如何在 Docker 中运行的 Jupyter notebook 中完成代码

Question

I followed the instructions on the official Tensorflow Docker hub to be able to use a GPU-ready docker image with an exposed port for local Jupyter notebook development; including installation of the required nvidia-docker

I can import packages no problem, but code completion doesn't work when hitting tab. Eg:

import tensorflow as tf       # works fine
tf. <tab>                     # nothing happens

I can use the code if I type I manually. I can execute the following:

import numpy as np            # no complaints
np.arange(0, 10)              # returns expected result

Interestingly, if I type tf. and hit shift-<double-tab> I get the usually full docstring:

---

I used the following command to pull the docker image and run it:

docker run -u $(id -u):$(id -g) -it --runtime=nvidia --rm \
    -v $(realpath ~/Documents/jupyter_notebooks):/tf/notebooks \
    -p 8888:8888 tensorflow/tensorflow:nightly-gpu-py3-jupyter

On top of binding a local folder to the container, this maps my user ID to the running container, so I am not working from root 's home folder (my docker installation is made to belong to root . I can open the notebook directory in the browser and start a new Python3 notebook, as usual.

I am running Ubuntu 16.04, opening Jupyter notebooks in Chrome (I use vimium extension, but it is switched off and I restarted the container after figuring it might be messing with keyboard commands sent to Chrome).

$  docker --version
Docker version 18.09.1, build 4c52b90

Answer 1

Wenmin Wu's tabnine plugin is definitely a better auto-completion service than the built-in.

I want to point out that IPython uses Jedi to do auto-completion and some other magics. However, the compatibility between the duo is problematic every now and then. Besides waiting for an IPython update, an easy fix is this magic:

%config IPCompleter.use_jedi=False

Answer 2

I came across a quick and easy solution here - just run one of the Jupyter magic commands:

%config IPCompleter.greedy=True

Also worked in a Jupyter notebook running within PyCharm. As soon as I ran that command, PyCharm began indexing the project, meaning code completion worked in scripts as well as in the notebook (which also hadn't been working!)

It should be possible to add the config parameter to a global Jupyter config file.

---

There is a second possible solution on the same thread, using a notebook extension, but I haven't tested it.

Answer 3

I wrote a plugin for jupyter notebook which provide code autocompletion based on deep learning model. It's C/S model, and can be run everywhere. You can try this: https://github.com/wenmin-wu/jupyter-tabnine This tool is now available on pypi . Simply issue following four lines command and enjoy it:)

pip3 install jupyter-tabnine
jupyter nbextension install --py jupyter_tabnine
jupyter nbextension enable --py jupyter_tabnine
jupyter serverextension enable --py jupyter_tabnine