Good news everyone! A new version of graph-tool is just out!
@graph_tool
@graph_tool
is a comprehensive and efficient Python library to work with networks, including structural, dynamical, and statistical algorithms, as well as visualization. 1/N
Good news everyone! A new version of graph-tool is just out!
@graphtool
Single line installation:
Anaconda ⤵️
conda create --name gt -c conda-forge graph-tool
Homebrew ⤵️
brew install graph-tool
Debian/Ubuntu⤵️
apt-get install python3-graph-tool
1/2
Taking a closer look at this paper, I really dislike it.
It's clear that newer papers *must* have a lower “disruption” score than older ones under a null model — they even confirm this in the supplemental material with a randomization test.
I'm happy to announce a new project, called Netzschleuder:
This is a network data catalogue and repository. It currently contains 257 datasets totaling 3,916 individual networks (94,956 if you include a big
@openstreetmap
trove).
1/5
A new year, a new version of graph-tool is out!!
@graphtool
Single line installation:
anaconda ⤵️
conda create --name gt -c conda-forge graph-tool
homebrew ⤵️
brew install graph-tool
debian/ubuntu⤵️
apt-get install python3-graph-tool
New paper!
“Scalable network reconstruction in subquadratic time”
TL;DR: It's now possible to reconstruct huge networks from observational data using statistical inference.
Explainer thread: 1/N
Just in time for
#netsci2020
, a new version of graph-tool has been released!
Single line installation:
anaconda ⤵️
conda create --name gt -c conda-forge graph-tool
homebrew ⤵️
brew install graph-tool
debian/ubuntu ⤵️
apt-get install python3-graph-tool
Good news everyone! A new version of graph-tool is just out!
@graphtool
Single line installation:
anaconda ⤵️
conda create --name gt -c conda-forge graph-tool
homebrew ⤵️
brew install graph-tool
debian/ubuntu⤵️
apt-get install python3-graph-tool
A new version of graph-tool is out!
@graphtool
It includes a 5x speedup of the SBM inference multilevel algorithm for principled community detection! Let it rip on your very large networks!
See the HOWTO here:
1/3
Just out on
@PhysRevE
: "Nonparametric weighted stochastic block models"
Want to find statistically significant hierarchical modules in weighted networks? Read the HOWTO and get the code here:
Our perspective with
@PiratePeel
and
@manlius84
just came out on
@NatureComms
: “Statistical inference links data and theory in network science”
We argue how Bayesian inferential methods are essential for a robust network science.
A new version of graph-tool is out!
It includes bug fixes, improvements, and a new merge-split algorithm for SBM inference (see updated documentation )
1/3
New on the
@arxiv
: "Statistical inference of assortative community structures"
With my student Lizhi Zhang, we show how you can use Bayesian inference to uncover statistically significant assortative communities in networks.
1/7
New on the
@arxiv
! "Disentangling homophily, community structure and triadic closure in networks",
Homophily/communities and triadic closure (triangles) are conflated properties in network analysis, and this method tells them apart. An explainer: 1/5
The zoom terms of service have been updated to require consent to use your call data for AI training. No opt-out possible.
The near-future of proprietary internet is bleak.
Switch to the free internet. Use
@jitsinews
,
@bigbluebutton
, etc.
Finally out on
@CambridgeUP
Elements:
“Descriptive vs. Inferential Community Detection in Networks”
Open access, and just in time for
@netsci2023
!
Enjoy! 1/3
New paper
@ScienceAdvances
"A network approach to topic models" w/
@martgerlach
&
@egaltmann
We show how using hierarchical SBMs removes serious limitations of LDA, and connects topic modelling to community detection. Code & more:
New in the
@arxiv
: “Implicit models, latent compression, intrinsic biases, and cheap lunches in community detection”
Joint with
@captainkirk1041
, we “back engineer” arbitrary community detection methods to reveal their hidden generative models! 1/11
Pre-prints are as “real” as any other kind of publication.
Peer-review is important, but it's not what determines the value of a publication.
Real scientists read and cite pre-prints all the time.
@arxiv
is invaluable in Physics & Math, with a very different culture than CS.
@tdietterich
@TaliaRinger
@mmitchell_ai
@ErikWhiting4
@arxiv
arXiv is a cancer that promotes the dissemination of junk "science" in a format that is indistinguishable from real publications. And promotes the hectic "can't keep up" + "anything older than 6 months is irrelevant" CS culture.
>>
New work on the
@arxiv_org
: "Network reconstruction and community detection from dynamics",
I show how coupling Bayesian network reconstruction from functional behavior with community detection enhances both tasks simultaneously. 1/4
Want do to a PhD in Network Science?
We are offering 5 fully funded fellowships to study at
@dnds_ceu
, in Vienna, Austria! 🇦🇹
Application deadline is February 1st, 2023. (Next Wednesday!)
More information:
Want to do a PhD in Network Science, Data Science or a related field? 🎓👩🔬📉
We are offering 5 fully-funded scholarships at
@dnds_ceu
@ceu
in Vienna, Austria 🇦🇹, to start in September 2021.
Application deadline: February 28th, 2021.
Read more at:
Exciting news! We are announcing a new undergraduate program in Quantitative Social Sciences at
@dnds_ceu
.
Our program combines rigorous ("hard sciences"-level) mathematics, statistics, and programming with the pillars of the social sciences. 1/3
New on the
@arxiv
: "Revealing consensus and dissensus between network partitions"
Community detection algorithms often give us multiple competing answers. One common solution is to build a consensus among them.
But what if there is no consensus?
1/7
Want do to a PhD in Network Science?
We are offering 6 fully funded fellowships to study at
@dnds_ceu
, in Vienna, Austria! 🇦🇹
Application deadline is February 28th, 2022.
More information:
Want do to a PhD in Network Science?
We are offering 6 fully funded fellowships to study at
@dnds_ceu
, in Vienna, Austria! 🇦🇹
Application deadline is February 28th, 2022.
More information:
Finally on the arXiv: "Reconstructing networks with unknown and heterogeneous errors"
Did you know you can reconstruct and make error estimates for networks, by making only a single noisy measurement?
Bored in home isolation and craving to infer the modular structure of some network data, but your code is too slow?
New in the arxiv: "Merge-split Markov chain Monte Carlo for community detection" 1/3
Finally, according to their definition, review papers would be “disruptive” because they funnel a bunch of citations. And a paper that does not cite anyone but is universally cited would not be “disruptive”. 🤷
New work on the
@arxiv_org
: "Network reconstruction and community detection from dynamics",
I show how coupling Bayesian network reconstruction from functional behavior with community detection enhances both tasks simultaneously. 1/4
I used
@Sci_Hub
today and was greeted by this GIF of Alexandra Elbakyan (), the site's inventor, who has single-handedly solved the problem of open access in science. It's amazing what can be accomplished with common sense, courage and skill. Massive kudos!
Just out on
@PhysRevX
: "Reconstructing Networks with Unknown and Heterogeneous Errors"
New method can reconstruct networks and provide error estimates for them, even when measurement uncertainties are unknown.
In Germany, the situation for post-docs has been notoriously horrid: If you don't get a permanent position in 6 years—you're out. Your contract cannot be renewed.
The solution the government just announced for this is:
🥁🥁🥁
Reduce the maximum post-doc time to 3 years.
WTF
In den letzten Wochen und Monaten habe ich unzählige Gespräche mit Interessenverbänden, dem Bundesministerium für Bildung und Forschung und innerhalb der Ampel-Koalition geführt. Heute können wir ein Ergebnis präsentieren! Wir reformieren das Wissenschaftszeitvertragsgesetz.
The shrinkage bias caused by L1 and cross-validation do not mix well, forcing an unnecessary trade-off between promoting sparsity and reducing bias.
In the end, your reconstructed network ends up with a bunch of fake edges. 👎
Besides being slow and annoying to use... 6/N
What a sorry state of affairs... Study with 161 social science researchers working on the *exact same data* reach radically different conclusions, almost symmetrically distributed among positive/negative effect.
1/5
We're hiring! Tenure-track Assistant Professor at
@dnds_ceu
@ceu
, in Vienna.
The focus is on social data science, social network science, or quantitative social science — broadly interpreted.
Deadline: February 20, 2024
For questions, get in touch!
Giorgio Parisi – awarded this year’s
#NobelPrize
in Physics – discovered hidden patterns in disordered complex materials. His discoveries are among the most important contributions to the theory of complex systems.
@_jgyou
True visa interview in the US consulate a while back:
- What do you do?
- I'm a Physicist.
- What do you actually research?
- Networks, complex systems, statistical physics.
- Oh, is this related to Markov Chains?
- !!!..... yes?
Just got the new book by
@dorogovtsev
&
@jffmendes
: “The Nature of Complex Networks.”
Looks fantastic!
Humbled to see my name in the “shout outs” and have a few of my papers cited. As someone who admired the authors' towering work as a PhD student way back, this feels surreal.
New on the
@arxiv
: "Descriptive vs. inferential community detection: pitfalls, myths and half-truths"
This is a selected review of the many dangers of using descriptive methods (e.g. modularity) with inferential goals, and existing solutions. 1/2
It's worse than this. It's a tiny side project at google without any public oversight that has become a central piece of academic infrastructure.
We clearly need a free and open version of google scholar, with accountability, an actual API, etc. Imagine how cool that would be.
Dear academia,
before your least favorite billionaire buys off
@SlackHQ
, you might want to take a look at
@element_hq
, a free, decentralized and secure alternative based on
@matrixdotorg
.
There is a free internet out there — we just need to use it!
Every point
@lpachter
makes about the unredeemable failures of UMAP is also valid for community detection in networks using modularity maximization.
The main difference is that with community detection we actually have robust inferential alternatives:
Kind of sucks when you've been telling people publicly that they should abandon UMAPs because they can be misleading and don't provide a faithful representation of the data... and then you look at a UMAP and it's really useful. 1/3
Finally out on
@PhysRevX
: "Disentangling homophily, community structure and triadic closure in networks"
It presents a tractable inference method to separate triadic closure from group-level preference.
Code here:
How about this set if 10⁶ x 10-dimensional points sampled i.i.d. from a Zipf distribution? Does the manual say anything about that?
Keep increasing the number of points, and watch the medusa get more interesting.
Friends don't let friends interpret UMAP plots.
There are lots of interesting things to say about dimred algs but this is pure clickbait.
UMAP + t-SNE are for high-dimensional data with many points. This plot is exactly the same toy ex, but the data is 100 * 1e4.
Read the user guide before you use a tool, it's not hard🤷♂️
New blog post (not April fools!): “Hidden models and latent compression in community detection”
This an overdue post on a joint work with
@captainkirk1041
published last year.
🚨Job Alert🚨
We're hiring an Assistant or Associate Professor in Machine Learning and Data Science to join us at
@dnds_ceu
in Vienna, Austria! 🇦🇹
We're searching for an expert in social data and/or network science.
Deadline: March 31
@graph_tool
's documentation has just gotten a facelift!
Besides an overall re-organization, now there's a new and improved quick start guide.
Check it out and tell your friends!
@deaneckles
Not critical enough by a long shot...
It's amazing how some researchers/field seem to have a fast lane to vanity journals like nature.
They desk reject everything and then somehow let go through sloppy analyses like this.
The Department of Network and Data Science
@dnds_ceu
at the Central European University
@ceu
is announcing *FIVE* fully funded fellowships for its PhD Program in Network Science! 👩🎓
Application deadline is: January 30 2020 📢
Tip of the day: Did you know you can install and use graph-tool using
@GoogleColab
?
This means you can use it without a local install, and collaborate with other people.
See the (simple) instructions here:
From that example you just start using it!
Just out on
@SciReports
: "Change points, memory and
epidemic spreading in temporal networks"
Simultaneous modelling of multiple time scales accurately represents epidemic spreading on temporal networks.
Joint work with
@laetitiagvn
,
@ISI_Fondazione
.
🚨Hot news for those interested in the statistics of networks and/or social systems! 👩🔬📉
We are hiring an Assistant or Associate Professor in Applied Statistics at the Department of Network and Data Science
@dnds_ceu
at CEU
@ceu
, in Vienna, Austria 🇦🇹!
This is excellent.
Now
@PhysRevLett
is explicitly embracing machine learning and neural networks as valid topics in physics.
@PhysRevX
and
@PhysRevE
should follow suit.
We need more ML work with a physics flavor.
@APSphysics
could offer a great counterpart to CS venues.
When comparing with the null model they compute only the z-score, getting values at most 4 or so. It's also besides the point — as usual, small meaningless deviations from the null model can be statistically “significant.” Effect size ≠ statistical significance.
It's *awesome* that arxiv requires LaTeX source files, and *everyone* knows they are public, a fact they make abundantly clear in the upload instructions — and it's also obvious.
Having the LaTeX source for *millions* of papers is a major public treasure, let's not lose it.
Yes, it's funny that this happened to a rather pompous paper, but we should be talking about the fact that it's just awful that arXiv requires uploading LaTeX and doesn't make sufficiently clear that it will be public, despite thousands of authors getting tripped up for decades.
@drugmonkeyblog
Data hoarder: "I fail to grasp the most basic scholarly obligations as a scientist, and think that those that profit from my work are leeches, and not 'actual scientists'."
Other scientists:
Friendly reminder that modularity maximization is not a good way of uncovering communities in networks. It overfits and underfits massively — yields no information on statistical significance at all — and is heavily biased in important ways.
#netsci2023
New on the
@arxiv
: "Statistical inference of assortative community structures"
With my student Lizhi Zhang, we show how you can use Bayesian inference to uncover statistically significant assortative communities in networks.
1/7
This is a very interesting work from
@jacomyma
, the creator of
@Gephi
, about network visualization!
He articulates many ideas about network visualization, from a point of view that is very new to me. I have not read it all yet but I'm enjoying it. 1/8
My PhD thesis is titled Situating Visual Network Analysis, and I will defend it June the 1st at 12:00 CET in Copenhagen. It will be accessible to an online audience. I will provide additional details when I have them.
Download:
Pathetic, worthless argument. Everyone started with a single paper.
And credentials *do not matter* in science. This is not a church, or the army.
This only goes to show how well educated and successful scientists can be abysmally asinine and vile.
Senior scientist work hard & write hundreds of papers to earn recognition.
Social media & preprint servers are turning a model successful for hundreds of years up-side-down.
Compare data reported by Google Scholar before dismissing giants like John Ioannidis. Shame on MIT.
Woot! My 2019 paper "Network reconstruction and community detection from dynamics" is trending in
@PhysRevLett
!
Read it (for free!) here:
Here's the explainer thread:
Code documentation is here:
It was always clear that graph neural networks don't actually need the graph for most tasks and don't perform so well when the node features are not doing the heavy lifting.
But it seems that often including the graph makes things *worse*:
Periodic reminder that using modularity maximization for network data analysis is virtually always a bad idea, in any context, and that superior inferential methods have existed for a long time.
🚨Job alert! 🚨
We are looking for an Assistant Professor to join our department
@dnds_ceu
at
@ceu
, in Vienna, Austria!
Spread the word!
Application Deadline: September 15, 2023
#networkscience
#netsci2023
Really nice short lecture on the ideas and history of complexity.
I'm always in awe at
@pholme
's encyclopedic knowledge of the history of science, in particular complex systems and networks. He is second to none!
@pholme
When are you writing a book about this? Can I pre-order?
For those wondering, a SBM fit of a network with 10^6 edges () on a laptop takes around 2 minutes for the simplest model (planted partition) and around 19 minutes for the most complex (nested SBM).
All with statistical regularization... No overfitting!
A new version of graph-tool is out!
@graphtool
It includes a 5x speedup of the SBM inference multilevel algorithm for principled community detection! Let it rip on your very large networks!
See the HOWTO here:
1/3
We're looking for an associate or full professor in Machine Learning to join our department
@dnds_ceu
!
We're looking for researchers invested in social or behavioral data and/or at the interface with network science.
Come and join
@ceu
in
@Stadt_Wien
!
@ylecun
@matthen2
@gabrielpeyre
This is true, but it is not the most convincing explanation.
The reversible trajectory is *possible* but exponentially suppressed due to the overwhelming number of non-reversible alternatives. The simulation chose this particular one, but it would never be observed in practice.
The city of Sāo Paulo, with a population bigger than entire Austria, just reached 100% of adults fully vaccinated.
Yet, here I am in Vienna under a lockdown, with only 65% of the population vaccinated.
Carl Sagan testifying before Congress in 1985 on how the greenhouse effect will change the global climate system.
“I think what is essential for this problem is a global consciousness. A view that transcends our exclusive identifications”
36 years later, did we listen to Carl?
@ramencult
It's not just that it costs money, Matlab is proprietary. You want to understand how it works? You can't. Do you want to improve it and share the improvement with the world? It's illegal.
Using Matlab (or any proprietary software) is anathema to science.
I developed a model-free algorithm that extracts features from data, without any prior assumption about how it is generated.
ⓘ 𝗢𝗳𝗳𝗶𝗰𝗶𝗮𝗹 𝘀𝗼𝘂𝗿𝗰𝗲𝘀 𝘀𝘁𝗮𝘁𝗲𝗱 𝘁𝗵𝗮𝘁 𝗶𝘀 𝗳𝗮𝗹𝘀𝗲 𝗮𝗻𝗱 𝗺𝗶𝘀𝗹𝗲𝗮𝗱𝗶𝗻𝗴