By Andrin Halali
Stats baby!
Researchers may be interested in writing their own custom estimator.
Archive
Boosting Python with Rust
By Raphaël Gomès
A basic mercurial command was benchmarked after being rewritten with Rust which showed exciting improvements.
Raphaël was excited to take this forward. However, initially this was not successful. Problems included data friction when passing data between languages. As well as missing features in Rusty Python. Work is going on to implement features in hg with mercurial with the aim to attaining the potential speed improvements that have been demonstrated.
Extending sudo in Python- Best of both worlds
By Peter Czanik
My first session at FOSDEM is brought to you by Peter Czanik. Videos and live streams are available on the FOSDEM website. Here is the session summary.
Need to restrict or monitor sudo access on your servers? Well sudo will be soon supporting Python based plugins and reporting.
Did you know it already supports recording sessions?
Machine Learning as a Service – Anand Chitipothu
This talk is about creating a simple end interface for running your machine learning code.
Anand is the co-founder of https://rorodata.com/ a Platform-as-a-Service,
designed for data scientists, for running machine learning code.
Machine learning libraries that were mentioned that I had not come across before:
PyTorch – https://pytorch.org/ – a deep learning framework for fast, flexible experimentation.
joblib – https://pypi.org/project/joblib/ – is a set of tools to provide lightweight pipelining in Python.
The Rorodata firefly tool creates a RESTful API for your client defined functions.
The config format is YAML and not unlike defining a Bitbucket pipeline.
The end user only needs to a Python function, define API with firefly, deploy and an endpoint on the paas is created.
You can add further system requirements to the runtimes available in rorodata.
It also supports configuring CORs domains.
You can define the size and scale of platform your code will be run against.
You can also use the power of rorodata on your own servers or cloud infrastructure using https://github.com/rorodata/rorolite NB: size/scale config are not available but otherwise can use the same code.
Assume the worst – Alex Chan
If your system allows a user interaction, say a chat function, ensure you can block people. Gave an example about a payment platform which was abused to harass an ex.
How to build services to reduce being open to abuse online.
# Diversify the team
Interestingly you cannot change your name in the history of git say.
# Real name policies
This does ensure people will be civil, truthful, is helpful for protecting vulnerable groups
# Robust privacy controls
Have the ability to ban/block malicious users (individual/platform)
Consider shadow blocking or muting / so the blocked user cannot se they have been blocked per say. As blocking can trigger retaliation.
This could be changeable in the past.
# Human moderation is best
Think about your moderators – they need access to counselling etc before burning out.