Progress_Summary.txt

# June 12
Finished the age detection task. Below is the summary.

1. Performance of best models
      A. All groups: 94.78% (using 95 features)
      B. Only cancer group: 92.68% (using 100 features)
      C. Only no-cancer group: 97.87% (using 30 features)

2. ToDo:
      A. Compare the list of selected 30 features (Age Detection on No-Cancer group) with the selected features for Cancer Detection on Young group. (Jianhai)
      B. Identify optimal the threshold of each of 30 CGs (Age Deteciton on Young Group) such that the cancer prediciton on young people can reach to the highest accuracy. (Jianhai)
      C. Build a model to classify 4 labels, which are Young Cancer, Young No-Cancer, Old Cancer and Old No-Cancer. (Ying)

# April 20
HSIC Lasso: feature selection - 111 influential features out of ~400k features
Bayesian Optimization: tune SVM
SVM: binary classifier

Best Model: 97.01% test accuracy. Test set = 40% of the original dataset

# April 15.
1. Siamese Network + One-shot learning
	A. use the original gene vector: about 400K in size
	B. map it into a matrix

2. Feature Selection: Hilbert Schmidt Independence Criterion Lasso (HSIC Lasso) 
	With selected features:
	A. KNN
	B. Random Forester
	C. SVM

3. Deep Neural Pursuit: ask for code