- Feature:譬如一隻大象有四隻腳、一個鼻子、體重。
- Labels:這是一隻大象嗎? 是或不是?
- Naive Bayes
- Supervised Classificatio\n
- clf = Classifier
- 從sklearn中的metrics
- 找accuracy_score,算精確直。
- Bayes Rule
- Prior Probability *Test evidence =Posterior P
- P(C)=1%
- P(Pos|C)=90%
- P(非C)=0.99
- P(Pos
- P(C|Pos) =P(C)*P(Pos|C)
- 陽姓的情況下有癌症的機率是=有癌症的被測出來是陽性的機率*全部人口中有癌症的機率
- 當Learning from document 時,可用Naive Bayes
- 反正就是靠機率去推測的
- SVM( Support Vector Machine)
- Maximize Distance to nearest point = margin
- outliers :會忽略掉
- kernel trick很重要
- low D(維度)->Map to High D->在轉換回來
- 這個方法叫做kernels,回到原本的空間。
- Not separable ->separable
- Non linear separation
- SVC =SV classifier
- C =control trade off between Smooth decision boundary and classifying training points correctly
- C越大,decision boundary 越是扭曲
- gamma =defines how far the influence of a single training example reaches
- low values =far(越平滑) , high values = close (越扭曲)
- 適合有明顯邊界的時候使用
- 數據太大的時候不好,因為是cube
- 噪音太多不太好,此時Naive Bayes比較好
- Decision Tree
- Non-linear ->linear (simple)
- 一直分割一直分割
- Entropy
- Measure of impurity a bunch of examples
沒有留言:
張貼留言