def test_apply_softmax(self):
x = np.random.randn(100, 10)
eig_x = mat.EigenMatrix(x)
eig_y = mat.empty((100, 10))
eig_x.apply_softmax(target=eig_y)
y = np.exp(x - x.max(axis=0))
y /= y.sum(axis=0)
diff = ((eig_y.asarray() - y)**2).sum()
self.assertAlmostEqual(diff, 0, places=5)
def test_add(self):
x = np.random.randn(10, 10)
y = np.random.randn(10, 10)
eig_x = mat.EigenMatrix(x)
eig_y = mat.EigenMatrix(y)
eig_z = mat.empty(x.shape)
z = x + y # Numpy add.
eig_x.add(eig_y, target=eig_z) # EigenMat add.
diff = ((eig_z.asarray() - z)**2).sum()
self.assertAlmostEqual(diff, 0)
def test_dot_transposed(self):
x = np.random.randn(500, 1000)
y = np.random.randn(600, 1000)
eig_x = mat.EigenMatrix(x)
eig_y = mat.EigenMatrix(y)
eig_z = mat.empty((x.shape[0], y.shape[0]))
z = x.dot(y.T)
mat.dot(eig_x, eig_y.T, target=eig_z)
diff = ((eig_z.asarray() - z)**2).sum()
self.assertAlmostEqual(diff, 0, places=5)
def test_add(self):
x = np.random.randn(10, 10)
y = np.random.randn(10, 10)
eig_x = mat.EigenMatrix(x)
eig_y = mat.EigenMatrix(y)
eig_z = mat.empty(x.shape)
z = x + y # Numpy add.
eig_x.add(eig_y, target=eig_z) # EigenMat add.
diff = ((eig_z.asarray() - z)**2).sum()
self.assertAlmostEqual(diff, 0)
def test_apply_softmax(self):
x = np.random.randn(100, 10)
eig_x = mat.EigenMatrix(x)
eig_y = mat.empty((100, 10))
eig_x.apply_softmax(target=eig_y)
y = np.exp(x - x.max(axis=0))
y /= y.sum(axis=0)
diff = ((eig_y.asarray() - y)**2).sum()
self.assertAlmostEqual(diff, 0, places=5)
def test_dot_transposed(self):
x = np.random.randn(500, 1000)
y = np.random.randn(600, 1000)
eig_x = mat.EigenMatrix(x)
eig_y = mat.EigenMatrix(y)
eig_z = mat.empty((x.shape[0], y.shape[0]))
z = x.dot(y.T)
mat.dot(eig_x, eig_y.T, target=eig_z)
diff = ((eig_z.asarray() - z)**2).sum()
self.assertAlmostEqual(diff, 0, places=4)
import matplotlib.pyplot as plt
import numpy as np
plt.ion()
import eigenmat as mat
mat.EigenMatrix.init_random(seed=1)
plt.figure(1)
plt.clf()
x = mat.empty((100, 100))
x.fill_with_randn()
plt.hist(x.asarray().flatten(), 100)
plt.figure(2)
plt.clf()
y = np.random.randn(100, 100)
plt.hist(y.flatten(), 100)
raw_input('Press Enter.')
