def find_best_smoother(x_tr, y_tr, x_dv, y_dv, smoothers):
"""find the smoothing value that gives the best accuracy on the dev data
:param x_tr: training instances
:param y_tr: training labels
:param x_dv: dev instances
:param y_dv: dev labels
:param smoothers: list of smoothing values to try
:returns: best smoothing value, scores of all smoothing values
:rtype: float, dict
"""
labels = set(
[u'worldnews', u'science', u'askreddit', u'iama', u'todayilearned'])
bestAcc = 0
returnDict = {}
for smoothing in smoothers:
#estimate_nb(x_tr,y_tr,smoothing);
theta_nb = estimate_nb(x_tr, y_tr, smoothing)
#dev_predict = clf_base.predict(x_dv,theta_nb,labels);
#train_predict = clf_base.predict(x_tr,theta_nb,labels);
y_hat = clf_base.predict_all(x_dv, theta_nb, labels)
accuracy = evaluation.acc(y_hat, y_dv)
print "accuracy: ", accuracy
if (accuracy > bestAcc):
bestAcc = accuracy
returnDict[smoothing] = accuracy
return bestAcc, returnDict
def find_best_smoother(x_tr, y_tr, x_dv, y_dv, smoothers):
"""
find the smoothing value that gives the best accuracy on the dev data
:param x_tr: training instances
:param y_tr: training labels
:param x_dv: dev instances
:param y_dv: dev labels
:param smoothers: list of smoothing values
:returns: best smoothing value
:rtype: float
"""
labels = list(set(y_tr))
best_acc = 0
best_smoother = None
scores = {}
for smoother in smoothers:
theta_i = estimate_nb(x_tr, y_tr, smoother)
y_hat = clf_base.predict_all(x_dv, theta_i, labels)
acc = evaluation.acc(y_hat, y_dv)
scores[smoother] = acc
if acc > best_acc:
best_acc = acc
best_smoother = smoother
return best_smoother, scores
def test_lr_d5_3_test():
# NOTE! This test is for the TAs to run
# You cannot pass this test without the true test labels.
# This is a sanity check to make sure your solution for 5.3 is not too crazy
global y_te
y_hat_te = evaluation.read_predictions('lr-best-test.preds')
assert_greater_equal(evaluation.acc(y_hat_te,y_te),.63)
def test_lr_d5_3_test():
# NOTE! This test is for the TAs to run
# You cannot pass this test without the true test labels.
# This is a sanity check to make sure your solution for 5.3 is not too crazy
global y_te
y_hat_te = evaluation.read_predictions('lr-best-test.preds')
assert_greater_equal(evaluation.acc(y_hat_te,y_te),.63)
def test_clf_base_d2_3():
global x_dv, y_dv, y_te, labels
y_hat = clf_base.predict_all(x_dv,hand_weights.theta_hand,labels)
assert_greater_equal(evaluation.acc(y_hat,y_dv),.41)
# just make sure the file is there
y_hat_te = evaluation.read_predictions('hand-test.preds')
eq_(len(y_hat_te),len(y_te))
def test_clf_base_d2_3():
global x_dv, y_dv, y_te, labels
y_hat = clf_base.predict_all(x_dv,hand_weights.theta_hand,labels)
assert_greater_equal(evaluation.acc(y_hat,y_dv),.41)
# just make sure the file is there
y_hat_te = evaluation.read_predictions('hand-test.preds')
eq_(len(y_hat_te),len(y_te))
def test_perc_d4_3_test():
# NOTE! This test is for the TAs to run
# You cannot pass this test without the true test labels.
# This is a sanity check to make sure your solution for 4.3 is not too crazy
global y_te
y_hat_te = evaluation.read_predictions('avp-test.preds')
# i get 66.8% accuracy
assert_greater_equal(evaluation.acc(y_hat_te,y_te),.645)
def test_perc_d4_3_test():
# NOTE! This test is for the TAs to run
# You cannot pass this test without the true test labels.
# This is a sanity check to make sure your solution for 4.3 is not too crazy
global y_te
y_hat_te = evaluation.read_predictions('avp-test.preds')
# i get 66.8% accuracy
assert_greater_equal(evaluation.acc(y_hat_te,y_te),.645)
def train_model(loss,
model,
X_tr_var,
Y_tr_var,
num_its=200,
X_dv_var=None,
Y_dv_var=None,
status_frequency=10,
optim_args={
'lr': 0.002,
'momentum': 0
},
param_file='best.params'):
# initialize optimizer
optimizer = optim.SGD(model.parameters(), **optim_args)
losses = []
accuracies = []
for epoch in range(num_its):
# set gradient to zero
optimizer.zero_grad()
# run model forward to produce loss
output = loss.forward(model.forward(X_tr_var), Y_tr_var)
# backpropagate and train
output.backward()
optimizer.step()
#print(output.item())
losses.append(output.item())
# write parameters if this is the best epoch yet
if X_dv_var is not None:
# run forward on dev data
_, Y_hat = model.forward(X_dv_var).max(dim=1)
# compute dev accuracy
acc = evaluation.acc(Y_hat.data.numpy(), Y_dv_var.data.numpy())
# save
if len(accuracies) == 0 or acc > max(accuracies):
state = {
'state_dict': model.state_dict(),
'epoch': len(accuracies) + 1,
'accuracy': acc
}
torch.save(state, param_file)
accuracies.append(acc)
# print status message if desired
if status_frequency > 0 and epoch % status_frequency == 0:
print("Epoch " + str(epoch + 1) + ": Dev Accuracy: " + str(acc))
# load parameters of best model
checkpoint = torch.load(param_file)
model.load_state_dict(checkpoint['state_dict'])
return model, losses, accuracies
def test_d2_2_predict():
global x_tr_pruned, x_dv_pruned, y_dv
y_hat,scores = clf_base.predict(x_tr_pruned[0],hand_weights.theta_hand,labels)
eq_(scores['pre-1980'],0.1)
assert_almost_equals(scores['2000s'],1.3,places=5)
eq_(y_hat,'2000s')
eq_(scores['1980s'],0.0)
y_hat = clf_base.predict_all(x_dv_pruned,hand_weights.theta_hand,labels)
assert_almost_equals(evaluation.acc(y_hat,y_dv),.3422222, places=5)
def test_avp_d4_3():
global y_dv, x_tr, y_tr
# run on a subset of data
theta_avp,theta_avp_history = perceptron.estimate_avg_perceptron(x_tr[:10],y_tr[:10],3)
assert_almost_equals(theta_avp[('science','what')],3.2258,places=2)
assert_almost_equals(theta_avp[('science','its')],0,places=2)
assert_almost_equals(theta_avp[('worldnews','its')],0.871,places=2)
y_hat_dv = evaluation.read_predictions('avp-dev.preds')
# i get 66.4% accuracy
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.64)
def test_d2_2_predict():
global x_tr_pruned, x_dv_pruned, y_dv
y_hat, scores = clf_base.predict(x_tr_pruned[0], hand_weights.theta_hand,
labels)
eq_(scores['pre-1980'], 0.1)
assert_almost_equals(scores['2000s'], 1.3, places=5)
eq_(y_hat, '2000s')
eq_(scores['1980s'], 0.0)
y_hat = clf_base.predict_all(x_dv_pruned, hand_weights.theta_hand, labels)
assert_almost_equals(evaluation.acc(y_hat, y_dv), .3422222, places=5)
def find_best_smoother(x_tr, y_tr, x_dv, y_dv, smoothers):
"""find the smoothing value that gives the best accuracy on the dev data
"""
scores = {}
labels = set(y_tr)
for s in smoothers:
theta_nb = estimate_nb(x_tr, y_tr, s)
y_hat = clf_base.predict_all(x_dv, theta_nb, labels)
scores[s] = evaluation.acc(y_hat, y_dv)
l = scores.values()
best = smoothers[np.argmax(l)]
return best, scores
def test_lr_d5_2():
global x_tr, y_tr, y_dv, y_te
# run on a subset of data
theta_lr,theta_lr_hist = logreg.estimate_logreg(x_tr[:10],y_tr[:10],3)
assert_almost_equals(theta_lr[('science','what')],.000402,places=4)
assert_almost_equals(theta_lr[('iama', 'missile')],-0.00031832285759249263,places=4)
assert_almost_equals(theta_lr[('iama',constants.OFFSET)],.00045298,places=4)
assert_almost_equals(theta_lr[('askreddit',constants.OFFSET)],0.,places=4)
# dev set accuracy
y_hat_dv = evaluation.read_predictions('lr-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.595)
def test_lr_d5_2():
global x_tr, y_tr, y_dv, y_te
# run on a subset of data
theta_lr,theta_lr_hist = logreg.estimate_logreg(x_tr[:10],y_tr[:10],3)
assert_almost_equals(theta_lr[('science','what')],.000402,places=4)
assert_almost_equals(theta_lr[('iama', 'missile')],-0.00031832285759249263,places=4)
assert_almost_equals(theta_lr[('iama',constants.OFFSET)],.00045298,places=4)
assert_almost_equals(theta_lr[('askreddit',constants.OFFSET)],0.,places=4)
# dev set accuracy
y_hat_dv = evaluation.read_predictions('lr-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.595)
def test_perc_d4_2():
global y_dv, x_tr, y_tr
# run on a subset of data
theta_perc,theta_perc_history = perceptron.estimate_perceptron(x_tr[:10],y_tr[:10],3)
eq_(theta_perc[('worldnews','its')],1)
eq_(theta_perc[('science','its')],0)
eq_(theta_perc[('science','what')],4)
eq_(theta_perc[('worldnews','always')],-1)
eq_(theta_perc_history[0][('science','what')],2)
y_hat_dv = evaluation.read_predictions('perc-dev.preds')
# i get 64.6% accuracy
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.62)
def test_perc_d4_2():
global y_dv, x_tr, y_tr
# run on a subset of data
theta_perc,theta_perc_history = perceptron.estimate_perceptron(x_tr[:10],y_tr[:10],3)
eq_(theta_perc[('worldnews','its')],1)
eq_(theta_perc[('science','its')],0)
eq_(theta_perc[('science','what')],4)
eq_(theta_perc[('worldnews','always')],-1)
eq_(theta_perc_history[0][('science','what')],2)
y_hat_dv = evaluation.read_predictions('perc-dev.preds')
# i get 64.6% accuracy
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.62)
def train_model(loss, model, X_tr_var, Y_tr_var,
num_its = 200,
X_dv_var = None,
Y_dv_var = None,
status_frequency=10,
optim_args = {'lr':0.002,'momentum':0},
param_file = 'best.params'):
# initialize optimizer
optimizer = optim.SGD(model.parameters(), **optim_args)
losses = []
accuracies = []
for epoch in range(num_its):
# set gradient to zero
optimizer.zero_grad()
# run model forward to produce loss
output = loss.forward(model.forward(X_tr_var),Y_tr_var)
# backpropagate and train
output.backward()
optimizer.step()
losses.append(output.data[0])
# write parameters if this is the best epoch yet
if X_dv_var is not None:
# run forward on dev data
_, Y_hat = model.forward(X_dv_var).max(dim=1)
# compute dev accuracy
acc = evaluation.acc(Y_hat.data.numpy(),Y_dv_var.data.numpy())
# save
if len(accuracies) == 0 or acc > max(accuracies):
state = {'state_dict':model.state_dict(),
'epoch':len(accuracies)+1,
'accuracy':acc}
torch.save(state,param_file)
accuracies.append(acc)
# print status message if desired
if status_frequency > 0 and epoch % status_frequency == 0:
print("Epoch "+str(epoch+1)+": Dev Accuracy: "+str(acc))
# load parameters of best model
checkpoint = torch.load(param_file)
model.load_state_dict(checkpoint['state_dict'])
return model, losses, accuracies
def test_avp_d4_3():
global y_dv, x_tr, y_tr
theta_avp,theta_avp_history = perceptron.estimate_avg_perceptron(x_tr[:10],y_tr[:10],3)
# with t=0 initialization
#assert_almost_equals(theta_avp[('science','what')],3.2,places=1)
# with t=1 initialization
assert_almost_equals(theta_avp[('science','what')],3.2258,places=1)
assert_almost_equals(theta_avp[('science','its')],0,places=2)
# with t=0 initialization
#assert_almost_equals(theta_avp[('worldnews','its')],0.866,places=1)
# with t=1 initialization
assert_almost_equals(theta_avp[('worldnews','its')],0.871,places=1)
y_hat_dv = evaluation.read_predictions('avp-dev.preds')
# i get 66.4% accuracy
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.64)
def find_best_smoother(x_tr, y_tr, x_dv, y_dv, smoothers):
'''
find the smoothing value that gives the best accuracy on the dev data
:param x_tr: training instances
:param y_tr: training labels
:param x_dv: dev instances
:param y_dv: dev labels
:param smoothers: list of smoothing values
:returns: best smoothing value
:rtype: float
'''
score = {}
for smoother in smoothers:
theta_nb = estimate_nb(x_tr, y_tr, smoother)
y_hat = clf_base.predict_all(x_dv, theta_nb, set(y_tr))
score[smoother] = (evaluation.acc(y_hat, y_dv))
return clf_base.argmax(score), score
def find_best_smoother(x_tr, y_tr, x_dv, y_dv, smoothers):
"""find the smoothing value that gives the best accuracy on the dev data
:param x_tr: training instances
:param y_tr: training labels
:param x_dv: dev instances
:param y_dv: dev labels
:param smoothers: list of smoothing values to try
:returns: best smoothing value, scores of all smoothing values
:rtype: float, dict
"""
smoother_acc = {}
labels = set(y_dv)
for smoother in smoothers:
theta = estimate_nb(x_tr, y_tr, smoother)
y_hat = clf_base.predict_all(x_dv, theta, labels)
smoother_acc[smoother] = evaluation.acc(y_hat, y_dv)
argmax = lambda x: max(x.iteritems(), key=lambda y: y[1])[0]
return argmax(smoother_acc), smoother_acc
def find_best_smoother(x_tr, y_tr, x_dv, y_dv, smoothers):
'''
find the smoothing value that gives the best accuracy on the dev data
:param x_tr: training instances
:param y_tr: training labels
:param x_dv: dev instances
:param y_dv: dev labels
:param smoothers: list of smoothing values
:returns: best smoothing value
:rtype: float
'''
accuracy = {}
genres = set(y_dv)
for smoother in smoothers:
accuracy[smoother] = evaluation.acc(
clf_base.predict_all(x_dv, estimate_nb(x_tr, y_tr, smoother),
genres), y_dv)
best_smoother = clf_base.argmax(accuracy)
return best_smoother, accuracy
def find_best_smoother(x_tr_pruned, y_tr, x_dv_pruned, y_dv, smoothers):
'''
find the smoothing value that gives the best accuracy on the dev data
:param x_tr: training instances
:param y_tr: training labels
:param x_dv: dev instances
:param y_dv: dev labels
:param smoothers: list of smoothing values
:returns: 1) best smoothing value, 2) a dictionary of smoothing values and dev set accuracy.
:rtype: 1) float, 2) dictionary
'''
smther_dict = {}
labels = set(y_tr)
for x in smoothers:
theta_nb = estimate_nb(x_tr_pruned, y_tr, x)
y_hat = clf_base.predict_all(x_dv_pruned, theta_nb, labels)
smther_dict[x] = evaluation.acc(y_hat, y_dv)
key_min = min(smther_dict.keys(), key=(lambda k: smther_dict[k]))
return smther_dict[key_min], smther_dict
def test_d7_3_bakeoff_dev4():
global Y_dv_var
acc = evaluation.acc(np.load('bakeoff-dev.preds.npy'), Y_dv_var.data.numpy())
assert_greater_equal(acc, 0.55)
def test_feats_d7_1():
global y_dv
y_hat_dv = evaluation.read_predictions('bakeoff-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.78)
def test_feats_d7_1_test():
global y_te
y_hat_te = evaluation.read_predictions('bakeoff-test.preds')
assert_greater_equal(evaluation.acc(y_hat_te,y_te),.722)
def test_lr_d5_3():
global y_dv
y_hat_dv = evaluation.read_predictions('lr-best-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.66)
def test_d5_5_accuracy():
global Y_dv_var
acc = evaluation.acc(np.load('logreg-es-dev.preds.npy'),Y_dv_var.data.numpy())
assert_greater_equal(acc,0.5)
def test_d7_3_bakeoff_dev4():
global Y_dv_var
acc = evaluation.acc(np.load('bakeoff-dev.preds.npy'),Y_dv_var.data.numpy())
assert_greater_equal(acc,0.55)
def train_model(loss, model, X_tr,Y_tr, word_to_ix, tag_to_ix, X_dv=None, Y_dv = None, num_its=50, status_frequency=10,
optim_args = {'lr':0.1,'momentum':0},
param_file = 'best.params'):
#initialize optimizer
optimizer = optim.SGD(model.parameters(), **optim_args)
losses=[]
accuracies=[]
for epoch in range(num_its):
loss_value=0
count1=0
for X,Y in zip(X_tr,Y_tr):
X_tr_var = prepare_sequence(X, word_to_ix)
Y_tr_var = prepare_sequence(Y, tag_to_ix)
# set gradient to zero
optimizer.zero_grad()
lstm_feats= model.forward(X_tr_var)
output = loss(lstm_feats,Y_tr_var)
output.backward()
optimizer.step()
loss_value += output.data[0]
count1+=1
losses.append(loss_value/count1)
# write parameters if this is the best epoch yet
acc=0
if X_dv is not None and Y_dv is not None:
acc=0
count2=0
for Xdv, Ydv in zip(X_dv, Y_dv):
X_dv_var = prepare_sequence(Xdv, word_to_ix)
Y_dv_var = prepare_sequence(Ydv, tag_to_ix)
# run forward on dev data
Y_hat = model.predict(X_dv_var)
Yhat = np.array([tag_to_ix[yhat] for yhat in Y_hat])
Ydv = np.array([tag_to_ix[ydv] for ydv in Ydv])
# compute dev accuracy
acc += (evaluation.acc(Yhat,Ydv))*len(Xdv)
count2 += len(Xdv)
# save
acc/=count2
if len(accuracies) == 0 or acc > max(accuracies):
state = {'state_dict':model.state_dict(),
'epoch':len(accuracies)+1,
'accuracy':acc}
torch.save(state,param_file)
accuracies.append(acc)
# print status message if desired
if status_frequency > 0 and epoch % status_frequency == 0:
print("Epoch "+str(epoch+1)+": Dev Accuracy: "+str(acc))
return model, losses, accuracies
def test_d3_3b_nb():
global y_dv
y_hat_dv = evaluation.read_predictions('nb-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv, y_dv), .46)
def test_lr_d5_3():
global y_dv
y_hat_dv = evaluation.read_predictions('lr-best-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.66)
def test_d4_2b_perc_accuracy():
global y_dv
# i get 43% accuracy
y_hat_dv = evaluation.read_predictions('perc-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv, y_dv), .43)
def test_d3_3b_nb():
global y_dv
y_hat_dv = evaluation.read_predictions('nb-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.46)
def test_feats_d7_1_test():
global y_te
y_hat_te = evaluation.read_predictions('bakeoff-test.preds')
assert_greater_equal(evaluation.acc(y_hat_te,y_te),.722)
def test_d5_5_accuracy():
global Y_dv_var
acc = evaluation.acc(np.load('logreg-es-dev.preds.npy'), Y_dv_var.data.numpy())
assert_greater_equal(acc, 0.5)
def test_d4_2b_perc_accuracy():
global y_dv
# i get 43% accuracy
y_hat_dv = evaluation.read_predictions('perc-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.43)
def train_model(loss,
model,
X_tr,
Y_tr,
word_to_ix,
tag_to_ix,
X_dv=None,
Y_dv=None,
num_its=50,
status_frequency=10,
optim_args={
'lr': 0.1,
'momentum': 0
},
param_file='best.params'):
#initialize optimizer
optimizer = optim.SGD(model.parameters(), **optim_args)
losses = []
accuracies = []
for epoch in range(num_its):
loss_value = 0
count1 = 0
for X, Y in zip(X_tr, Y_tr):
X_tr_var = prepare_sequence(X, word_to_ix)
Y_tr_var = prepare_sequence(Y, tag_to_ix)
# set gradient to zero
optimizer.zero_grad()
lstm_feats = model.forward(X_tr_var)
output = loss(lstm_feats, Y_tr_var)
output.backward()
optimizer.step()
loss_value += output.data[0]
count1 += 1
losses.append(loss_value / count1)
# write parameters if this is the best epoch yet
acc = 0
if X_dv is not None and Y_dv is not None:
acc = 0
count2 = 0
for Xdv, Ydv in zip(X_dv, Y_dv):
X_dv_var = prepare_sequence(Xdv, word_to_ix)
Y_dv_var = prepare_sequence(Ydv, tag_to_ix)
# run forward on dev data
Y_hat = model.predict(X_dv_var)
Yhat = np.array([tag_to_ix[yhat] for yhat in Y_hat])
Ydv = np.array([tag_to_ix[ydv] for ydv in Ydv])
# compute dev accuracy
acc += (evaluation.acc(Yhat, Ydv)) * len(Xdv)
count2 += len(Xdv)
# save
acc /= count2
if len(accuracies) == 0 or acc > max(accuracies):
state = {
'state_dict': model.state_dict(),
'epoch': len(accuracies) + 1,
'accuracy': acc
}
torch.save(state, param_file)
accuracies.append(acc)
# print status message if desired
if status_frequency > 0 and epoch % status_frequency == 0:
print("Epoch " + str(epoch + 1) + ": Dev Accuracy: " + str(acc))
return model, losses, accuracies
def test_feats_d7_1():
global y_dv
y_hat_dv = evaluation.read_predictions('bakeoff-dev.preds')
assert_greater_equal(evaluation.acc(y_hat_dv,y_dv),.78)