https://gitlab.cs.nuim.ie/u210251/cs401-hw2-great-fun
- numpy
- pandas
- sklearn
- matplotlib
- joblib
- random
- sysimport
- time
- sys
- Sigmoid
- Gaussian
- Polynomial
- C
- gama
- coef0
One of the Matrix: (From Test_Set which is 20%*Train_Set)
I take ten point as follows:
-
(FPR=5.55014606e-02, TPR=0.03597122)
-
(FPR=0.11613566, TPR=1.51898734e-01)
-
(FPR=0.17574512, TPR=2.56085686e-01)
-
(FPR=0.24665982, TPR=3.54430380e-01)
-
(FPR=0.32168551, TPR=4.50827653e-01)
-
(FPR=0.40904419, TPR=5.51119766e-01)
-
(FPR=0.50154162, TPR=6.51411879e-01)
-
(FPR=0.60842754, TPR=7.54625122e-01)
-
(FPR=0.72970195, TPR=8.50048685e-01)
-
(FPR=0.89825283, TPR=9.51314508e-01)
![0.75_Polynomial_ROC_by_hand_20k20-12-19 16.59.35](./result_graph/0.75_Polynomial_ROC__2020-12-19 16.59.35.jpg)
![0.75_Polynomial_ROC__2020-12-19 16.59.35](./result_graph/0.75_Polynomial_ROC__2020-12-19 16.59.35.png)
Polynomial is the best choice
- Gaussian.
I had try C=0.01 to C=1.0
- Sigmoid
I had try C=0.01 to C=1.0
- improvement of accuracy.
- I‘m trying to do it
-
Since the accuracy of this model is over 50%, the overall accuracy should be much higher if multiple predictions are made (e.g., 100 times) and the result with the most occurrences is taken as the final result.
-
May be ensemble learning is better.