吴恩达机器学习 - 课程总结

2018-12-03 本文已影响0人 YANWeichuan

监督学习

算法	h_θ(x)	cost function	求解	特性
线性回归	$h_θ(x) = θ_0+ θ_1x$	$J( θ_0, θ_1) = \frac{1}{2m}\sum_{i=1}^{m}{(h_\theta(x^i) - y^i)^2}$	$θ_1 = θ_1 + α{\frac{1}{m}\sum_{i=1}^{m}{(h_θ(x^i) - y^i)x^i}}$
逻辑回归	$0<= h_θ(x) <= 1$ $h_θ(x) = g (θ^Tx)$ $g(z) = {\frac{1}{1+ e^{-z}}}$	$J(θ)=-{\frac{1}{m}}\left[\sum_{i=1}^{m}{y_i^{(i)}log(h_θ(x^{(i)})} + {(1-y_i^{(i)})log(1-(h_θ(x^{(i)})))}\right]$	$θ_j = θ_j - α{\frac{1}{m}\sum_{i=1}^{m}{(h_θ(x^i) - y^i)x^i}}$
神经网络	sigmoid	$J(Θ)=-{\frac{1}{m}}\left[\sum_{i=1}^{m}\sum_{j=1}^{K}{y_k^{(i)}log(h_Θ(x^{(i)})_k)} + {(1-y_k^{(i)})log(1-(h_Θ(x^{(i)}))_k)}\right] + {\frac{λ}{2m}}\sum_{l=1}^{L-1}\sum_{i=1}^{S_l}\sum_{j=1}^{S_l+1}Θ_{ji}^l$	反向传播
SVM		$min C{\frac{1}{m}}\sum_{i=1}^{m}{y_i^{(i)}Cost_1(h_θ(x^{(i)})} + {(1-y_i^{(i)})Cost_0(h_θ(x^{(i)}))}$	无	决策边界