Open basickarl opened 6 years ago
Yes, I agree -- need to flesh out both readme and the samples.
I'll answer in brief the install one for anyone who looks at the issue in the meantime. If all you want to use is node with an already built graph (support for already trained models is forthcoming), then just npm install tensorflow
works, since it pulls down the TensorFlow libraries as well.
However, if you want to actually build a graph in python (or train a model), then you'll want to use the Python framework and APIs and for that, you'll need to do pip install tensorflow==1.4.1
.
Note the updated readme with installation instructions.
I'll leave this open until I have an end-to-end sample demonstrating training a model in python, and using it in node.js.
@nikhilk If you do update the documentation I'll go straight over it and some cases make a pull request. I like dumbing down these kinds of things and making things crystal clear.
Next update to the readme will come with next few features being tackled/added.
In the meantime, if something is unclear in how its currently written, please share any thoughts on improvements here in this issue (or open a new issue, if appropriate).
@nikhilk The other bug was filed under due to me following the new instructions. You explain everything needed to know as a Javascript developer on Linux. I like that you also specifically state the .so files required for it to work.
I'm guessing that TENSORFLOW_LIB_TYPE and TENSORFLOW_LIB_VERSION are specific to this library only? I can't find them on Tensoflows website.
Would be nice if you added three sections into the installation section, one specifically for each OS (Linux, Mac, Windows) (if anything is needed specifically), for example, you state:
export TENSORFLOW_LIB_TYPE=gpu
export TENSORFLOW_LIB_VERSION=1.5.0
export TENSORFLOW_LIB_PATH=path-to-custom-binaries
This not Windows specific, so having this under Linux and Mac would make sense. If you understand what I mean.^^
One thing a lot of libraries in machine learning miss out are proper full-scale examples. Nodejs is a part of the web community and the web community I can say aren't the best of programmers. So simple impossible-to-not-understand examples would make this library go very far.
Perhaps you meant this however as part of the better documentation though.
I also have a request for how to install this from scratch. I'm guessing one must install the TensorFlow C API, then install this? Or will this node module install that also (which would be awesome).