Bakers Hughes

Selections

• Kmeans, centroid based
• BIRCH, works better on large sets vs kmeans

Evaluation

plots orginal data set finds the max and minimum distances and the standerad variation of the diffrences between eveypoint in the orginal data set and its clustered paired.

Process

V1

Started with multiple models several clustering models such as kmeans, spectural, BIRCH, OPTICS, DBSCAN, FeatureAgglomeration, kmeans_plusplus, and affinity_propagation. As well as clustering using KDE's.

V2

due to long load times and trouble evaluting centers point and cluster indices the pool of vibal methods was reduced too. kmeans, spectural, OPTICS, BIRCH, DBSCAN, and KDE + kmeans.

V3

Due too lack of time the pool was reduced kmeans and BIRCH as well as a layered clustering method mixing the two models.

Finals Notes

• Birch accurracy improves with lowered threshhold hovever it also raise compute time.
• Nothing to found on kmeans random seeds.
• Layered model clustering ratio could see improvements.
• 3d Hist's ugly idk _(-v-)_/

Built With

• numpy
• pandas
• python
• sklearn