Predict housing prices in Austin TX with tidymodels and xgboost

[@chrisjackson8526]

These are honestly some of the most helpful/instructional videos that I've ever found on statistical modeling.

[@My-NaMeS_jEfF]

Thank you Julia I’m voting for you tonight because you rock!!!!

[@sarizwan1986]

Nicely done Prof. Julia. I hope you take viewer requests?? :) Can you do a video on ensemble modelling for multiclass classification problems and then derive how important the variables are based on the ensemble. Thank you very much for doing these screencasts consistently and regularly. I can't really stress how important these are to new learners like me.

[@deanmait]

Thank you. Another great screencast. The systematic step through is a treat.

[@collinguidry9867]

This was really informative! I was not expecting latitude and longitude passed straight into the model to work so well.

[@method341]

Hi Julia, love your videos. I was wondering if you could do one on web scraping? In particular stock market data. Thanks

[@jesuisrobert808]

I liked at the end where you had the graph of items that correlated most with price. It reinforced the geo map of prices. West of city houses generally do have bigger lots. It would have been interesting to see the trend of data prices over time. Also, where are the jobs? Type of jobs. That plays a huge role in the latitude - price correlation. For the longest time, tech jobs were north and west. Now we are seeing more jobs in central and Tesla and Oracle just built a presence in South Austin. Central Austin has more to do with activities, IMO.

[@Leonhard-Euler]

Good video. Two points: 1) When generaing top_words, the clause "filter(!word %in% as.character(1:5))" could be improved to "filter(nchar(word) > 1)", this could eliminate all single-letter word 2) It looks like the hasSpa logical column is not convert to numeric before tuning, this cause the tuning step fail on my box. I simply add a clause "mutate(hasSpa = as.numeric(hasSpa))" when generating austin_split, this works around the problem and the tuning step suceeds.

[@mosesotieno1629]

The videos are magnificently great! I love them. Thank you very much

[@deannanuboshi1387]

Great video. Is there a way to obtain the confidence interval of the final prediction?

[@JuliaSilge]

xgboost doesn't generate a confidence interval, but if you are using a model that does, you can get that as shown here: www.tidymodels.org/start/models/ You might be interested in bootstrap prediction intervals: markjrieke.github.io/workboots/

[@EsinaViwn9]

Demonstrated EDA attempts are perfect.

[@AnantaPradhan]

really nice!! Could you upload some prediction models using raster dataset?

[@vitorklein]

Love your videos, you have a great teaching skills

[@alanjiang2930]

Thanks for sharing, Julia!

[@charlesnjorogekamau8681]

This has been very helpful. Where can I access this specific dataset?

[@georgecooke5639]

Cheers Julia, these videos are great. 👍👍

[@tankUpp]

on 19:28, Julia mentioned to adjust the p-value because she trained many models. Any books someone can refer that touches on this topic? I don't understand the reason behind. ty!

[@JuliaSilge]

I think the Wikipedia article could be a good place to start and at least helps you know about the right vocabulary to look up for more info: en.wikipedia.org/wiki/Multiple_comparisons_problem

[@tankUpp]

@@JuliaSilge Huge ty for introducing this new term for me!!! I would have probably find it a few years later :)

[@N1loon]

Amazing! I learn so much from each of your videos, thank you!

[@황성근-t9k]

As always, Thanks for useful codes and info

[@darmaw22]

Julia, Many Thanks!

[@EsinaViwn9]

I have noticed that parentheses are colorized differently (so it is easy to track where is the closing one). How do you configure this? What is the Rstudio theme used?

[@JuliaSilge]

That part is not a theme, but an option in RStudio. You can check that out here: posit.co/blog/rstudio-1-4-preview-rainbow-parentheses/

[@EsinaViwn9]

@@JuliaSilge thanks!

[@afazdadash9562]

Thank you

[@mniyas3612]

Hi! Recently I have encountered a problem in my work. My variables are several maps in raster format, the value of which I extracted at the sampling points, and I have to generate the pollution zoning map from them with the xgb regression model. As long as the data is in Excel format, everything is going well, but I do not know with what function I can generalize the model on raster maps. I could not do that with the predict function.

[@JuliaSilge]

You can `predict()` on lat/long points that were not part of your training set but you'll need to generate a new grid of points for prediction

[@reshmilb2527]

please avoid background colour black. use eyesight friendly colour

[@joekannoo]

You’re the best Julia!

[@harisjaved1379]

Hi Julia thank you. Can you please deal with a more complicated dataset? Very sparse dataset or imbalanced classes please? Thank you

[@violinplayer7201]

Awesome, thanks for sharing!

[@MrJruda]

Just curious, is there a particular reason you decided to use xgboost instead of something else?

[@JuliaSilge]

xgboost tends to give you really good performance on dense tabular data like this, so it's a good option for a "competitive" situation like SLICED

[@MrJruda]

can't figure out why r keeps giving me this message Error in workflow(austin_recipe, xgb_spec) : unused arguments (austin_recipe, xgb_spec) everything worked perfectly till here

[@JuliaSilge]

Looks like you need to update to the latest version of workflows from CRAN

[@journey-in-pixels]

Much appreciated 🙏

[@MrAbhimufc]

Love the content!

[@ravi281381]

Thanks for the wonderful screencast! I have a question. Why do we use price_total in the glm?

[@JuliaSilge]

That's like the "failures" in successes out of failures for a binomial model.

[@alexandregeorgelustosa5969]

Top 👏👏👏

[@Akbar_Ato]

Julia, what an amazing content. Kudos! I am religiously watching each one of them. You know you've got the best Tidy Tuesdays. Meanwhile, I am making house price prediction by using XGBoost regression. I have 5 categorical predictors for one of numerical predictors. When I run, I get this error over and over again 🙁I have its reprex, just in case. [1] "Error in as_indices_impl(): ! Must subset columns with a valid subscript vector. x Subscript has the wrong type quosures. i It must be numeric or character."

[@JuliaSilge]

That's great that you have created a reprex! I recommend that you post on RStudio Community, where folks can see the reprex and help you understand where your problem is: rstd.io/tidymodels-community