Thread by @HunainShiwani

Interested in learning more about #DataScience, #MachineLearning or #AI? I’ve got a few places and resources for medics to start with. Anyone can do it with enough time and effort! Soon enough you’ll be making your own neural networks

1/16. A thread 🧵.

2/16. Everyone has their preferences with programming languages. However if you’re starting from scratch, I highly recommend #Python. It is easy to learn, has a wide variety of applications and you will find it is much easier to perform even the most basic of statistics.

3/16. It also gives you access to multiple libraries that are used heavily by the machine learning community such as #Keras, #TensorFlow and #PyTorch.

4/16. Learning #Python lets you do just about anything. A lot of the work in machine learning isn’t just the actual design of the neural network and tuning parameters. It is sorting out data and files in a way that makes it easy to process (a process called “data-wrangling”)

5/16. From writing a short script to rename all 5000+ patient files in a folder…

6/16. …to quickly plotting out the difference in demographics between 2 groups of patients

7/16. A good place to start is #CodeAcademy. It’ll help you get started with some of the basics of #Python

codecademy.com/catalog/langua…

8/16. #freeCodeCamp is also another excellent resource to get started with Python

freecodecamp.org

9/16. #freeCodeCamp also has a machine learning pathway which gives you the basics of a neural network and principles

10/16. To advance your skills, try your hand at working on statistical libraries such as #Pandas, #NumPy and #SciPy. Practice doing t-tests or chi-squared tests on data you already have. Understanding these frameworks are essential for machine learning.

realpython.com/python-statist…

11/16. Learn how to visualise your data. Excellent graphical skills allow you to really communicate what your data *MEANS*. Learning #matplotlib and #Seaborn are great place to start as there are lots of snippets of code available throughout the internet for you to copy.

12/16. Now let's get into the meat of what you’re wanting to learn: how to design a neural network. It is really important to understand how a neural network functions before trying to create one and apply it to your data. YouTube is your friend here.

13/16. I found @Sentdex absolutely fantastic when learning how to design neural networks from scratch. If you can watch every single one of his videos and implement them, you’re halfway to becoming a real machine learning expert.

14/16. @Sentdex explains common problems you encounter as you’re building your neural networks. When self-teaching, this kind of experience is invaluable. If you’re reading this; I partially owe my achievements to your videos!

youtube.com/playlist?list=…

15/16. When you build up the confidence, attempt some #kaggle competitions. The real gems are in the solutions of previous competition winners. Reading how they attempted to solve a problem can help you identify clever solutions to your own problems

kaggle.com

16/16. I hope you enjoyed the (very) short primer! I've not even touched upon the subtopics within #ML such as types of CNN architectures but I hope this thread kickstarts your own learning!

Great way for trainees to dip their feet into AI! @TraineesBSIR @UK_NIRS @BSNRTrainees @JuniorBSGAR @IRadResearch @_the_SRT @BSCImaging @VascResearchNet @TheBJCA @ASiTofficial

@MIC_Foundation @surgicalMIC @YSRC_UK @YoungNIR @RiiSEdinburgh @ETF_IRtrainees @STARSurgUK @RouleauxClub @BSGAR_UK @ESHNRSociety @SSRbone @bncs_imaging @BSCMR @BSTImaging

This sounds a bit silly but it really is so easy to program a neural network by copying and pasting a few lines of code. Understand what a neuron is. Understand what layers are.