def test1():
(x, y) = DataModel.loadData("..\\train.csv")
(x_train, x_cv, y_train, y_cv) = DataModel.splitData(x, y)
x_sub = x_train[:500,:]
y_sub = y_train[:500]
s = SimpleNN.SimpleNN([784, 70, 10])
#s = Train.trainGradientDescent(s, x_sub, y_sub, 5)
s = Train.trainSciPy(s, x_sub, y_sub, 5)
acc_cv = accuracy_score(y_cv, [s.predictClass(w) for w in x_cv])
print("Accuracy on CV set: {0}", acc_cv)
from scipy.optimize import check_grad
import SimpleNN
import Train
mat = loadmat('..\\ex4weights.mat')
s = SimpleNN.SimpleNN([400, 25, 10])
s.theta = [np.transpose(mat["Theta1"]), np.transpose(mat["Theta2"])]
data = loadmat("..\\ex4data1.mat")
x = data["X"][:500,:]
y = data["y"][:500]
#check_grad(func = lambda p: s.computeCost(s.combineTheta(p), x, y, 0.5),
# grad = lambda p: s.computeGrad(s.combineTheta(p), x, y, 0.5),
# x0 = s.combineTheta(s.theta))
#(cost, grad) = s.computeCostGrad(s.theta, x, y, 1)
s = Train.trainSciPy(s, x, y, 0.05)
predictions = [s.predictClass(w) for w in x]
err_rate = np.mean([1 if pred != check else 0 for (pred, check) in zip(y, predictions)])
print("Error rate with pre-computed paramaters: {0}".format(err_rate))
