def train_classifier(train_data, train_labels, dev_data, dev_labels, seed):
learning_rate = 0.005
classifier = ml.Classifier(dim=4, lr=learning_rate, seed=seed)
last_acc = None
for i in range(100): # number of epochs
for data, label in zip(train_data, train_labels):
classifier.learn(data, label)
current_acc = classifier.calculate_acc(dev_data, dev_labels)
print(current_acc)
if last_acc and last_acc > current_acc - 0.001:
break
last_acc = current_acc
return classifier, current_acc
def train_classifier(train, dev, seed):
learning_rate = 0.005
classifier = ml.Classifier(dim=4, lr=learning_rate, seed=seed)
last_acc = None
for i in range(100): # number of epochs
for example in train.get_data():
classifier.learn(example.get_data(), example.get_label())
current_acc = classifier.calculate_acc(dev.get_data_batch(),
dev.get_label_batch())
if last_acc and last_acc > current_acc - 0.001:
break
last_acc = current_acc
return classifier, current_acc
if datapoint[3] not in ['Alaska', 'Wyoming']: # not legally allowed to perform experiment in these states
if datapoint[4] == "smoking":
smoking_label = 1
else:
smoking_label = 0
dev_data.append((datapoint[0], int(datapoint[1]), datapoint[2], datapoint[3]))
dev_labels.append(smoking_label)
## Whew, we're finished loading the data. Let's train a classifier five times with different random seeds.
## We use the learn function to update our parameters
accs = []
for i in range(5):
print("Seed", i)
learning_rate = 0.005
classifier = ml.Classifier(dim=4, lr=learning_rate, seed=i)
last_acc = None
for i in range(100): # number of epochs
for j in range(len(train_data)):
classifier.learn(train_data[j], train_labels[j])
current_acc = classifier.calculate_acc(dev_data, dev_labels)
print(current_acc)
if last_acc and last_acc > current_acc - 0.001:
break
last_acc = current_acc
accs.append(current_acc)
## Finally, let's print out our predictions on the dev set:
dev_predictions = classifier.predict_batch(dev_data)
num_printed = 0