2_Classification

Assignment2

在kaggle上的结果(评判标准是Categorization Accuracy)：

• simple_baseline = 0.88675

• strong_baseline = 0.89102

• private score为0.89110对应的排名是99/285

• 手写linear的cross entropy的gradient descent的performance往往会比用keras写神经网络要来的好

• 特征工程是进一步优化的思路

method	Public Score	Private Score
predict_adagrad_gd	0.88878	0.89110
predict_adagrad_gd	0.88929	0.89052
predict_adagrad_sgd	0.88842	0.88863
predict_keras	0.88769	0.88900
predict_keras	0.88798	0.88675