A Website in a Container Running CNN

These files can be deploied into a docker container and will establish a website with flask structure, on which users can upload their image about a digit, then the website will use CNN algroithm to indentify the number(not exactly) and report to users. Then the log will be stored into the database of cassandra.

catalogue

1,Train the CNN model from mnist
2,Build the docker image
3,Run the cassandra to get access to database
4,Deploy the image into a container
5,Open the browser and go to http:/localhost:4000
6,upload the image and get response of the result
7,Go to cassadra to check the upload log

configuration required:

1,Docker to build image and run container.
2,Cassandra to provide database to store log.

CNN model from mnist dataset

The website uses CNN algroithm to recongize images, so there must be a trained model. There is a trained model in ./form, and if you want to train your own model, execute

• FILE: ./model.py
  python model.py
  

to create your own model. In saver.save(sess,"./form/model.ckpt") you can change your path to store your model and go to for i in range(5000): to change times of training.

If you just want to recongnize your local image, just drag your file into the current catalogue, then execute

• FILE: ./show.py xxx.png/jpg/jpeg
  python show.py xxx.png/jpg/jpeg
  

to get the result. You can use image of any size and color, but only jpg,png,jpeg form are accepted. If you have created your own model, go to saver=tf.train.import_meta_graph('./form/model.ckpt.meta') and saver.restore(sess, tf.train.latest_checkpoint('./form')) and change the path into yours.

build docker image from these files

We will build a docker image from these files, preparing to deploy it into a container. Make sure that you have installed docker. then execute
docker build -t name .
in current catalogue. The -t name means adding a tag to the docker image, so you can type your tag on the position of name. Wait for a few minutes, the docker iamge will be created. Use
docker image ls
to check all the docker image that you have created.

Toubleshooting

• The dot.in the first commad is NOT a type mistake, this dot means you create your docker image from the current path. If you miss the dot, the building will fail.
  
• In the Dokcerfile, the docker will install modules that the image need from pypi.python.org. Becasue I am in USA, I can install them successfully, but if you are in China, you minght be affected by the GFW and cannot install them from this source. Then you should change the pypi.python.org into some mirror in China, such as pypi.tuna.tsinghua.edu.cn/simple.
  

run cassandra and connect to database

We use cassandra to provide a database to store the log data. Make sure that you have installed cassandra. Then execute
docker run --name yourname-cassandra -p 9042:9042 -d cassandra:latest
to start the cassandra service. You can use your ownrun name to replace yourname. Then execute
docker run -it --link yourname-cassandra:cassandra --rm cassandra cqlsh cassandra
to cqlsh, in which you can create the keyspace and table, then you can insert your log data. If you want to test the cqlsh only, run

• FILE: ./communicate_with_cassandra.py
  python communicate_with_cassandra.py xxx.png/jpg/jpeg '0'
  

to create a keyspace named spaceforcnn and a table named cnntable then insert a row into the table. xxx.png/jpg/jpeg and '0' are two parameters needed by the python file, the first parameter must be a image file and the second must be a string.

Troubleshoot

• If you cannot communicate with cassandra, execute
  docker inspect -f '{{.NetworkSetting.IPAdress }}' yourname-cassandra
  to get the IP of cassandra container. If it's not 172.17.0.2, go to cluster = Cluster(contact_points=['172.17.0.2'],port=9042) and change the contact_points to correct IP.

deploy your image into container

Now we deploy our image into container and run the image. Execute
docker run --name yourname -p 4000:80 --rm -d yourimagename
to deploy the image into container and run it. The yourname is the position to set your name of container, and yourimagename is the tag of your image. -p 4000:80 means exposing your local port 4000 and connecting to the 80 port of your container.

go to the website and upload your image

• Now open your browser and go to
  http://localhost:4000
  and you can see the greeting page. Then go to
  http://localhost:4000/upload
  and you can see the page to upload your image. Upload your image, then you can see the recongnized number of the image. Go to
  http://localhost:4000/upload/filename
  to check the image that you last upload.

• Or you can execute
  curl -F "file=@pathofyourimage" http://localhost:4000/upload
  to get the same result.

check the upload log

Now go to terminal of sqlsh, execute
use spaceforcnn;
to get into the keyspace. Then execute
select * from cnntable;
to get all the uplaod log data of the website.

Troubleshoot

• Remember to type ; at the end of the command to tell the sqlsh that you have finished your command.