Machine learning and metagenomics
Adam R. Rivers
Cross-JGI informatics talk
October 1, 2015
The netflix challenge
- 100,480,507 ratings
- 480,189 users
- 17,770 movies
- Most users rate only 200 movies
- 10% improvement = $1,000,000
DNAnerd
MrsDarcy85
ARivers_JGI
Train
Predict
| 5 | 4 | 2 | 5 |
|---|---|---|---|
| 4 | 5 | 1 | 2 |
| 2 | 1 | 5 |
What is machine learning
Useful when:
- a mechanistic model cannot be built
- The number of random variables is large
- Amount of training data is large
Evaluated with formal methods
A group of statistical methods historically associated with CS/AI for:
- Classification
- Prediction
- Clustering
- Dimensionality reduction
Machine learning applications
Supervised
Supervised
Unsupervised
Unsupervised