Surprise
tensorflow
Our great sponsors
Surprise | tensorflow | |
---|---|---|
8 | 221 | |
6,178 | 182,323 | |
- | 0.7% | |
0.0 | 10.0 | |
12 months ago | 4 days ago | |
Python | C++ | |
BSD 3-clause "New" or "Revised" License | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
Surprise
-
Recommender Systems: Surprise library installation on m1 mac
Something is wrong with the repo. The compiler fails with this error clang: error: no such file or directory: 'surprise/similarities.c' If you go to the repo, you'll see the file is indeed missing: https://github.com/NicolasHug/Surprise/tree/master/surprise
-
Recommender systems question
Scikit-surprise is a useful package and has pretty good documentation to help make the leap from conceptual understanding to code. If you want to understand the various implementations, the package is open source and available on GitHub. I can’t speak for optimal computational efficiency but I think that it’s premature to worry about that while you’re still making the transition from concept to functionality.
- Surprise – a simple recommender system library for Python
-
Dislike button would improve Spotify's recommendations
I spent the latter half of 2019 trying to build this as a startup. Ultimately I pivoted (now I do newsletter recommendations instead), but if I hadn't made some mistakes I think it could've gotten more traction. Mostly I should've simplified the idea to make it easier to build. If anyone's interested in working on this, here's what I would do:
(But first some background: The way I saw it, you can split music recommendation into two tasks: (1) picking a song you already know that should be played right now, and (2) picking a new song you've never heard of before. (Music recommendation is unique in this way since in most other domains there isn't much value in re-recommending items). I think #1 is more important, and if you nail that, you can do a so-so job of #2 and still have a good system.)
Make a website that imports your Last.fm history. Organize the history into sessions (say, groups of listen events with a >= 30 minute gap in between). Feed those sessions into a collaborative filtering library like Surprise[1], as a CSV of `, , 1` (1 being a rating--in this case we only have positive ratings). Then make some UI that lets people create and export playlists. e.g. I pick a couple seed songs from my listening history, then the app uses Surprise to suggest more songs. Present a list of 10 songs at a time. Click a song to add it, and have a "skip all" button that gets a new list of songs. Save these interactions as ratings--e.g. if I skip a song, that's a -1 rating for this playlist. For some percentage of the suggestions (20% by default? Make it configurable), use Last.fm's or Spotify's API to pick a new song not in your history, based on the songs in the current playlist. Also sometimes include songs that were added to the playlist previously--if you skip them, they get removed from the playlist. Then you can spend a couple minutes every week refreshing your playlists. Export the playlists to Spotify/Apple Music/whatever.
As you get more users, you can do "regular" collaborative filtering (i.e. with different users) to recommend new songs instead of relying on external APIs. There are probably lots of other things you could do too--e.g. scrape wikipedia to figure out what artists have done collaborations or something. In general I think the right approach is to build a model for artist similarity rather than individual song similarity. At recommendation time, you pick an artist and then suggest their top songs (and sometimes pick an artist already in the user's history, and suggest songs they haven't heard yet--that's even easier).
This is the simplest thing I can think of that would solve my "I love music but I listen to the same old songs everyday because I'm busy and don't want to futz around with curating my music library" problem. You wouldn't have to waste time building a crappy custom music app, and users won't have to use said crappy custom music app (speaking from personal experience...). You wouldn't have to deal with music rights or integrating with Spotify/Apple Music since you're not actually playing any music.
If you want to go further with it, you could get traction first and then launch your own streaming service or something. (Reminds me a bit of Readwise starting with just highlights and then launching their own reader recently). I think it'd be neat to make an indie streaming service--kind of like Bandcamp but with an algorithm to help you find the good stuff. Let users upload and listen to their own MP3s so it can still work with popular music. Of course it'd be nicer for users in the short term if you just made deals with the big record labels, however this would help you not end up in Spotify's position of pivoting to podcasts so you can get out of paying record labels. And then maybe in a few decades all the good music won't be on the big labels anyway :).
Anyway if anyone is remotely interested in building something like this, I'll be your first user. I really need it. Otherwise I'll probably build it myself at some point in the next year or two as a side project.
[1] http://surpriselib.com/
-
Show HN: The Sample – newsletters curated for you with machine learning
I'm planning to build a business on this, so probably won't open-source it--but I'm always looking for interesting things to write about! I write a weekly newsletter called Future of Discovery[1]; I might write up some more implementation details there in a week or two. In the mean time, most of the heavy lifting is done by the Surprise python lib[2]. It's pretty easy to play around with, just give it a csv of , , and then you can start making rating predictions. Also fastText[3] is easy to mess around with too. Most of the code I've written just layers things on top of that, e.g. to handle exploration-vs-exploitation as discussed in another thread here.
Recently I've been factoring out the ML code into a separate recommendation service so it can different kinds of apps (I just barely made this essay recommender system[4] start using it for example).
I'm happy to chat about recommender systems also if you like, email's in my profile.
[1] https://findka.com
[2] http://surpriselib.com/
[3] https://fasttext.cc/
[4] https://essays.findka.com
tensorflow
- TensorFlow-metal on Apple Mac is junk for training
-
🔥🚀 Top 10 Open-Source Must-Have Tools for Crafting Your Own Chatbot 🤖💬
To get up to speed with TensorFlow, check their quickstart Support TensorFlow on GitHub ⭐
- One .gitignore to rule them all
-
10 Github repositories to achieve Python mastery
Explore here.
-
GitHub and Developer Ecosystem Control
Part of the major userbase pull in GitHub revolves around hosting a considerable number of popular projects including Angular, React, Kubernetes, cpython, Ruby, tensorflow, and well even the software that powers this site Forem.
-
Non-determinism in GPT-4 is caused by Sparse MoE
Right but that's not an inherent GPU determinism issue. It's a software issue.
https://github.com/tensorflow/tensorflow/issues/3103#issueco... is correct that it's not necessary, it's a choice.
Your line of reasoning appears to be "GPUs are inherently non-deterministic don't be quick to judge someone's code" which as far as I can tell is dead wrong.
Admittedly there are some cases and instructions that may result in non-determinism but they are inherently necessary. The author should thinking carefully before introducing non-determinism. There are many scenarios where it is irrelevant, but ultimately the issue we are discussing here isn't the GPU's fault.
-
Can someone explain how keras code gets into the Tensorflow package?
and things like y = layers.ELU()(y) work as expected. I wanted to see a list of the available layers so I went to the Tensorflow GitHub repository and to the keras directory. There's a warning in that directory that says:
-
Is it even possible to design a ML model without using Python or MATLAB? Like using C++, C or Java?
Exactly what language do you think TensorFlow is written in? :)
-
How to do deep learning with Caffe?
You can use Tensorflow's deep learning API for this.
-
When the documentation has TODOs
Since you've specifically mentioned ML, here's Tenserflow's GitHub. I'm sure a quick glance through that will change your mind.
What are some alternatives?
LightFM - A Python implementation of LightFM, a hybrid recommendation algorithm.
PaddlePaddle - PArallel Distributed Deep LEarning: Machine Learning Framework from Industrial Practice (『飞桨』核心框架,深度学习&机器学习高性能单机、分布式训练和跨平台部署)
scikit-learn - scikit-learn: machine learning in Python
Prophet - Tool for producing high quality forecasts for time series data that has multiple seasonality with linear or non-linear growth.
python-recsys - A python library for implementing a recommender system
Pandas - Flexible and powerful data analysis / manipulation library for Python, providing labeled data structures similar to R data.frame objects, statistical functions, and much more
Crab - Crab is a flexible, fast recommender engine for Python that integrates classic information filtering recommendation algorithms in the world of scientific Python packages (numpy, scipy, matplotlib).
LightGBM - A fast, distributed, high performance gradient boosting (GBT, GBDT, GBRT, GBM or MART) framework based on decision tree algorithms, used for ranking, classification and many other machine learning tasks.
MLflow - Open source platform for the machine learning lifecycle
Keras - Deep Learning for humans