Train a set of models

[JJanowiak]

set-up a ML pipeline to train a set of ML models based on the sets of features calculated. Details to be confirmed later.

[Alex-AMC]

[[Descriptors]] have been calculated, all 208 of them. [[Fragments]] are included amonst these as seen in RDkit

[[Models]] to develop - an array, free for all.

Starting

• [[Multivariate Linear Model]]
• [[Random Forest]]
• [[gxboostedTrees]]
• [[Logistical Regression]]

---

[[Neural Network]]

• scikit-learn - have a play.

[[Pre-Processing required]]

• Training : Test Split 85:15
• Scaling would be required for most of them (minus the RF)

[[Optimisation]]

• [[Feature Reduction]]
• [[Cross Validation]]

Carry out work in [[Jupyter]]

• If model works, save as a [[pickle]]

Notes:

• Scikit-learn allows you to write pipelines which allows you to do [[Pre-Processing]] -> [[Model Building ]] -> [[Validation]]
• What would be useful to know can we think of any new hypotheses (?)

[Alex-AMC]

Trained multivariant regression model and random forest regressor with feature selection optimisation and hypermeter optimisation done, the best result was a 93% accuracy with an error of 29K error.