def calculate_accurracy(root, noOfAcids, kMers, train_file, test_file, laplace_alpha, train_end_index = -1):
csv_path = os.path.join(root, test_file)
test_x, test_y = bs._load_dataset(csv_path)
res, _, _ = bs.result_bayes(root, train_file, test_x, kMers, noOfAcids, laplace_alpha, train_end_index)
#Find index of elements where we predicted cleavable
trueIndices = np.where(np.array(test_y) == 1)
#Find index of elements where we predicted nonCleavable
falseIndices = np.where(np.array(test_y) == 0)
#Generate results
accuracy = ((np.sum(res[0,trueIndices]) + (np.size(falseIndices) - np.sum(res[0,falseIndices])))/len(test_x))
return accuracy
def part1(root='./Dataset',
trainfile='q2_train_set.txt',
testfile='q2_test_set.txt'):
#Load Datasets
noOfMers = 8
noOfAcids = 20
csv_path = os.path.join(root, trainfile)
train_x, train_y = bs._load_dataset(csv_path)
csv_path = os.path.join(root, testfile)
test_x, test_y = bs._load_dataset(csv_path)
#Train
myRes, _, _ = bs.result_bayes(root, trainfile, test_x, noOfMers, noOfAcids)
#Find index of elements where we predicted cleavable
trueIndices = np.where(np.array(test_y) == 1)
#Find index of elements where we predicted nonCleavable
falseIndices = np.where(np.array(test_y) == 0)
#Generate results
print("Real cleavable number: \t",
np.size(trueIndices), "\t Number predicted true cleavable:\t",
np.sum(myRes[0, trueIndices]), "\t Accuracy:\t",
np.sum(myRes[0, trueIndices]) / np.size(trueIndices))
print("Real nonCleavable number:\t", np.size(falseIndices),
"\t Number predicted true nonCleavable:\t",
np.size(falseIndices) - np.sum(myRes[0, falseIndices]),
"\t Accuracy:\t",
(np.size(falseIndices) - np.sum(myRes[0, falseIndices])) /
np.size(falseIndices))
print(
"Total test size:\t\t", len(test_x),
"\t Number predicted true in total:\t",
np.sum(myRes[0, trueIndices]) +
(np.size(falseIndices) - np.sum(myRes[0, falseIndices])),
"\t Accuracy:\t",
((np.sum(myRes[0, trueIndices]) +
(np.size(falseIndices) - np.sum(myRes[0, falseIndices]))) /
len(test_x)))
def part6(root='./Dataset', trainfile='q2_train_set.txt'):
def rotate(angle):
ax.view_init(azim=angle)
csv_path = os.path.join(root, trainfile)
train_x, train_y = bs._load_dataset(csv_path)
no_of_rows, _ = train_x.shape
centroid = np.mean(train_x, axis=0)
std = np.std(train_x, axis=0)
Z = (train_x - centroid) / std
Z_transpose = Z.T
#covariance matrix with up to a constant k
cov_mat_wk = np.matmul(Z_transpose, Z)
eig_values, eig_col_vectors = np.linalg.eig(cov_mat_wk)
idx = eig_values.argsort()[::-1]
eig_values_sorted = eig_values[idx]
eig_col_vectors_sorted = eig_col_vectors[:, idx]
Z_centered = np.matmul(Z, eig_col_vectors_sorted)
PC1 = Z_centered[:, 0]
PC2 = Z_centered[:, 1]
PC3 = Z_centered[:, 2]
PVE = np.sum(eig_values_sorted[0:3]) / np.sum(eig_values)
print("PVE: ", PVE)
#plot
plt.close('all')
fig = plt.figure()
ax = fig.gca(projection='3d')
ax.scatter(PC1, PC2, PC3, c=PC1, linewidth=0.1)
ax.set_xlabel('PC 1')
ax.set_ylabel('PC 2')
ax.set_zlabel('PC 3')
ax.view_init(azim=50)
animation.FuncAnimation(fig,
rotate,
frames=np.arange(0, 365, 1),
interval=0.1)
print("Please close figures to continue...")
plt.show()
def part5(root='./Dataset',
trainfile='q2_train_set.txt',
testfile='q2_test_set.txt'):
noOfAcids = 20
kMers = 8
csv_path = os.path.join(root, trainfile)
train_x, train_y = bs._load_dataset(csv_path)
csv_path = os.path.join(root, testfile)
test_x, test_y = bs._load_dataset(csv_path)
##CALCULATE PROBABILITIES##
N = len(train_x)
#Find number of ones for each feature
cleavable = [
row for index, row in enumerate(train_x) if train_y[index] == 1
]
N11 = np.array(cleavable).sum(axis=0)
N01 = len(cleavable) - N11
#Not Cleavables
notCleavable = [
row for index, row in enumerate(train_x) if train_y[index] == 0
]
N10 = np.array(notCleavable).sum(axis=0)
N00 = len(notCleavable) - N10
N1dot = N10 + N11
N0dot = N00 + N01
Ndot1 = len(cleavable)
Ndot0 = len(notCleavable)
##
sum_term1 = N11 * (np.log2((N * N11) / (N1dot * Ndot1)))
sum_term2 = N01 * (np.log2((N * N01) / (N0dot * Ndot1)))
sum_term3 = N10 * (np.log2((N * N10) / (N1dot * Ndot0)))
sum_term4 = N00 * (np.log2((N * N00) / (N0dot * Ndot0)))
sum_const = 1 / N
I_UC = np.multiply(sum_const,
(sum_term1 + sum_term2 + sum_term3 + sum_term4))
##
I_UC[np.where(np.isnan(I_UC))] = np.Inf
I_UC_sort_indices = np.argsort(I_UC)[::-1]
I_UC_sorted = I_UC[I_UC_sort_indices]
##
trueIndices = np.where(np.array(test_y) == 1)
falseIndices = np.where(np.array(test_y) == 0)
learningParams = np.empty(shape=(train_x.shape[0], 0))
testParams = np.empty(shape=(test_x.shape[0], 0))
accuracies = []
for i in range(1, noOfAcids * kMers):
learningParams = np.hstack(
(learningParams, train_x[:, I_UC_sort_indices[i - 1:i]]))
testParams = np.hstack(
(testParams, test_x[:, I_UC_sort_indices[i - 1:i]]))
res, _, _ = bs._bayes(learningParams, train_y, testParams, kMers,
noOfAcids)
accuracies.append(
((np.sum(res[0, trueIndices]) +
(np.size(falseIndices) - np.sum(res[0, falseIndices]))) /
len(test_x)))
print("Max accuracy:\n",
np.array(accuracies)[np.where(accuracies == np.max(accuracies))[0]])
print("k = ", np.where(accuracies == np.max(accuracies))[0])
#plot
plt.close('all')
plt.plot(range(1, noOfAcids * kMers), accuracies, '-k', linewidth=1)
plt.ylabel('Accuracy (%)')
plt.xlabel('k')
plt.grid(True)
plt.title("k vs. Accuracy")
print("Please close figures to continue...")
plt.show()
