:1704.00028.Cited by 4187, accelerated WGAN

2/23/2021
Read more from ShuYuHuang
Published on HackMD