Submitted by: | Sec. | B.N. |
---|---|---|
Alaa Allah Essam Abdrabo | 1 | 13 |
-
-
- Using built in function that scales values in the range [0, 1]:
- That results in :
- scales values to have mean 0 and standard deviation 1:
- That results in :
- normalization implementation from scratch based on min & max of train and normalize test data with same min & max of train :
- That results in :
3. normalization implementation from scratch based on min & max of train and normalize test data with same min & max of train :
we notice that the average accuracy in case of using normalization is higher than without normalization
- had no missing values
- no text as all data points are numeric
-
-
- C : is the hyperparameter "Regularization Constant" that determines to what extent the soft margine would be(15.0)
- B : is the beta in hyperplane equation [h(x)=B1X1+B2X2+....+b] has number of values according to number of features
- b : is the bias in previous equation
- Learning rate: 0.001
- number of iterations: 500
- For binary classification only two classes of data were used so, the last 100 points were extracted being of 2 classes
- In this algorithm equations were designed according to classes with codes of -1 & 1 then all 0 class were just encoded to be -1 and the other class was already of code 1