Machine Learning- HW1

HW1-15

• Very simple implementation of PLA algorithm. Really worth to try by anyone interesting to Machine Leaning.
• Though the input x has only 4 features in this problem, we need to add one more feature as bias during learning, meaning that the weight vector should be 5 dimensions too.
• hw1-16, hw1-17 : 調參數實驗

HW1-18

• 如果資料確定是線性可分，使用vanilla PLA，反之使用pocket PLA但會慢很多。
• pocket PLA：weight更新發法同vanilla PLA，但會有一個最佳的res_weight對整個training dataset的 error rate最低，所以每次更新完weight都要檢查現在的是否比之前保存的更好，這時候必須遍歷全部data統計錯誤率，也就是比較慢的原因。
• hw1-19, hw1-20 : 調參數實驗

Ref:

• My Github Implementation of Course Assignment
• NTU Machine Learning Foundations (Coursera)