Regression
预测模型
1.Model
Linear model：
y
=
b
+
∑
w
i
x
i
y = b+ \sum w_i x_i
y=b+∑wixi
2.Goodness of Function
Loss Function : output（ how bad it is）
L
(
f
)
=
∑
i
=
1
N
(
y
n
−
(
b
+
w
x
j
i
)
)
2
L(f) = \sum_{i=1}^{N} (y^n -(b+w x_j ^i))^2
L(f)=i=1∑N(yn−(b+wxji))2
3.Best Function
w
∗
,
b
∗
=
arg
min
w
,
b
∑
i
=
1
N
(
y
n
−
(
b
+
w
x
j
i
)
)
2
w^{*},b^* = \mathop{\arg\min_{w,b}} \sum_{i=1}^{N} (y^n -(b+w x_j ^i))^2
w∗,b∗=argw,bmini=1∑N(yn−(b+wxji))2
Gradient Descent （无约束最优化问题）
min
x
∈
R
n
f
(
x
)
\mathop{\min_{x \in R^n}} f(x)
x∈Rnminf(x)
If in the convex (eg linear regression) will find the global minimum
Algorithm
- get x 0 x^{0} x0
- calculate the
g
k
=
g
(
x
k
)
g_k=g(x^{k})
gk=g(xk) ,when
g
k
<
ε
g_k < \varepsilon
gk<ε and
x
∗
=
x
k
x^{*} = x^{k}
x∗=xk,end the algorithm. otherwise ,
f ( x ( k ) − λ k g k ) = min λ ≥ 0 ( f ( x ( k ) − λ g k ) ) f(x^{(k)}-\lambda_{k} g_k) = \mathop{\min_{\lambda \geq0}}(f(x^{(k)}-\lambda g_k)) f(x(k)−λkgk)=λ≥0min(f(x(k)−λgk))
885
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
