The NHS-R Community started in 2018 to promote the use of R in the UK Health and Care System. It's since grown to be a large voluntary community that is wider than the NHS with people being interested in more than R. The videos on this RU-vid channel include many of the conferences, webinars, workshops and talks by and for the NHS-R Community.
NHS-R Community is hosted by the Strategy Unit, part of the NHS Midlands & Lancashire Commissioning Strategy Unit.
Hopefully the subtitles help but we recorded the podcasts through Zoom and on work laptops often so the sound is not always the best and varied according to equipment. Thanks for commenting!
Hello, thanks for your comment. This workshop was run by Shannon Pileggi who wrote a blog about the workshop www.pipinghotdata.com/talks/2022-11-11-debugging/ and the GitHub repository is github.com/shannonpileggi/debugging-nhsr.
Thanks for your comment, we'll put the resources in the description and they also go on the NHS-R Community website nhsrcommunity.com/events/february-2023-webinar-introduction-to-causal-inference-and-propensity-modelling-christel-swift-bbc/ for reference.
Thank you so much guys. This was by far the most interesting video on forecast I've seen on RU-vid. I will apply these models on hospitalization dataset and check which one yield better results. All the best to you!
Really nice tutorial of the whole workflow! I would make some considerations: 1) Maybe train-test split could be done at very first, before EDA and surely before imputing missing values 2) That barplot about significance in logistic regression is very very ugly and eventually misleading for various reasons. I would recommend you to not use it anymore, especially in tutorials. It's more useful to plot regression coefficients (or odds ratios) to immediately see positive and negative effects with a central breakpoint and eventually sort bars by significance, eventually putting in evidence those predictors under the established level of significance. A coord_flip() would be also appreciated, to have horizontal labels :)
Indeed - I would say that is a good approach. 2) agree - it is just for demo purposes, but points noted. Have a look at this package I developed for plotting odds ratios: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-HO0Mm6_LCGE.html.
@@giuseppedejanlucido - great to field your comments and being connected. Thanks for your help in the guidance around data vizz (my least favourite part of DS). Cheers
Hi!..First of all, thank you very much for this amazing tutorial!!...I do not see the "Git" tab in Rstudio...How do I fix that?...Thanks again and good luck!
Thanks for making this tutorial available to all - I found it really interesting. At 57:05 someone mentions that their graph of p-values from the logistic regression looks different to the one being displayed in the tutorial. Mine also looked very different and I noticed that re-running the whole process, including the SMOTE step, would result in a different graph each time, e.g. the p-value for patient age varied between 0.005 and 0.50 in a small number of replicate runs! The over-sampling approach is clearly not producing reliable results here - but I guess that's a very useful lesson in itself.
Yes, one way to control this behaviour would be to set a random seed set.seed(123) - this would make the results reproducible. However, I don't include this in the code, as it is easy to put this into production without realising. Essentially, this is controlled by rsample as it will take a different sample of data every time you run the experiment. I explain later on that K-Fold cross validation is a more useful estimate, as it averages over a number of splits on the training set.
Tom, Very nice presentation until the end when you started running through loads of code. It was very hard to keep up since you kept changing things (and often not explaining why). A lot of these base functions like .Last etc I had not seen before and found confusing.
For most part very clear. Although it seems to me that functional programming is simply programming with functions. There seems to be a tendency to over-explain simple ideas and concepts in general. Also calling a function "average" and having it run "median" is simply confusing.
miss Rebeca. Thank you so much for sharing your knowledge on Change point analysis. this is really very helpful in my undergraduate studies. I have a small question. Do we need to make the time series data stationary before applying the changepoint method you have discussed here and in the other lecture. can you please help me with this. Thank you
I'm guessing it was recorded on MS Teams so we are a bit constrained by Microsoft and the secure networks we use. But yes, we could do better with quality and also with subtitles which we haven't really done anything with in previous recordings.
Thank u for this Is there a way to automatically split tables based on size of boxes in a presentation? I am using officer package with customised layout
