This is a notebook tutorial for TensorFlow (mainly thorugh Keras) on MNIST data
You will go through building a simple fully connected (dense - DNN) network, then improve it using convolution (CNN), and then you will explore RNN (LSTM) for the same problem
Windows users can use this bootcamp at: https://github.com/awslabs/aws-ai-bootcamp-labs
Note that there are a few new AMI, choose the one with Conda:
"Deep Learning AMI (Amazon Linux) Version 1.0 - ami-77eb3a0f
Deep Learning AMI with Conda-based virtual environments for Apache MXNet, TensorFlow, Caffe2, PyTorch, Theano, CNTK and Keras"
Make sure that you have the keypair you are using or download the new one that you created
Connecting to the instance and opening an SSH tunnel for Jupyter on port 8888 (Ubuntu or Amazon Linux):
ssh -i user.pem -L localhost:8888:localhost:8888 ubuntu@ec2-ip-ip-ip-ip.region.compute.amazonaws.com
ssh -i user.pem -L localhost:8888:localhost:8888 ec2-user@ec2-ip-ip-ip-ip.region.compute.amazonaws.com
git clone https://github.com/guyernest/TensorFlowTutorials.git
jupyter notebook
In the jupyter terminal start TensorBoard and point it to the log directory used in the notebook
tensorboard --logdir=~/TensorFlowTutorials/logs/
Opening SSH tunnel for TensorBoard default port 6006 (Ubuntu or Amazon Linux):
ssh -i user.pem -L localhost:6006:localhost:6006 ubuntu@ec2-ip-ip-ip-ip.region.compute.amazonaws.com
ssh -i user.pem -L localhost:6006:localhost:6006 ec2-user@ec2-ip-ip-ip-ip.region.compute.amazonaws.com
Append the port number after the /proxy/ URL, for example:
https://.notebook..sagemaker.aws/proxy/6006/