def generate_paths(self, fixed_seed=True, day_count=365.):
if self.time_grid is None:
self.generate_time_grid()
M = len(self.time_grid)
I = self.paths
paths = np.zeros((M, I))
paths_ = np.zeros_like(paths)
paths[0] = self.initial_value
paths_[0] = self.initial_value
if self.correlated is False:
rand = sn_random_numbers((1, M, I), fixed_seed=fixed_seed)
else:
rand = self.random_numbers
for t in range(1, len(self.time_grid)):
if self.correlated is False:
ran = rand[t]
else:
ran = np.dot(self.cholesky_matirx, rand[:, t, :])
ran = ran[self.rn_set]
paths_[t] = (paths_[t - 1] + self.kappa *
(self.theta - npmaximun(0, paths_[t - 1, :])) *
self.volatility * np.sqrt(dt) * ran)
paths[t] = np.maximun(0, paths_[t])
self.instrument_values = paths
def train(epoch):
model.train()
train_loss = 0
temperature = opt.temperature
for bi, (data, _) in enumerate(train_loader):
data = data.to(device)
optimizer.zero_grad()
recon_output, encode_output = model(data, temperature)
loss = loss_function(recon_output, data, encode_output)
loss.backward()
train_loss += loss[0].item()
optimizer.step()
if (bi + 1) % 100 == 0:
temperature = np.maximun(temperature * np.exp(-ANNEAL_RATE * bi),
temperature_min)
if (bi + 1) % opt.log_interval == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, bi * len(data), len(train_loader.dataset),
100. * bi / len(train_loader), loss.data[0] / len(data)))
print('====> Epoch: {} Average loss: {:.4f}'.format(
epoch, train_loss / len(train_loader.dataset)))
def test(epoch):
model.eval()
test_loss = 0
temperature = opt.temperature
with torch.no_grad():
for bi, (data, _) in enumerate(test_loader):
data.to(device)
recon_output, encode_output = (data, temperature)
test_loss += loss_function(recon_output, data,
encode_output)[0].item()
if (bi + 1) % 100 == 0:
temperature = np.maximun(
temperature * np.exp(-ANNEAL_RATE * bi), temperature_min)
if (bi + 1) % 100 == 0:
n = min(data.size(0), 8)
comparison = torch.cat([
data[:n],
recon_output.view(opt.batch_size, 1, 28, 28)[:n]
])
save_image(comparison.data.cpu(),
'results/reconstruction_' + str(epoch) + '.png',
nrow=n)
test_loss /= len(test_loader.dataset)
print('====> Test set loss: {:.4f}'.format(test_loss))
def compute_ap(recall,precision):
'''
compute the average precision given the recall and precision curve
#arguments:
recall:list
precision:list
'''
#correct Ap calculation
mrec = np.concatenate(([0.0],recall,[1.0]))
mpre = np.concatenate(([0.0],precision,[0.0]))
#calculate the precision envelope
for i in range(mpre.size-1,0,-1):
mpre[i-1] = np.maximun(mpre[i-1],m[i])
#to calculate area under PR curve ,look for points
#where x axis (recall) changes value
i = up.where(mrec[1:] != mrec[:-1])[0]
ap = np.sum((mrec[i+1] - mrec[i])* mpre[i+1])
return ap
deltas[0], axis=0, keepdims=True) * lr
neural_net[l].w = neural_net[l].w - out[l][1].T @ deltas[0] * lr
return out[-1][1]
def create_nn(topology, act_f):
nn = [] # Vector de capas
for l, layer in enumerate(topology[:-1]):
nn.append(neural_layer(topology[l], topology[l + 1], act_f))
return nn
# Funciones de activacion
sigm = (lambda x: 1 / (1 + np.e**(-x)), lambda x: x * (1 - x))
relu = lambda x: np.maximun(0, x)
# Dataset
n = 500 # Population
p = 2 # Characteristics
from sklearn.datasets import make_blobs
X, Y = make_blobs(n_samples=n,
cluster_std=0.5,
n_features=2,
centers=[(-1, 1), (1, 1)])
Y = Y[:, np.newaxis]
#show info
plt.scatter(X[Y[:, 0] == 1, 0], X[Y[:, 0] == 1, 1], color='salmon')
plt.scatter(X[Y[:, 0] == 0, 0], X[Y[:, 0] == 0, 1], color='skyblue')
plt.axis('equal')
plt.show()
# H is hidden dimension; D_out is output dimension
N, D_in, H, D_out = 64, 1000, 100, 10
# Create random input and output data
x = np.random.randn(N, D_in)
y = np.random.randn(N, D_out)
# Randomly initialize weights
w1 = np.random.randn(D_in, H)
w2 = np.random.randn(H, D_out)
lr = 1e-6
for t in range(500):
# forward pass
h = x.dot(w1)
h_relu = np.maximun(h, 0)
y_pred = h_relu.dot(w2)
loss = np.square(y_pred - y).sum()
print(t, loss)
#backprop
grad_y_pred = 2 * (y_pred - y)
grad_w2 = h_relu.T.dot(grad_y_pred)
grad_h_relu = grad_y_pred.dot(w2.T)
grad_h = grad_h_relu.copy()
grad_h[h < 0] = 0
grad_w1 = x.T.dot(grad_h)
#update weights
w1 -= lr * grad_w1
def relu(z):
return np.maximun(0, z)
def relu(x):
return np.maximun(0, x)
def LeakyReLU(x, alpha): ### Leaky Rectified Linear Unit activation
## If 'alpha' is equal to zero, then it becomes a standard ReLU
return np.maximun(x, alpha * x)
def run_cppi(risky_r,
safe_r=None,
start=1000,
floor=0.8,
drawdown=None,
riskfree_rate=0.03,
m=3):
"""
Runs a backtest of the CPPI strategy, given a set of returns for the risky asset
Returns a dictionary containing: Asset Value History, Risk budged History, Risky Weight History
"""
#set up CPPI Parameters
dates = risky_r.index
n_steps = len(dates)
account_value = start
floor_value = start * floor
peak = start
m = m
if isinstance(risky_r, pd.Series):
risky_r = pd.DataFrame(risky_r, columns["R"])
if safe_r is None:
safe_r = pd.DataFrame().reindex_like(risky_r)
safe_r.values[:] = riskfree_rate / 12 #fast way to fill with numbers
account_history = pd.DataFrame().reindex_like(risky_r)
risky_w_history = pd.DataFrame().reindex_like(risky_r)
cushion_history = pd.DataFrame().reindex_like(risky_r)
for step in range(n_steps):
if drawdown is not None:
peak = np.maximun(peak, account_value)
floor_value = peak * (1 - drawdown)
cushion = (account_value - floor_value) / account_value
risky_w = m * cushion
risky_w = np.minimum(risky_w, 1) #wont leverage
risky_w = np.maximum(risky_w, 0) #wont go short
safe_w = 1 - risky_w
risky_alloc = account_value * risky_w
safe_alloc = account_value * safe_w
##update the account value for this timestamp
account_value = (risky_alloc *
(1 + risky_r.iloc[step])) + (safe_alloc *
(1 + safe_r.iloc[step]))
## save the current values
cushion_history.iloc[step] = cushion
risky_w_history.iloc[step] = risky_w
account_history.iloc[step] = account_value
risky_wealth = start * (1 + risky_r).cumprod()
backtest_result = {
"Wealth": account_history,
"Risky Wealth": risky_wealth,
"Risk Budget": cushion_history,
"Risky Allocation": risky_w_history,
"m": m,
"start": start,
"risky_r": risky_r,
"safe_r": safe_r
}
return backtest_result
def forward(self,inputs):
self.output = np.maximun(0,inputs)
import numpy as np
import matplotlib.pyplot as plt
D = np.random.randn(1000, 500)
hidden_layer_sizes = [500] * 10
#***************active funtion "tanh"
nonlinearities = ['tanh'] * len(hidden_layer_sizes)
#***************active funtion "tanh"
#nonlinearities = ['relu']*len(hidden_layer_sizes)
act = {'relu': lambda x: np.maximun(0, x), 'tanh': lambda x: np.tanh(x)}
Hs = {}
for i in xrange(len(hidden_layer_sizes)):
X = D if i == 0 else Hs[i - 1] #input layer
fan_in = X.shape[1]
fan_out = hidden_layer_sizes[i]
#-------little random number
#W = np.random.randn(fan_in, fan_out) * 0.01
#-------zero
#W = 0
#-------big number
W = np.random.randn(fan_in, fan_out) * 1
#-------xavier and tanh (xavier and relu)
#W = np.random.randn(fan_in, fan_out) / np.sqrt(fan_in)
#-------Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification by He et al., 2015
#W = np.random.randn(fan_in, fan_out) / np.sqrt(fan_in/2)
H = np.dot(X, W)
H = act[nonlinearities[i]](H)
Hs[i] = H # cache result on this layer
np.modf(x) : 将数组各元素的小数和整数部分以两个独立数组形式返回 np.cos(x) np.cosh(x) np.sin(x) np.sinh(x) np.tan(x) np.tanh(x) : 计算数组各元素的普通型和双曲型三角函数 np.exp(x) ：计算数组各元素的指数值 np.sign(x) ：计算数组各元素的符号值，1(+)，0，-1(-) """ a = np.array([2, 3, -4, 5]) print(np.sign(a)) # [ 1 1 -1 1] print(np.modf(a)) # (array([ 0., 0., -0., 0.]), array([ 2., 3., -4., 5.])) print(np.square(a)) # [ 4 9 16 25] """ 二元函数 + - * / ** : 两个数组各元素进行对应运算 np.maximun(x,y) 或np.fmax() : 元素级的最大值 np.minimun(x,y) 或np.fmin() : 元素级的最小值 np.mod(x, y) : 元素级的模运算 np.copysign(x, y) : 将数组y中各元素值的符号赋值给数组x对应的元素 > < >= <= == != : 算术比较，产生布尔型数组 """
