Practical Guide to Applied Conformal Prediction in R P̶y̶t̶h̶o̶n̶ 😏
Spent some time replicating Chapter 7 in R instead of Python today. Using {tidymodels} & {ggplot2}, all in a handy dandy Quarto.
#rstats
Here's another Julia + R tech stack. Oceananigans.jl fluid dynamics displayed with {rayshader} in R, wave shading? 🌊
All on a single quarto doc 🚀
#TidyTuesday
#BYOD
As someone who studied physics, ex-quant, now leading a DS&A team; I randomly learned about conformal prediction intervals for the first time this year. Applied it & felt like a ML wizard.
So excited to dive into this book and deeper into conformal prediction!
Here's another Julia + R tech stack. Oceananigans.jl fluid dynamics displayed with {rayshader} in R, wave shading? 🌊
All on a single quarto doc 🚀
#TidyTuesday
#BYOD
Given all the requests over the last few days, dropped this repo last night 🎅🏻
Repo includes R code complimenting both Chapters 7 & 8 from Practical Guide to Applied Conformal Prediction ⭐
Topics are Regression, Time Series, & Forecasting 🎄
Enjoy! ☃️
I could talk about RStudio IDE, Quarto, Shiny, Python, & R all day long. . But . .
.. Have I told you about our new R Shiny app built on a Quarto doc that has a backend using {reticulate} which runs R & Python code dynamically together?
🐍🚀📃🥳
And a "few" lines might be an overstatement. In R, using the {probably} library, calibrate your model with ONE line of code. 🔥🤯
probably::int_conformal_quantile(object, train_data, cal_data, level = 0.95, ...)
REPO UPDATE 🎉
Check out the new section in Practical Guide to Applied Conformal Prediction in R. This covers importing
@nixtlainc
's {statsforecast} python library in R for conformal prediction 🚀
Yes, you can use Python libraries in R, thanks to {reticulate} 🐍
All my favorite tools are agnostic or have bindings.
I love the fact I can pick up Quarto 📃, Arrow 🏹, Polars 🐻❄️ & switch between R, Python, & Julia with no issues.
Chaining open source libraries has never been easier.
Ok... How does {tidymodels} low-key drop conformal prediction intervals in the {probably} extension? It's literally exactly what I need to add bounds on a xgboost point forecaster. Check out probably::int_conformal_quantile, you'll probably like it
#rstats
#pydata
#posit
SO CLEAN 🧽
TidierOrg is truly bringing {tidyverse} to Julia. This week, I tested out Julia on the popular
#tidytuesday
R repo.
The fact is, every R programmer just became a Julia programmer. Unlocking a whole new set of packages for R users.
#julialang
#rstats
#tidierjl
The new dashboard tool on
@grid_status
is so dope!
I'm screwing around with a +/- 7 day hourly trend tracker & embedded an AI assistant 🤯, yeah web widgets are supported, even got R working 😂🤫
GREAT SAND DUNES NATIONAL PARK 🏜️🏔️
Really impressed with
#rayvista
this extension package makes
#rayshader
so easy to use, anyone can make a map.
#rstats
Created this yesterday in a few minutes using one function. The great sand dunes are an amazing place 🤗
#colorado
How to use NHITS, NBEATS, & NBEATSx in R 🎉🎉
If you are interested in Deep Learning in R, check out this implementation.
The backend is
@nixtlainc
's neuralforecast. Yes, this is Python and R, the love story.
#rstats
#python
#MachineLearning
{treesnip} is a non-CRAN pkg but jives with the rest of the tidymodels ecosystem.
Although not on CRAN, {catboost} is supported in treesnip! Making it easy to split, resample, cv, and add conformal predictions with {probably}
BOB ROSS COLORS 🎨
I created an algo to scrape blended colors from all 403 Bob Ross painting episodes. Data vizzers can finally feel like painters. 📈🖌️
Comes with two scale functions for ggplot2!
#rstats
#tidytuesday
#ggplot2
#dataviz
Here's my Julia solution for Day One of
#AdventOfCode
Using {TidierData} 🧽 which looks and feels like {dplyr}🔧, making my initial statements, "I refuse to use R for AOC", carry little weight. 😅
Will try a language/syntax I'm less familiar with for the next challenge.
What is your preferred date format? Below, I'll explain why Excel's date format (wtf) should be international standard 👇🏻
🧵 1/44999
#rstats
#pydata
#excel
I've read tons of Wind Power Forecasting whitepapers, benchmarking novelty neural net solutions with shotty results in production.
Within a few minutes one can train a constrained xgboost with conformal calibration and get better results.
... Spooky!
❗New Package Alert❗
{alertr} is a R package that has a single function called alert‼️
This is an audio alert which plays the infamous Metal Gear Solid 'alert' noise‼️
Add it to a for loop, if-statement, ML tuning, or whatever!
devtools::install_github("frankiethull/alertr")
#TidyTuesday
#rstats
#dataviz
#ggplot2
Had fun putting together my first TidyTuesday. Thank you
@Emil_Hvitfeldt
for the dataset! Leveraged ggthemes, ggdark, & wesanderson to come up with a fresh new look. A "FiveThirtyEight & DARJEELING LIMITED met in the dark", theme.
It's always fun bridging open source libraries 🌉
This one involves simulating fluid dynamics in Julia and visualizing the results in R.
Waterlily.jl + rayshader, a perfect combo for unlimited
#dataviz
fun? I think so.. 🌊☀️
2023 was full of suspense, LLM races, new TV series, OpenAI drama.
But tbh my most watched topic of the year was lightGBM, XGBoost, and CatBoost struggling to pass CRAN checks.
#rstats
#multithreading
#trees
Catching up on
#TidyTuesday
's! Went a different route on {babynames} via
@hadleywickham
.
Ever wonder why 'Jessica' is more likely friends with 'Sarah'? Why is mom hanging out with 'Donna' all the time? They may follow similar popularity windows 🤫😉
#rstats
#dataviz
#ggplot2
Had wayyy too much fun this morning with: 1) movie scripts, 2) sentiment, & 3) stream plots.
Here, I explored the word-emotion streams of A NEW HOPE then labeled a few 'anomaly scenes'. Those scenes also stood out to me as a kid!
#dataviz
#rstats
#StarWars
#ggplot2
#tidyverse
#tidytuesday
#rstats
#ggplot2
#PrettyCols
Explored the Bigfoot spottings today. Obviously pop culture plays a role. But I have a suspicion about geospatial applications and the disappearance of Big Foot.
#TidyTuesday
#rstats
#ggplot2
So many characters & traits this week!
A few filter layers on notable characters & trait counts; removing highly correlated traits; w/ a finale of parsing down to what I'd call 'heroes' and 'prevalent' traits via PCA;
Here's a guide to heroism:
Hold it together, Texas. I want my job to be easy next week. 🤞🏻
Would also be interesting to see N1 contingencies, i.e. a DA committed CC trips offline in RT, where does the reserve margin go to and will that trigger an ORDC event.
With cold hazards approaching next week, we built a dashboard to track the forecasted impact on load & renewable generation in ERCOT
With 80 GW+ forecasted loads, a record winter demand could be on the way
Follow the live updating current forecast here:
Practical Guide to Applied Conformal Prediction in R P̶y̶t̶h̶o̶n̶ 😏
Spent some time replicating Chapter 7 in R instead of Python today. Using {tidymodels} & {ggplot2}, all in a handy dandy Quarto.
#rstats
@predict_addict
@mdancho84
@topepos
@VincentAB
Hopefully useful to compare to the ipynb!
I highlight {tidymodels} and {probably} for Chapter 7. For Chapter 8, provide a quick ts solution in {fable} (shout-out to
@robjhyndman
) then switch over to {modeltime} workflow.
I need to look into {marginaleffects} a bit more!
Sierra Blanca Massif in CO 🏔️
Created in R using {rayshader}
Absolutely love these mountains 🖤
Just south of the Great Sand Dunes, Blanca Peak watches over the San Luis Valley with it's massif friends
I've been using
@maxk
's
@grid_status
for a minute. The open source GitHub project, the website, and API. Whether it be monitoring or forecasting.
MISO is trending toward their monthly peak today. Yet what sticks out is bad meter data. This is just the surface of ISO data issues
@WallSt_SunTzu
I'm already running tidymodels workflows in production.
If someone is already doing tidy modeling in R & wants conformal prediction intervals on their points preds, it's just one extra line of code.
Tidymodels is a prerequisite for these. Info here:
Sometimes presentation slides just lack something special.
Then this week I switched from PowerPoint to Revealjs --holy sh💩t --, like you can just put an interactive plotly object in a slide deck. 🤯🥳
All on a single Quarto doc! 📃🚀
👀 Based on the cadence of Mojo releases and the lack of a new version this past week, I’m predicting the next version of
#MojoLang
will be open-sourced, hopefully on an MIT License.
Would like to announce the release of an experimental pkg!
This pkg has one goal. Web scrape daily gas pricing PNGs & convert the image pixels into tabular time series data.
#rstats
#datascience
@aachmano
🎉 We are thrilled to announce the release of the latest version of mlforecast a
#Python
library for Scalable
#machinelearning
🤖 for
#timeseries
#forecasting
🚀 This version comes with exciting new features that are sure to make forecasting even more efficient and accurate
🧵
Just a reminder to new followers, I created a pkg that plays the MGS infamous `alert!` sound.
Three ways to use:
1) add it to a model tuning routine
2) modify your RING motion sensor sound
3) place on your coworkers' task scheduler when they step away
You would not use spoon 🥄 as a fork 🍴 or a knife. Why would data scientists try to use forecasting to solve anomaly detection problems. This is so wrong.
Anomaly detection is an own domain that has its own methods.
One of the pitfalls of
#datascientists
is trying to treat
Let me know when and where 🚀
I came across STUMPY last week thanks to
@predict_addict
. Was just about to {reticulate} it for testing in R. Was checking if someone already had but found zilch.
Ok... How does {tidymodels} low-key drop conformal prediction intervals in the {probably} extension? It's literally exactly what I need to add bounds on a xgboost point forecaster. Check out probably::int_conformal_quantile, you'll probably like it
#rstats
#pydata
#posit
Two big updates:
1. Tidier.jl comes with a new README introducing the package.
Also includes an FAQ:
- Why should Julia users look at Tidier.jl?
- Is Tidier.jl intended for R users?
- Which functions are re-exported?
- How to install/update it?
Project LaMb, my little local AI assistant, small enough for a Raspberry PI. 🐑
Localized LLMs are getting easier to build by the day.
It's unreal how easy it is to create an AI assistant that speaks and listens!
Pairing {whisper} 👂, {gpt4all} 🤖, and {pyttsx3} 🗣️
Did you know you can now download any data you see on Grid Status?
It's as easy as finding your favorite graph and clicking the download button.
Try it yourself:
#rstats
#dataviz
#maps
Exploring 1.5GB of military missions in SE Asia. Filtering down to 'Laos' & 'Airstrike'. Can see here the USA bombed Laos at a rate of nearly one bomb per second for 9 years. . . Thanks for sending me down the rabbit hole,
@WilliamTchuggs
.
#defundtheCIA