DL - xinshuaiqi/My_books GitHub Wiki

2017 DL + Genomics + Biomedicine

L1 ML101

Loss function:

  • square loss
  • Gradient descent
    • Stochastic GD
    • Batch GD

Logistic regression

  • softmax loss

Regression performance

  • RSE: residual standard error
  • R squared: goodness of fit

Common performance measures

  • Accuracy, ACC = TP + TN / N
  • Error rate = 1- ACC = FP + FN /N
  • Sensitivity/ Recall = TP /TP + FP = TP /All P
  • Specificity = TN /TN + FP
  • Precision = TP / All P
  • FDR = FP / All P
  • ROC =(FPR, TPR)
  • PR

![1538487025735](C:\Users\evrpa\OneDrive - Monsanto\GitHub\My_books\ML\DL\Lectures2017\Lecture1\1538487025735.png)

Overfitting and underfitting solution:

  • Train + Validation + Test

=0.8 + 0.2

=0.64 + 0.16 + 0.2

  • 5 fold Cross validation + test

= 0.16 + 0.16 + 0.16 + 0.16 + 0.16 + 0.2

L2 Genomics101

1538487733056

epigenetics这一块要加强