# Machine learning cơ bản --- ![](https://hackmd.io/_uploads/B12suFR-6.png) ### Đạo hàm tiến Tính de/db. ![](https://hackmd.io/_uploads/Hkd1AY0ZT.png =x300) ### Đạo hàm lùi (back propagation) https://nttuan8.com/bai-4-backpropagation/ Tính de/db,de/da. ![](https://hackmd.io/_uploads/SkYMAKAZp.png =x250) Trường hợp này ta sẽ thấy tính de/da tiện hơn. VD: ![](https://hackmd.io/_uploads/SJ0COK0ba.png =x400) Đạo hàm cục bộ (màu đỏ) là hệ số. Đạo hàm từng phần (màu xanh lá). Tổng của tất cả(đạo hàm từng phần tại điểm đến * đạo hàm cục bộ của điểm đến đó) vd tại dy/db=dy/dd * dd/db + dy/de * de/db https://drive.google.com/drive/folders/1KDifaACbAuomUiyBsiZfN09gFoKa5iqo BT logistic regression ![](https://hackmd.io/_uploads/r1qssK0ba.png =x150) ![](https://hackmd.io/_uploads/H18uiYAb6.png =x300) ![](https://hackmd.io/_uploads/Sy3tiKC-a.png) ## Linear regression model ![](https://hackmd.io/_uploads/SJ-0B8yMa.png =x200) ![](https://hackmd.io/_uploads/Hym4YUkz6.png =x300) ![](https://hackmd.io/_uploads/H1Vo08kzT.png) ### Polynomial ![](https://hackmd.io/_uploads/Sk_U1vkGT.png =x30)![](https://hackmd.io/_uploads/HkFPZvkGp.png =x200) ### Gaussian ![](https://hackmd.io/_uploads/H1PiJv1zT.png =x60)![](https://hackmd.io/_uploads/Sy6OWvyfa.png =x200) ### Sigmoid ![](https://hackmd.io/_uploads/r1iT1PJzp.png =x70)![](https://hackmd.io/_uploads/S119WDJG6.png =x200) ![](https://hackmd.io/_uploads/BJNjxDJMT.png =x250) ### Integer Encoding ![](https://hackmd.io/_uploads/SJUvTPyGp.png =x150) ### One-Hot Encoding ![](https://hackmd.io/_uploads/HJN8aDJzp.png =x150) ## Logistic Function ![](https://hackmd.io/_uploads/Byek2dkf6.png) ![](https://hackmd.io/_uploads/B1rx2_kGT.png) ![](https://hackmd.io/_uploads/r1gZh_yMT.png) ### Loss measure ![image.png](https://hackmd.io/_uploads/BkLgOTkQT.png) ![](https://hackmd.io/_uploads/S1LlauJG6.png =x140) ## Tree Định lượng: last output là số. Định tính: last output là loại. ### Entropy ![image](https://hackmd.io/_uploads/H1xVxbsm6.png =x50) ![image](https://hackmd.io/_uploads/SkBql-iQ6.png =x70) ![image](https://hackmd.io/_uploads/ry_E4bs7T.png =x140) ### Gini ![image](https://hackmd.io/_uploads/ry_MV-imp.png =x160) ### Misclassification ![image](https://hackmd.io/_uploads/ByIBP-jma.png =x150) ![image](https://hackmd.io/_uploads/SJ0AN-iQ6.png) ### Evaluating Association Rules ![image](https://hackmd.io/_uploads/BJ5P5sXE6.png =x150) ### Regression tree ![image](https://hackmd.io/_uploads/SJsi9iQVa.png =x200) ![image](https://hackmd.io/_uploads/HygkFboXa.png =x200) -y: giá trị trung bình e là độ lỗi % (có thể sai nếu -y~0) cv: cross-validation ### Multivariate Trees Tại một nút xài hơn nhiều 1 thuộc tính. ## Bayes ### Maximum likelihood ![image](https://hackmd.io/_uploads/B1dUSiH4a.png =x200) ![image](https://hackmd.io/_uploads/BJJaVjS46.png) ![image](https://hackmd.io/_uploads/ByU1SsH4T.png =x200) ## Language model ### Estimating N-Gram Probabilities ![image](https://hackmd.io/_uploads/r12ub3LrT.png =x200) ![image](https://hackmd.io/_uploads/r1AYfhLS6.png =x230) ![image](https://hackmd.io/_uploads/Sy6PihUBp.png =x200) ## AdaBoost https://viblo.asia/p/adaboost-buoc-di-dau-cua-boosting-gAm5yrGwKdb https://drive.google.com/drive/folders/1KDifaACbAuomUiyBsiZfN09gFoKa5iqo https://scikit-learn.org/stable/auto_examples/ensemble/plot_adaboost_twoclass.html ![image](https://hackmd.io/_uploads/HkD4IFvra.png =x220) ![image](https://hackmd.io/_uploads/By_Y8Fwr6.png =x250) ![image](https://hackmd.io/_uploads/rJo0tItrT.png =x250) ## Atificial neural network (ANN) ### Activation ![image](https://hackmd.io/_uploads/HkdJ3T7Pp.png =x170)![image](https://hackmd.io/_uploads/Bk4E3pXDa.png =x200) ### Softmax ![image](https://hackmd.io/_uploads/Bkttha7D6.png =x250) ### Cost func ![image](https://hackmd.io/_uploads/Hy6a6TXw6.png =x120) ![image](https://hackmd.io/_uploads/ry0CT67wp.png =x100) ### AdaGrad ![image](https://hackmd.io/_uploads/HJc0lAmva.png =x250) ![image](https://hackmd.io/_uploads/rJH2bCXva.png =x200) ### Back propagation ![image](https://hackmd.io/_uploads/HkhRNCXPp.png =x200) ![image](https://hackmd.io/_uploads/Sk8yBR7Pa.png =x200) ![image](https://hackmd.io/_uploads/HysHfAXP6.png =x250) ![image](https://hackmd.io/_uploads/rkDhGRQPa.png =x350)![image](https://hackmd.io/_uploads/BkexmA7v6.png =x90) ## Convolution neutral network https://www.youtube.com/watch?v=gv693bG9K98&t=2626s ![image](https://hackmd.io/_uploads/ByZhaDrnp.png) ![image](https://hackmd.io/_uploads/HyUsOkVP6.png =x200) ### Convolution ![image](https://hackmd.io/_uploads/SyOTIkVDp.png =x200) P(Padding): thêm viền ngoài ảnh. S(Strike): khoảng cách giữa 2 lần chiếu lên ảnh. F(kernal size) ### Pooling reduce the number of feature-map coefficients to process ![image](https://hackmd.io/_uploads/SyzyCJNvT.png =x150) ![image](https://hackmd.io/_uploads/B1epFAJVDT.png =x250) ![image](https://hackmd.io/_uploads/B1a4TvS36.png =x200) ### Một số mạng https://www.youtube.com/watch?v=gv693bG9K98&t=2626s * VGG-16(nay đã có VGG-19) * Inception(GoogleLeNet) * ResNet * Darknet-53 * ## Tham khảo https://drive.google.com/drive/folders/1kmCWMLzL5sez5RK4v4Q7jXjrVTcqDAVU https://machinelearningcoban.com/2017/01/04/kmeans2/