import exemplos from diff.revediff import var, const, df import numpy as np f = lambda x: 2 * x + 1 xi = var(1) x = np.array([xi, xi + 1, xi + 2]) dx = df(f, x) print(f(x))
import exemplos
from diff.revediff import var, const, df
xs = [i / 10 for i in range(11)]
ys = [x**2 + 0.7 * x + 0.1 for x in xs]
loss = lambda x, y, a, b, c: (a * x**2 + b * x + c - y)**2
err = lambda a, b, c: sum(
[loss(xs[i], ys[i], a, b, c) for i in range(len(xs))])
a = var(1)
b = var(0)
c = var(1)
lr = 0.01
for k in range(10000):
tam = len(xs)
da, db, dc = df(err, a, b, c)
a = a - lr * da
b = b - lr * db
c = c - lr * dc
print(a, b, c)
import exemplos
from diff.revediff import var, const, df
xs = [i / 10 for i in range(11)]
ys = [2 * x + 0.5 for x in xs]
loss = lambda x, y, a, b: (a * x + b - y)**2
err = lambda a, b: sum([loss(xs[i], ys[i], a, b) for i in range(len(xs))])
a = var(1)
b = var(0)
lr = 0.01
for k in range(100):
tam = len(xs)
da, db = df(err, a, b)
a = a - lr * da
b = b - lr * db
print(a, b)
import exemplos
from diff.revediff import var, const, df
f = lambda x: x**2 - 3 * x + 1
## gradiente descent para encontrar df(x) = 0
x = var(0) # chute inicial
lr = 0.01 # tamanho do passos
err = 0.001 # |df(x) - df(xi)|
n = 1000 #numero máximo de iterações
for i in range(n):
dy = df(f, x) # return dy
x = var(x - dy * lr)
# ajustando os passos
if abs(dy) <= err: # verificando se
print("stop at it: {} x = {}".format(i + 1, x))
break
if i == (n - 1):
print("não houve convergência>: it{}".format(n))
#teste.py import exemplos from diff.revediff import df, var, const, new_df f = lambda x, y: x / (x**2 + y**2) x = var(1) y = const(1) fxy, dx = new_df(f)(x, y) print(fxy, dx)