def test_mse_outputdim():
x = np.array([[1,2], [3,4]])
y = np.array([[1,2], [3,4]])
v1 = AD(1., 'v1')
v2 = AD(1., 'v2')
with pytest.raises(AssertionError):
boomdiff.loss_function.linear_mse(x, y, [v1, v2])
def test_array_ops_2arrays(ar_x, ar_y):
assert (ar_x + ar_y)[0].func_val == 4.5
assert (ar_x + ar_y)[1].func_val == 15
assert (ar_x + ar_y)[0].partial_dict['x_0'] == 1
assert (ar_x + ar_y)[0].partial_dict['y_0'] == 1
assert (ar_x + ar_y)[1].partial_dict['x_1'] == 1
assert (ar_x + ar_y)[1].partial_dict['y_1'] == 1
assert (ar_x - ar_y)[0].func_val == -1.5
assert (ar_x * ar_y)[0].func_val == 4.5
assert (ar_y / ar_x)[0].func_val == 2
assert (ar_x - ar_y)[1].partial_dict['x_1'] == 1
assert (ar_x - ar_y)[1].partial_dict['y_1'] == -1
assert (ar_x * ar_y)[1].partial_dict['x_1'] == 5
assert (ar_x * ar_y)[1].partial_dict['y_1'] == 10
assert (ar_y / ar_x)[1].partial_dict['x_1'] == -0.05
assert (ar_y / ar_x)[1].partial_dict['y_1'] == 0.1
assert (AD(1, 'x') +
AD.from_array(np.array([1, 2, 3, 4])))[1].func_val == 3
assert (AD(1, 'x') -
AD.from_array(np.array([1, 2, 3, 4])))[1].func_val == -1
assert (AD(1, 'x') *
AD.from_array(np.array([1, 2, 3, 4])))[1].func_val == 2
assert (AD(1, 'x') /
AD.from_array(np.array([1, 2, 3, 4])))[1].func_val == 0.5
assert (AD(10, 'x')**AD.from_array(np.array([1, 2, 3,
4])))[1].func_val == 100
def test_ce_outputdim():
x = np.array([[1,2], [3,4]])
y = np.array([[1,0], [1,1]])
v1 = AD(1., 'v1')
v2 = AD(1., 'v2')
with pytest.raises(AssertionError):
boomdiff.loss_function.logistic_cross_entropy(x, y, [v1, v2])
def test_rowsum():
x = np.array([[1, 2, 3], [4, 5, 6]])
v1 = AD(0.0, 'v1')
v2 = AD(0.0, 'v2')
v3 = AD(0.0, 'v3')
f = boomdiff.loss_function._rowsums(x, [v1, v2, v3])
assert f[0] == AD(0.0, {'v1': 1.0, 'v2': 2.0, 'v3':3.0})
assert f[1] == AD(0.0, {'v1': 4.0, 'v2': 5.0, 'v3': 6.0})
def test_optim_exc(var1, var2):
opt = boomdiff.optimize._gradient_descent.GD(learning_rate=0.1)
loss = lambda: var1**2 + var2**2
opt.step(loss, var_list=[var1, var2], learning_rate=0.2, record=True)
assert var1 == AD(60, {'var1': 1})
assert var2 == AD(0.6, {'var2': 1})
with pytest.raises(Exception):
opt.step(loss, var_list=[var1, var2], learning_rate='a')
def test_optim_lists(var1, var2):
opt = boomdiff.optimize._gradient_descent.GD(learning_rate=0.1)
loss = lambda: var1**2 + var2**2
opt.minimize(loss,
var_list=[var1, var2],
steps=2,
learning_rates=[0.2, 0.4],
record=False)
assert var1 == AD(12.0, {'var1': 1})
assert var2 == AD(0.12, {'var2': 1})
def test_equality_partial():
# Test equality of partial dictionary
x = AD(12, 'x1')
y = AD(12, 'x2')
assert (x == y) == False
def test_equality_fval():
# Test equality of func_val
x = AD(12, 'x1')
z = AD(13)
assert (x == z) == False
def test_evaluate():
x = AD(4.0, 'x')
y = AD(5., 'y')
f = 2*x + 2*y
assert f.evaluate() == (18.0, {'x': 2.0, 'y': 2.0})
def test_name():
x = AD(2.5, 'y')
assert x.name() == ['y']
def v2():
v2 = AD(0.0, 'v2')
return v2
def test_inequality_partial():
# Testinequality of partial dictionary
x = AD(12, 'x1')
y = AD(12, 'x2')
assert x != y
def test_inequality():
# Test basic inequality
x = AD(12, 'x1')
y = AD(9)
assert x != y
def test_nondict_init():
# Test that if der_dict passed, it is dictionary
with pytest.raises(ValueError):
x = AD(1.5, 9.3)
def test_nonnum_dict_init():
# Test values of passed dictionary are int or float
with pytest.raises(ValueError):
x = AD(3.6, {'x1': 'bad'})
def test_badbaseval():
# Test initialize base class without float or integer type
with pytest.raises(ValueError):
x = AD('v')
def test_bast_nonstringname():
with pytest.raises(ValueError):
x = AD(3.0, 3)
def test_base_setstr():
x = AD(1.0, 'x2')
assert x == AD(1.0, {'x2': 1.})
def test_improper_logbase():
x = AD(3)
with pytest.raises(Exception):
AD.log(x, base = AD(4))
def test_base():
x = AD(1, {'x1':1})
assert x.func_val == 1
print(x.func_val)
assert x.partial_dict['x1'] == 1
assert len(x.partial_dict.keys()) == 1
def test_equality_obj():
# test equality with non AD object
x = AD(12, 'x1')
assert (x == 12) == False
def test_set_nonatt():
# Test that set_params(att) must be one of 'func_val' or 'partial_dict'
with pytest.raises(ValueError):
x = AD(1.3)
x.set_params('evaluation_point', 2)
def test_inequality_rev():
# Test reverse of inequality
x = AD(12, 'x1')
y = AD(12, 'x1')
assert (x != y) == False
def test_set_funcval():
# If setting 'func_val', must be type float or int
with pytest.raises(ValueError):
x = AD(13.2)
x.set_params('func_val', '2.3')
def v1():
v1 = AD(0.0, 'v1')
return v1
def test_setparam_dictvals():
# 'partial_dict' values must be integers or floats
with pytest.raises(ValueError):
x = AD(12)
x.set_params('partial_dict', {'x1': '3'})
def test_funcval():
x = AD(3.0)
assert x.value() == 3.0
def test_equality():
# Test equality of operations
x = AD(12, 'x1')
y = AD(12)
assert x == y
def test_set_dictionary():
# If set 'partial_dict', must be dictionary
with pytest.raises(ValueError):
x = AD(13.2)
x.set_params('partial_dict', 2)
def test_ders():
x = AD(3.0, 'x')
y = AD(4.0, 'y')
f = x + y
assert f.ders() == {'x': 1.0, 'y': 1.0}