├── attach.sh
├── getip.sh
├── run.sh
├── Dockerfile
└── README.md


/attach.sh:
--------------------------------------------------------------------------------
1 | docker exec -it tensorflow_2.11.1 bash
2 | 


--------------------------------------------------------------------------------
/getip.sh:
--------------------------------------------------------------------------------
1 | docker container inspect tensorflow_2.11.1 | grep IPAddress
2 | 
3 | 


--------------------------------------------------------------------------------
/run.sh:
--------------------------------------------------------------------------------
1 | docker run -u $(id -u):$(id -g) -di --rm -p 8888:8888 -v /opt:/opt --name tensorflow_2.11.1 tensorflow_2.11.1
2 | 
3 | 


--------------------------------------------------------------------------------
/Dockerfile:
--------------------------------------------------------------------------------
 1 | FROM tensorflow/tensorflow:2.11.1
 2 | 
 3 | RUN pip install numpy pandas matplotlib scikit-learn seaborn jupyter notebook pyright python-language-server python-lsp-server && mkdir /.local && chmod 777 /.local && mkdir /.cache && chmod 777 /.cache
 4 | 
 5 | EXPOSE 8888
 6 | 
 7 | VOLUME /opt:/opt
 8 | 
 9 | CMD ["/bin/bash"]
10 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # Docker container for Tensorflow 2.11.1
 2 | 
 3 | This Dockerfile creates a docker image with Tensorflow version 2.11.1 that is compatible with at least ST Microelectronics X-CUBE-AI version 10.
 4 | 
 5 | If a model is created with a higher version of Tensorflow, X-Cube-AI, as of now, might fail to convert the model to be used on the microprocessor.
 6 | 
 7 | The created docker image also has some standard ML libraries, such as Numpy, Pandas and SciKit learn and the Jupyter Notebook server so development can be done using the image.
 8 | 
 9 | # How to use it?
10 | 
11 | Create the image by running **./build.sh**. If some other libraries are required, modify the pip install command on the Dockerfile as required.
12 | 
13 | Create a running container by running the **./run.sh** script.
14 | 
15 | Modify this script to map at least the host volume where your files (Jupyter notebooks reside) to the container **/opt** directory.
16 | At the moment, the /opt host volume is mapped to the internal /opt volume.
17 | 
18 | Check the running container IP by running the **./getip.sh** script.
19 | 
20 | Attach to the container by running the **./attach.sh** script.
21 | 
22 | At the prompt docker container bash prompt, run the following command: **jupyter notebook --ip 0.0.0.0**
23 | 
24 | Using the previous found IP, and the outputed token/url of the starting log of the jupyter notebook, use a browser to access it.
25 | 
26 | For example: **http://172.17.0.12:8888/tree?token=a3fb1692994fc26a29466691a18159879604ced685196c6b** where the IP address 172.17.0.12 is the containers IP address.
27 | 
28 | # How to use GPU or other version
29 | 
30 | From available Tensorflow docker images, choose the one with the required version and having the GPU libraries enabled. Check out https://hub.docker.com/r/tensorflow/tensorflow
31 | 


--------------------------------------------------------------------------------