'''
np.random.seed(0)
sgd_test(Sequential([Linear(2,3), Tanh(), Linear(3,2), SoftMax()], NLL()), test_1_values)
'''
# TEST 2: sgd_test for ReLU activation and SoftMax output
'''
np.random.seed(0)
sgd_test(Sequential([Linear(2,3), ReLU(), Linear(3,2), SoftMax()], NLL()), test_2_values)
'''
######################################################################
# TEST 3: you should achieve 100% accuracy on the hard dataset (note
# that we provided plotting code)
'''
X, Y = hard()
nn = Sequential([Linear(2, 10), ReLU(), Linear(10, 10), ReLU(), Linear(10,2), SoftMax()], NLL())
disp.classify(X, Y, nn, it=100000)
'''
# TEST 4: try calling these methods that train with a simple dataset
def nn_tanh_test():
np.random.seed(0)
nn = Sequential([Linear(2, 3), Tanh(), Linear(3, 2), SoftMax()], NLL())
X, Y = super_simple_separable()
nn.sgd(X, Y, iters=1, lrate=0.005)
return [
np.vstack([nn.modules[0].W, nn.modules[0].W0.T]).tolist(),
np.vstack([nn.modules[2].W, nn.modules[2].W0.T]).tolist()
# TEST 2: sgd_test for ReLU activation and SoftMax output
np.random.seed(0)
sgd_test(Sequential([Linear(2,3), ReLU(), Linear(3,2), SoftMax()], NLL()), test_2_values)
######################################################################
# TEST 3: you should achieve 100% accuracy on the hard dataset (note
# that we provided plotting code)
X, Y = hard()
nn = Sequential([Linear(2, 10), ReLU(), Linear(10, 10), ReLU(), Linear(10,2), SoftMax()], NLL())
disp.classify(X, Y, nn, it=100000)
# TEST 4: try calling these methods that train with a simple dataset
def nn_tanh_test():
np.random.seed(0)
nn = Sequential([Linear(2, 3), Tanh(), Linear(3, 2), SoftMax()], NLL())
X, Y = super_simple_separable()
nn.sgd(X, Y, iters=1, lrate=0.005)
return [np.vstack([nn.modules[0].W, nn.modules[0].W0.T]).tolist(),
np.vstack([nn.modules[2].W, nn.modules[2].W0.T]).tolist()]
def nn_relu_test():
np.random.seed(0)
nn = Sequential([Linear(2, 3), ReLU(), Linear(3, 2), SoftMax()], NLL())