Medium publication with community-written tutorials and opinion pieces on machine learning, data science, and deep learning.
HN commenters mostly link their own project write-ups hosted there, such as vehicle detection, chatbots, and editing text in images.
- Identify your goals. Are you blogging for extra income? Do you want to connect with a specific kind of reader, and grow a like-minded community? Do you want to be the best resource on the internet for a subject you love? Are you trying to get a new job? Will you write a book about this subject someday? Your goals inform how you'll promote your content.
- Write THE authoritative post on a subject you are deeply passionate about. For me, this is "tiny machine learning," especially computer vision for single-board computers. A few of my posts are keyword-authoritative for "Raspberry Pi TensorFlow."
patio11's salary negotiation post is a great example of an authoritative post. I've been linking my team-mates to this blog post 2x times a year since it was published, during raise/perf seasons. https://www.kalzumeus.com/2012/01/23/salary-negotiation/
- Publish on a consistent schedule. I struggle with this one, but every successful blog I know posts at least weekly. You'll have more hits the more times you step up to the plate.
- Publish via multiple channels. I run a self-hosted Ghost blog & newsletter, and use Zapier to cross-post to medium.com. Make sure you provide a canonical URL, which tells search engines that your blog is the original source.
- Submit guest posts to a popular publication (again, use a canonical link back to your primary domain)
Example: https://medium.com/towards-data-science/3-ways-to-install-tensorflow-2-on-raspberry-pi-fe1fa2da9104
If you…
It's disappointing to see that the benchmark results are so opaque. I hope we see reproducible results soon, and hopefully from Mistral themselves.
1. We don't know what the evaluation setup is. It's very possible that the ranking would be different with a bit of prompt engineering.
2. We don't know how large each dataset is (or even how the metrics are calculated/aggregated). The metrics are all reported as XY.ZW%, but it's very possible that the .ZW% -- or even Y.ZW% -- is just noise.[1]
3. We don't know how the datasets were mined or filtered. Mistral could have (even accidentally!) filtered out particularly data points that their model struggled with. (E.g., imagine good-meaning engineer testing a document with Mistral OCR first, finding it doesn't work, and deducing that it's probably bad data and removing it.)
[1] https://medium.com/towards-data-science/digit-significance-in-machine-learning-dea05dd6b85b
Very interesting. I've implemented something similar[1] using a pure Collaborative Filtering approach[2][3], that I think works better for me, but it's unable to recommend unpopular repositories.
The New York Times recommender system uses a hybrid approach (Content Based + Collaborative Filtering) called Collaborative Topic Modeling on top of LDA[4]. It would be interesting to try that.
[1]: https://github-recs.appspot.com/
[2]: https://medium.com/towards-data-science/recommending-github-repositories-with-google-bigquery-and-the-implicit-library-e6cce666c77
[3]: https://github.com/jbochi/facts
[4]: https://open.blogs.nytimes.com/2015/08/11/building-the-next-new-york-times-recommendation-engine/
Not really. This is people looking at Wikipedia.
Guess: Who's the Alan with the most pageviews in Wikipedia? Who's the Steve?
I left the answers here:
- https://medium.com/towards-data-science/bigquery-without-a-credit-card-discover-learn-and-share-199e08d4a064
Basically a show-off for BigQuery, but to answer this specific question: The most viewed Alan and Steve in Wikipedia are the ones closer to HN.
SELECT title, SUM(views) views
FROM `fh-bigquery.wikipedia_v3.pageviews_2019`
WHERE DATE(datehour) BETWEEN '2019-01-01' AND '2019-01-10'
AND wiki='en'
AND title LIKE r'Alan\_%'
GROUP BY title
ORDER BY views DESC
LIMIT 10
Good work!
I did this project as well about a week ago. I will shamelessly also share my writings on this project: https://medium.com/towards-data-science/teaching-cars-to-see-vehicle-detection-using-machine-learning-and-computer-vision-54628888079a .
The HOG + SVM method is quite slow and not as accurate as a deep learning approach. Before jumping onto semantic segmentation, I recommend re-implementing this project or more generally solve this problem using a Regional Convolution Neural Network architecture (R-CNN) like Faster R-CNN[1] or YOLO[2] for instance.
[1]: https://arxiv.org/abs/1506.01497
[2]: https://arxiv.org/abs/1506.02640
It's a surprisingly difficult task with quite a bit of research history.
We looked at different solutions extensively (https://medium.com/towards-data-science/editing-text-in-images-with-ai-03dee75d8b9c) and ended up building a tool to solve the problem: https://www.producthunt.com/posts/textify-2
Eventually models will get to the point where they can do this well natively but for now the best we can do is a post-processing step.