def test_Exponent():
a = Tensor.randn(8, 10, requires_grad=True)
o = MLlib.exp(a)
o.backward()
if not_close(a.grad.data, o.data):
raise AssertionError
def test_ExponentWithMathModule():
a = Tensor.randn(8, 10, requires_grad=True)
o = math.e**a
o.backward()
if not_close(a.grad.data, o.data):
raise AssertionError
def test_Cos():
a = Tensor.randn(6, 8, requires_grad=True)
o = MLlib.cos(a)
o.backward()
if not_close(a.grad.data, -np.sin(a.data)):
raise AssertionError
def test_Tan():
a = Tensor.randn(6, 8, requires_grad=True)
o = MLlib.tan(a)
o.backward()
if not_close(a.grad.data, 1 / (np.cos(a.data))**2):
raise AssertionError
def __init__(self, in_features, out_features):
self.bias = Tensor(0., requires_grad=True)
self.weight = Tensor.randn(out_features, in_features)
self.weight.requires_grad = True
from MLlib import Tensor
from MLlib.regularizer import LinearRegWith_Regularization
from MLlib.regularizer import L1_Regularizer
from MLlib.optim import SGDWithMomentum
from MLlib.utils.misc_utils import printmat
import numpy as np
np.random.seed(5322)
x = Tensor.randn(10, 8) # (batch_size, features)
y = Tensor.randn(10, 1)
reg = LinearRegWith_Regularization(8,
L1_Regularizer,
optimizer=SGDWithMomentum,
Lambda=7)
# Regularizer,optimizer and Lambda as per user's choice
printmat("Total Loss", reg.fit(x, y, 800))