摘要: 逻辑回归的另一种观点 \[{h_\theta }\left( x \right) = \frac{1}{{1 + {e^{ - {\theta ^T}x}}}}\] 如果y=1,我们希望hθ(x)≈1,对应θTx >> 0 如果y=0,我们希望hθ(x)≈0,对应θTx << 0 对于一个样本(x, 阅读全文
posted @ 2018-11-01 19:15 qkloveslife 阅读(227) 评论(0) 推荐(0) 编辑
摘要: 2001年Bank和Bill做了这么一个实验 区分容易混淆的词,如(to, two, too) 比如:For breakfast I ate two eggs. 他们用了不同的算法: Perceptron (Logistic regression) Winnow Memory-based Naïve 阅读全文
posted @ 2018-11-01 11:38 qkloveslife 阅读(196) 评论(0) 推荐(0) 编辑
摘要: 对于癌症检测的例子来说,y=1代表有癌症(1代表数目比较小的类) Precision/Recall \[\Pr ecision = \frac{{True \bullet positive}}{{predicted \bullet positive}} = \frac{{True \bullet p 阅读全文
posted @ 2018-11-01 10:07 qkloveslife 阅读(1411) 评论(0) 推荐(0) 编辑
摘要: Rcommended approach Start with a simple algorithm that you can implement quickly. Implement it and test it on your cross-validation data. Plot learnin 阅读全文
posted @ 2018-11-01 09:12 qkloveslife 阅读(700) 评论(0) 推荐(0) 编辑
摘要: “Small” neural network (fewer parameters; more prone to underfitting) Computationally cheaper "Large" neural network (more parameters; more prone to o 阅读全文
posted @ 2018-11-01 02:07 qkloveslife 阅读(315) 评论(0) 推荐(0) 编辑
摘要: 学习曲线 “训练误差”和“交叉验证误差”如下 \[\begin{array}{l}{J_{train}}\left( \theta \right) = \frac{1}{{2{m_{train}}}}\sum\limits_{i = 1}^{{m_{train}}} {{{\left( {{h_\t 阅读全文
posted @ 2018-11-01 01:52 qkloveslife 阅读(352) 评论(0) 推荐(0) 编辑