search

deepklarity / jupyter-text2code

A proof-of-concept jupyter extension which converts english queries into relevant python code
MIT License
2.1k stars 360 forks source link

Installation error on Mac OS/X #8

Closed babycamel closed 3 years ago

babycamel commented 4 years ago

Failed to install. Can you clarify the installation instructions please.

dk-crazydiv commented 4 years ago

Hi, Could you please elaborate on the issue you're facing. #7 is one possible scenario. Another issue I faced was caching of TensorflowHub. I am able to run the plugin on Mac using the pip install . installation, followed by uninstalling faiss and installing faiss-cpu.

babycamel commented 4 years ago

Does the code require Tensorflow to be installed? I don't have that installed. What version of Python is required?

Thanks

Rodney Beard

On Mon, Sep 14, 2020 at 2:26 PM dk-crazydiv notifications@github.com wrote:

Hi, Could you please elaborate on the issue you're facing. #7 https://github.com/deepklarity/jupyter-text2code/issues/7 is one possible scenario. Another issue I faced was caching of TensorflowHub https://www.tensorflow.org/hub/caching. I am able to run the plugin on Mac using the pip install . installation, followed by uninstalling faiss and installing faiss-cpu.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/deepklarity/jupyter-text2code/issues/8#issuecomment-692294838, or unsubscribe https://github.com/notifications/unsubscribe-auth/AADZ4SUTRAQV7EWPXN23FQDSFZ4AXANCNFSM4RL5LJAQ .

dk-crazydiv commented 4 years ago

Yes. Python3+, tensorflow2 is needed. After cloning this repository locally, you'll need to run pip install . from repo's base directory to get the plugin to install. It will install dependencies (with some warnings, which could be ignored). Then you'll need to uninstall faiss and install faiss-cpu. I'll commit a CPU-only installation process with instructions, with common fail cases in a couple of days.

wjkowalski commented 4 years ago

I'm on Mac OS as well and was able to download and install the package in Terminal, but opening Jupyter Notebooks in VS Code I don't see Text2Code in my list of installed extensions and it doesn't come up when I search for it. faiss-cpu is installed as well.

dk-crazydiv commented 4 years ago

Hi. We've updated the README with setup instructions for cpu-only systems. Could you please try installing with those instructions. It should work on chrome browser with jupyter notebook command executed from terminal. I haven't used jupyter with VSCode, so I am not sure if the plugin is supposted to work there or not.

The extension installed will be available with name mopp. I'll refactor it to jupyter-text2code in next commit.

bytesemantics commented 4 years ago

I installed yesterday before the git updates - heres the issues I've observed:

1) Installed OK - but I hadn't installed jupyter nbextensions package - this should be in the dependencies 2) Installed jupyter nbextensions package. Then manually enabled the mopp extension (using a --user switch) 3) Very slow to start up on first notebook use (log message should be added to indicate that the trained model is being pulled) 4) Getting 404 mopp.js GET calls

This morning - I refreshed codebbase with a git pull - and tried again 1) 404 mopp GET calls (note: not getting mopp.js 404 issues as above)

I am curious - having used Rasa extensively....why you didn't use it. Its a much cleaner approach than the Tensorflow approach followed. I'd be happy to help.

dk-crazydiv commented 4 years ago

@bytesemantics Thank you for the detailed feedback. 1 & 2 - I'll look into it. On my machine, I never faced this. Maybe I have nbextensions installed globally. I'll push a fix soon. 3 - Added with latest commit 4 - We ignored this as a harmless 404 for the PoC, as it happens first time notebook is opened, and then it gets loaded and works fine. Maybe some document.ready kind of syntax is missing for loading frontend extension.

Regarding 404 mopp GET calls, one reason is serverextension isn't loaded. Could you please have a look at jupyter notebook logs and share if you find error there? If extension is enabled, you'll see loading Sentence Encoder message in terminal. Also, to confirm, you installed with pip install .?

For the PoC, we were playing with both generative(huggingface gpt2) and intent+ner based models. For intent models, we experimented with many different sentence-embedders in pytorch/tensorflow, some of them finetuned on Sentence Similarity and paraphrasing tasks. I do have a primitive Rasa pipeline which does average of word embeddings and NER but didn't have enough vocab/training data to build a supervised classifier and wasn't performing as good as the current embeddings.