def get_c1_score(y_inp, y_pre, msk):
y_predicted = []
y_actual = []
for i in range(len(y_pre)):
if (msk[i // 700][i % 700 + 50] == 1):
y_predicted.append(np.argmax(y_pre[i], 0))
y_actual.append(y_inp[i // 700][i % 700 + 50])
print("F1 score results : \n", c_metric(y_actual, y_predicted))
print("Predicted : \n", c_metric(y_predicted, y_predicted))
def res_on_cv_data():
no_of_batches = len(data_cv) // batch_size
shuffle(data_cv)
correct = 0
y_actual = []
y_predicted = []
for batch_no in range(5):
x = []
y = []
max_length = 0
seq_length = []
for data in data_cv[batch_no * batch_size:batch_no * batch_size +
batch_size]:
seq_length.append(len(data[0]))
max_length = max(max_length, len(data[0]))
x.append(data[0])
y.append(data[1])
x_padded = np.array(
[row + [-1] * (max_length - len(row)) for row in x])
for y_ in y:
y_actual.append(y_)
print("Appending in y_actual", y_)
y = np.array(y)
accuracy_known = model.cross_validate(x_padded, y, seq_length)
y_predicted_ = model.predict(x_padded, y, seq_length)
for y_ in y_predicted_:
print("Appending in y_preicted", y_)
y_predicted.append(y_)
correct += accuracy_known * batch_size
accuracy = (correct * 100) / (batch_size * no_of_batches)
print("Accuracy on train data : ", accuracy)
print(c_metric(y_actual, y_predicted))
def res_on_cv_data(model, data_cv):
no_of_batches = len(data_cv.keys())
y_predicted = []
y_actual = []
correct = 0
total_samples = 0
for batch_no in range(no_of_batches - 1, -1, -1):
print("Iteration number, batch number : ", epoch, batch_no)
data_batch = data_cv[batch_no]
batch_size = len(data_batch[1])
x = data_batch[0]
y = data_batch[1]
freq = data_batch[2]
seq_length = data_batch[3]
x = np.array(x)
y_actual.extend(y)
y = np.array(y)
freq = np.array(freq)
accuracy_kmown = model.cross_validate(x, y, seq_length, freq)
y_predicted_ = model.predict(x, y, seq_length, freq)
y_predicted.extend(y_predicted_)
correct += accuracy_kmown * batch_size
total_samples += batch_size
accuracy = (correct * 100) / (total_samples)
correct_1 = len([i for i, j in zip(y_actual, y_predicted) if i == j])
print("Accuracy on cv data : ", accuracy, correct, correct_1)
print("Lengths of y_actual and y_predicted : ", len(y_actual),
len(y_predicted))
print(c_metric(y_actual, y_predicted))
def res_on_cv_data():
no_of_batches = len(data_cv) // batch_size
no_of_batches = 3
shuffle(data_cv)
zero_100_list = [[0] * 100]
correct = 0
y_actual = []
y_predicted = []
for batch_no in range(no_of_batches):
if (batch_no % 5 == 0):
print("res_on_cv_data : ", batch_no)
x = []
y = []
max_length = 0
seq_length = []
for data in data_cv[batch_no * batch_size:batch_no * batch_size +
batch_size]:
seq_length.append(len(data[0]))
max_length = max(max_length, len(data[0]))
x.append(data[0])
y.append(data[1])
x_n = [row + (zero_100_list) * (max_length - len(row)) for row in x]
x_padded = np.array(x_n)
y_actual.extend(y)
y = np.array(y)
accuracy_known = model.cross_validate(x_padded, y, seq_length)
y_predicted_ = model.predict(x_padded, y, seq_length)
y_predicted.extend(y_predicted_)
correct += accuracy_known * batch_size
accuracy = (correct * 100) / (batch_size * no_of_batches)
correct_1 = len([i for i, j in zip(y_actual, y_predicted)])
print("Accuracy on train data : ", accuracy, correct, correct_1)
print(c_metric(y_actual, y_predicted))
