def pegasos_sw(X_train, y_train, lambda_reg=1, max_it=1000, tol=1e-4):
W = Counter()
s = 1
t = 1
epoch = 1
objective = 1e5
objective2 = 10
m = len(y_train)
while abs(objective - objective2) > tol and epoch <= max_it:
objective2 = objective
objective = 0
for j in range(m):
t = t + 1
step = 1 / (t * lambda_reg)
review = X_train[j]
result = y_train[j]
scale = -(step * lambda_reg)
cond = result * s * util.dotProduct(W, review)
if cond < 1:
s = (1 + scale) * s
util.increment(W, step * result / s, review)
else:
s = (1 + scale) * s
objective += max(0, 1 - cond)
objective = objective / m
objective = objective + lambda_reg / 2 * (s**2) * util.dotProduct(W, W)
epoch += 1
return s, W
def loss(x,y,l,w):
loss = (l*dotProduct(w,w))/2
m = len(x)
for i in range(m):
loss = loss + (max(0, 1- y[i]*dotProduct(w,x[i])))/m
return loss
def pegasos(X_train, y_train, lambda_reg=1, max_it=1000, tol=1e-6):
w = Counter()
t = 1
epoch = 1
objective = 1e5
objective2 = 10
m = len(y_train)
while abs(objective - objective2) > tol and epoch <= max_it:
objective2 = objective
objective = 0
for j in range(m):
t = t + 1
step = 1 / (t * lambda_reg)
review = X_train[j]
result = y_train[j]
scale = -(step * lambda_reg)
cond = result * util.dotProduct(w, review)
if cond < 1:
util.increment(w, scale, w)
util.increment(w, step * result, review)
else:
util.increment(w, scale, w)
objective += max(0, 1 - cond)
objective = objective / m
objective = objective + lambda_reg / 2 * util.dotProduct(w, w)
epoch += 1
return w
def getIntensity(self, pos):
"""Returns the appropriate intensity of the sound being played assuming intensity falls off at 1/r^2"""
#Camera doesnt have position so im just using the position of the followed object (of 1st camera)
camPos = glad.renderer.cameraList[0].objectFollowed.getPos()
r=(pos-camPos)#separation vector
if r.isNullVector(): #if the vector is null, sound will be max anyways
sin = 1
cos = 1
else:
#calculate angles to determine where sound is coming from
cos = dotProduct(r.getNormalized(),Vector(-1,0))
sin = dotProduct(r.getNormalized(), Vector(0,1))
#Calculate intensity for left and right channels
#when sound is directly to the side have 80 percent come from that side speaker
#hopefully this will give some directional sounds
k = 130000 #arbitrary constant to calculate sound intensity
if r.isNullVector():
intensity = k #removes division by zero error
else:
intensity = k/r.getMagnitude()**2
#major is the percent of the sound intensity from the side with the greater intensity
a=0.68 #max percent of the intensity coming from one side
major = (a*0.5)/((0.5*cos)**2+(a*sin)**2)**0.5 #equation for an ellipse
if r[0] <= 0:
right = major
left = 1-major
else:
left = major
right = 1-major
right *= intensity
left *= intensity
if right > 1: right = 1
if left > 1: left = 1
return left,right
def kmeanspredictor(x):
assignment = 0
min_dist = 1000000
for j in range(NUM_CLUSTERS):
cur_dist = util.dotProduct(x, x) - 2 * util.dotProduct(
centroids[j], x) + pre_computed_centroid_dots[j]
if cur_dist < min_dist:
assignment = j
min_dist = cur_dist
return centroid_vals[assignment]
def predictor(x):
centroid_ind = 0
minDist = float('inf')
for k in range(len(centroids)):
cur_dist = util.dotProduct(x, x) - 2 * util.dotProduct(
centroids[k], x) + pre_computed_centroid_dots[k]
min_dist = float('inf')
if cur_dist < min_dist:
assignment = k
min_dist = cur_dist
return predictor_list[i](x)
def predictor(x):
if x == None:
return -1
if util.dotProduct(featureExtractor(x), weights) > 0:
return 1
else:
return 0
def pegasos_fast(x, y, l):
w = dict()
temp_w = dict()
t = 2
s = 1
temp_loss = 0
flag = True
while flag:
for j in range(len(x)):
t = t + 1
n = 1/(l*t)
s = (1-n*l)*s
if y[j]*(dotProduct(w, x[j])) < s:
temp = x[j].copy()
increment(temp, (n*y[j]-1), temp)
increment(w,(1/s), temp)
temp_w = w.copy()
increment(temp_w, s-1, temp_w)
loss_real = loss(x,y,l,temp_w)
if abs(temp_loss - loss_real) < 10**-2:
flag = False
temp_loss = loss_real
increment(w, s-1, w)
return w
def lassoLossGradient(features, weights, true_value, tuning_parameter):
"""Computes the value of the training loss gradient (with respect to the
weight vector) at a specific example.
Training loss includes a lasso (L1) regularization term.
Args:
features (dict): A sparse vector of feature values.
weights (dict): A sparse vector of feature weights.
true_value (int): The true value of an example.
tuning_parameter (double): Coefficient of the lasso regularization term.
Returns:
A sparse vector (dict) representing the gradient value.
"""
# Standard squared loss
gradient = {}
scale = 2 * (dotProduct(features, weights) - true_value)
# Lasso term: add gradient of the lasso term to the scaling factor (i.e.
# add gradient of |tuning_parameter| * (1-norm of weights)
weight_signs = [np.sign(weights[w]) for w in weights]
for w in weights:
gradient[w] = tuning_parameter * np.sign(weights[w])
increment(gradient, scale, features)
return gradient
def findExampleStatsFn(examples, weights, featureExtractor, examineFn):
summ = 0
correct = 0
tot = 0
summNeg = 0
correctNeg = 0
totNeg = 0
for example in examples:
prompt, response = example[0]
if examineFn(prompt, response):
phi = featureExtractor(example[0])
score = dotProduct(weights, phi)
if example[1] == 1:
summ += score
if score > 0:
correct += 1
tot += 1
if example[1] == -1:
summNeg += score
if score < 0:
correctNeg += 1
totNeg += 1
if tot > 0:
print "Average Score (+): {0}".format(1.0*summ/tot)
print "Average Correct (+): {0}".format(1.0*correct/tot)
if totNeg > 0:
print "Average Score (-): {0}".format(1.0*summNeg/totNeg)
print "Average Correct (-): {0}".format(1.0*correctNeg/totNeg)
def SparseGradChecker(loss_func,
gradient_loss_func,
x,
y_val,
theta,
epsilon=0.01,
tolerance=1e-4):
"""Question 3.2: Implement Generic Gradient Checker for Sparse Matrices.
Check that the function gradient_loss_func returns the correct gradient for
the given x, y_val, and theta.
Let d be the number of features. Here we numerically estimate the
gradient by approximating the directional derivative in each of
the d coordinate directions:
(e_1 = (1,0,0,...,0), e_2 = (0,1,0,...,0), ..., e_d = (0,...,0,1)
The approximation for the directional derivative of J at the point
theta in the direction e_i is given by:
( J(theta + epsilon * e_i) - J(theta - epsilon * e_i) ) / (2*epsilon).
We then look at the Euclidean distance between the gradient
computed using this approximation and the gradient computed by
gradient_loss_func(x, y_val, theta). If the Euclidean
distance exceeds tolerance, we say the gradient is incorrect.
Args:
loss_func - A function that computes the loss for (x, y_val, theta).
gradient_loss_func - A function that computes gradient for (x, y_val, theta).
x - A single row in the design matrix, represented by a dict/Counter object. (key length = num_features)
y_val - the label for the corresponding x_row (-1 or 1)
theta - the parameter vector, dict/Counter object. (key length = num_features)
epsilon - the epsilon used in approximation
tolerance - the tolerance error
Return:
A boolean value indicate whether the gradient is correct or not
"""
true_gradient = gradient_loss_func(x, y_val, theta)
approx_grad = dict.fromkeys(theta.keys(), 0.0)
for key in theta.iterkeys():
# Compute the approximate directional derivative in the chosen direction
# Avoid copying since it's so slow.
theta_key_original = theta[key]
theta[key] += epsilon
plus_loss = loss_func(x, y_val, theta)
theta[key] = theta_key_original - epsilon
minus_loss = loss_func(x, y_val, theta)
theta[key] = theta_key_original # restore theta
approx_grad[key] = (plus_loss - minus_loss) / (2 * epsilon)
util.increment(approx_grad, -1,
true_gradient) # approx_grad - true_gradient
error = math.sqrt(util.dotProduct(
approx_grad,
approx_grad)) # np.linalg.norm(approx_grad - true_gradient)
if error > tolerance:
print 'gradient doesn\'t match approximation. Error:', error
return (error < tolerance)
def regularizationLossGradient(features, weights, true_value,
tuning_parameter):
"""Computes the value of the training loss gradient (with respect to the
weight vector) at a specific example.
Training loss includes a ridge (L2) regularization term.
Args:
features (dict): A sparse vector of feature values.
weights (dict): A sparse vector of feature weights.
true_value (int): The true value of an example.
tuning_parameter (double): Coefficient of the ridge regularization term.
Returns:
A sparse vector (dict) representing the gradient value.
"""
# Standard squared loss
gradient = {}
scale = 2 * (dotProduct(features, weights) - true_value)
# Regularization term: add gradient of the regularization term to the
# scaling factor (i.e. add gradient of |tuning_parameter| *
# (2-norm of weights)^2
increment(gradient, tuning_parameter, weights)
increment(gradient, scale, features)
return gradient
def pegasos_grad(X,y,w,lamb):
tmp = y*dotProduct(w,X)
if 1-tmp > 0:
an1 = increment({},lamb,w)
ans = increment(an1,y,X)
else:
ans = increment({},lamb,w)
return ans
def PegasosSubgradientLoss(x, y_val, theta, lambda_reg):
'''Question 3.2: The Subgradient of the Pegasos Loss function.'''
margin = y_val * util.dotProduct(theta, x)
subgrad = theta.copy()
util.scale(subgrad, lambda_reg)
if margin < 1:
util.increment(subgrad, -y_val, x)
return subgrad
def PercentageWrong(X, y, theta):
'''Question 4.3: The percentage incorrect when using theta to predict y from X.'''
num_wrong = 0
for i, x in enumerate(X):
estimate_sign = np.sign(util.dotProduct(theta, x))
if estimate_sign != y[i]:
num_wrong += 1
return 1.0 * num_wrong / len(y)
def per_loss(x,y,w):
cnt = 0
total = len(y)
for i in range(total):
if np.sign(dotProduct(w, x[i])) != np.sign(y[i]):
cnt = cnt + 1
error = (cnt/total)*100.0
return error
def learnBoostedRegression(examples, num_iters, step_size, num_trees):
"""Learns a linear regression model using boosted trees.
Args:
examples: An array of training examples.
num_iters (int): Number of training iterations.
step_size (int): Stochastic gradient descent step size.
num_trees (int): Number of gradient boosting trees.
Returns:
A predictor function that outputs a price (int) given a single input
tuple.
"""
list_weights = []
objectives = [cur[1] for cur in examples]
filename = "boostedtree_" + str(num_trees -
1) + "_" + str(cross_val_seg) + ".p"
if num_trees > 1 and SAVE:
(list_weights, num_trees_prev, num_iters_prev) = pickle.load(
open(os.path.join("boostedtree_weights", filename), "rb"))
for k in range(num_trees):
if k >= len(list_weights):
print ""
print "TREE " + str(k + 1) + " OF " + str(num_trees)
curWeights = defaultdict(int)
for i in range(num_iters):
for ind in range(len(examples)):
x = examples[ind][0]
gradient = regression.lassoLossGradient(
x, curWeights, objectives[ind], .5)
increment(curWeights, -step_size / (i + 1), gradient)
if VERBOSE:
print "Training progress: " + str(
100.0 * (i + 1) / num_iters) + "%"
list_weights.append(curWeights)
else:
curWeights = list_weights[k]
for j in range(len(examples)):
x, y = examples[j]
objectives[j] = objectives[j] - dotProduct(x, curWeights)
if VERBOSE: print "COMPLETE"
if SAVE:
filename = "boostedtree_" + str(num_trees) + "_" + str(
cross_val_seg) + ".p"
pickle.dump((list_weights, num_trees, num_iters),
open(os.path.join("boostedtree_weights", filename), "wb"))
# Define the predictor function
def predictor(x):
return sum(dotProduct(x, curWeight) for curWeight in list_weights)
return predictor
def trainAndTest():
# Import the training and test data as numpy arrays
train_array = util.csvAsArray('data/train_updated.csv')
test_array = util.csvAsArray('data/test.csv')
# Generate a list of (feature vector, value) tuples for the training data
feature_names = util.getCsvHeaders('data/train_updated.csv')
train_examples = []
k_examples = []
for i in range(len(train_array)):
feature_count = range(len(train_array[i]) - 1)
feature_values = [train_array[i][j] for j in feature_count]
feature_vector = featurize(feature_values, feature_names)
output = train_array[i][len(train_array[0]) - 1]
train_examples.append((feature_vector, output))
k_examples.append(feature_vector)
# Train a k-means model on the training data and evaluate its mean
# squared error with the test data
random.shuffle(train_examples)
for i in range(0, NUM_SPLITS, 2):
startTest = i * len(train_examples) / NUM_SPLITS
endTest = (i + 1) * len(train_examples) / NUM_SPLITS
currentTrainExamples = train_examples[0:startTest] + train_examples[
endTest:len(train_examples)]
(centroids, assign, loss, loss_list,
centroid_vals) = kmeans(currentTrainExamples, NUM_CLUSTERS, 500)
currentBoostedExamples = [(currentTrainExamples[ind][0],
loss_list[ind])
for ind in range(len(currentTrainExamples))]
boostedRegPredictor = learnBoostedRegression(currentBoostedExamples, 500, \
0.00000000001, num_trees=NUM_B_TREES)
pre_computed_centroid_dots = [
util.dotProduct(centroids[ind], centroids[ind])
for ind in range(NUM_CLUSTERS)
]
def kmeanspredictor(x):
assignment = 0
min_dist = 1000000
for j in range(NUM_CLUSTERS):
cur_dist = util.dotProduct(x, x) - 2 * util.dotProduct(
centroids[j], x) + pre_computed_centroid_dots[j]
if cur_dist < min_dist:
assignment = j
min_dist = cur_dist
return centroid_vals[assignment]
def boostedKPredictor(x):
return kmeanspredictor(x) + boostedRegPredictor(x)
print "leaving out the", (
i + 1
), "th segment of the data, the validation error for the regression is:", util.evaluatePredictor(
boostedKPredictor, train_examples[startTest:endTest])
def test3c1():
weights = {}
for _ in range(100):
k = ''.join(random.choice(string.ascii_lowercase) for _ in range(5))
v = random.uniform(-1, 1)
weights[k] = v
data = submission.generateDataset(100, weights)
for phi, y in data:
grader.require_is_equal(util.dotProduct(phi, weights) >= 0, y == 1)
def gradLoss(phiX, w, y):
score = util.dotProduct(w, phiX)
margin = score * y
if margin < 1:
for name, feature in phiX.iteritems():
phiX[name] = -1 * y * feature
return phiX
else:
return 0
def PlotScoresAgainstAccuracy(X_training, y_training, X_testing, y_testing,
lambda_reg):
'''Question 4.5.
Divides the training set into buckets by score, and creates a bar chart showing the accuracy of
each bucket.
'''
NUM_BUCKETS = 10
theta = Pegasos(X_training, y_training, lambda_reg)
# Calculate the score for each row in a list
scores = [util.dotProduct(theta, x) for x in X_testing]
low_score = min(scores)
high_score = max(scores)
# f(score) -> bucket
score_to_bucket_func = lambda score: int(
round((NUM_BUCKETS - 1) * (score - low_score) /
(high_score - low_score)))
# Make a list of empty lists with NUM_BUCKETS elements
# Each entry is a list of the indexes of X's rows that fall in the same score bucket.
score_histogram = [[] for _ in range(NUM_BUCKETS)]
for row_index, score in enumerate(scores):
bucket = score_to_bucket_func(score)
score_histogram[bucket].append(row_index)
bucket_means = [0.0] * NUM_BUCKETS
bucket_accuracy = [0.0] * NUM_BUCKETS
for bucket, row_indices in enumerate(score_histogram):
# calculate the percentage wrong loss for each bucket
# make a scatter plot of these
bucket_scores = [scores[row_index] for row_index in row_indices]
bucket_score_mean = abs(np.mean(bucket_scores))
bucket_means[bucket] = bucket_score_mean
bucket_score_std = np.std(bucket_scores)
# print 'Bucket', bucket, 'ranges from', min(bucket_scores), 'to', max(bucket_scores)
# print 'Bucket', bucket, 'mean:', bucket_score_mean
# print 'Bucket', bucket, 'stdev:', bucket_score_std
X_bucket = [X_testing[row_index] for row_index in row_indices]
y_bucket = [y_testing[row_index] for row_index in row_indices]
bucket_accuracy[bucket] = 100 * (
1.0 - PercentageWrong(X_bucket, y_bucket, theta))
fig, ax = plt.subplots()
ax.set_xlabel('Mean Score for Bucket')
ax.set_ylabel('Percentage Correct')
ax.set_title('Pegasos Sentiment Analysis: Score vs. Accuracy')
width = 0.4
positions = range(0, len(bucket_accuracy))
rects1 = ax.bar(positions, bucket_accuracy, width, color='b', alpha=0.8)
plt.xticks(rotation=-45)
ax.set_xticks([pos + width for pos in positions])
ax.set_xticklabels(["%0.1f" % mean for mean in bucket_means])
plt.show()
def find_center(ex_index, example, precomputed_x, precomputed_quantities,
centroids):
assign = 0
min_dist = 1,000
for i in range(K):
cur_dist = precomputed_x[ex_index] - 2 * util.dotProduct(
centroids[i], example) + precomputed_quantities[i]
if cur_dist < min_dist:
assign = i
min_dist = cur_dist
def getAnswerProbs(weights, questionData): proposedAnswers = questionData["proposedAnswers"] correctIndex = questionData["correctAnswerIndex"] answerScores = [] for aIndex, proposed in enumerate(proposedAnswers): score = dotProduct(weights, featureExtractor(proposed)) answerScores.append(score) return softmax(answerScores)
def Pegasos(X, y, lambda_reg, max_epochs=1000, check_gradient=False):
'''Question 4.2.
Finds the sparse weight vector that minimizes the SVM loss function on X and y.
'''
print 'Running Pegasos with regularization parameter', lambda_reg
loss_func = lambda x, y_val, theta: PegasosLoss(x, y_val, theta, lambda_reg
)
gradient_loss_func = lambda x, y_val, theta: PegasosSubgradientLoss(
x, y_val, theta, lambda_reg)
# Initialize theta to have zero for every word mentioned in any review
theta = {key: 0.0 for x in X for key in x.keys()}
t = 2 # NOTE: This normally starts at zero, but that causes a divide-by-zero error.
weight_scalar = 1.0
for epoch in range(max_epochs):
# print '--Epoch', epoch
old_theta = theta.copy()
for j, x in enumerate(X):
t += 1
eta = 1.0 / (t * lambda_reg)
margin = y[j] * weight_scalar * util.dotProduct(theta, x)
# NOTE that the gradient is not differentiable at 1.0, so we don't check it near there.
if check_gradient and abs(margin - 1.0) > 0.01:
if SparseGradChecker(loss_func, gradient_loss_func, x, y[j],
theta):
print 'Computed gradient doesn\'t match approximations.'
sys.exit(1)
grad = gradient_loss_func(x, y[j], theta)
util.increment(theta, -eta, grad)
else:
weight_scalar *= 1.0 - 1.0 / t
if margin < 1:
util.increment(theta, eta * y[j] / weight_scalar, x)
util.increment(old_theta, -1, theta)
util.scale(old_theta, weight_scalar)
total_change = math.sqrt(util.dotProduct(old_theta, old_theta))
# print '----Change from previous theta:', total_change
if total_change < 0.01:
break
util.scale(theta, weight_scalar)
return theta
def learn(self, trainExamples):
numIters = 10
step = 0.0001
for i in range(numIters):
for feature_vec, y in trainExamples:
score = util.dotProduct(self.weights, feature_vec)
dloss = {}
if score*y > 1:
continue
else:
util.increment(dloss, -y, feature_vec)
util.increment(self.weights, -step, dloss)
def percent_error(X, y, w):
correct = 0
pos = 0
total = len(y)
for i in range(total):
sign_value = np.sign(dotProduct(X[i], w))
if y[i] == sign_value:
correct += 1
if sign_value > 0:
pos += 1
print pos
return 1-float(correct)/total
def chooseEval(examples, weights):
correct = 0
for i in range(len(examples)):
prompt = examples[i][0][0]
response1 = examples[i][0][1]
randomInt = random.randint(0, len(examples)-1)
response2 = examples[randomInt][0][1]
# Relies on random to break loop
while response1 == response2 or (neg_restrict_bad and isBadTurn(response2)) or response1[0].caller == response2[0].caller:
randomInt = random.randint(0, len(examples)-1)
response2 = examples[randomInt][0][1]
guess1 = (prompt, response1)
phi1 = swda_feature_extractor(guess1)
score1 = dotProduct(weights, phi1)
guess2 = (prompt, response2)
phi2 = swda_feature_extractor(guess2)
score2 = dotProduct(weights, phi2)
if(score1 > score2):
correct = correct + 1
if(score1 == score2):
correct = correct + .5
return 1.0 * correct / len(examples)
def test3c0():
ans = 0.05
grader.require_is_equal(
ans,
util.dotProduct({
'Movie': 0.1,
'is': 0.2,
'good': 0.25
}, {
'Movie': 0.1,
'is': 0.2,
'very': -0.25,
'bad': -0.25
}))
def predict(self, stories):
if self.args.verbose > 3:
print 'Predicting'
if self.args.verbose:
print "weights are ", self.weights
ANS_LETTERS = ['A', 'B', 'C', 'D']
ans = []
for story in stories:
if self.args.verbose > 0:
print story.name
print formatForPrint(story.rawPassage), "\n"
print story.rawQuestions, "\n"
print story.rawAnswers, "\n"
for qid in range(len(story.questions)):
scores = []
for aid in range(len(story.answers[qid])):
score = max([(util.dotProduct(self.weights, self.extractFeatures(story, sid, qid, aid)),sid) for \
sid in range(len(story.passageSentences))])
scores.append((score, aid))
# if question contains "n't | not", and begin
# with "what, who, whose", select the minium score.
s = story.rawQuestions[qid][1].strip()
if re.search('^(who|what|whose).*(n\'t|not)',
s,
flags=re.IGNORECASE):
answer = min(scores)[1]
else:
answer = max(scores)[1]
ans.append(ANS_LETTERS[answer])
if self.args.verbose > 0:
if answer != story.correctAnswers[qid]:
print 'WRONG: %s: correct answer %s, predicted answer %s, scores %s' \
%(story.rawQuestions[qid][0], ANS_LETTERS[story.correctAnswers[qid]], ANS_LETTERS[answer], scores)
else:
print 'RIGHT: %s: correct answer %s, predicted answer %s, scores %s' \
%(story.rawQuestions[qid][0], ANS_LETTERS[story.correctAnswers[qid]], ANS_LETTERS[answer], scores)
if self.args.verbose > 0:
print "\n"
return ans
def pegasos_SGD(X,y,lamb,num_iter):
w = {}
t = 1
s = 1
for i in range(num_iter):
for j in range(len(X)):
t += 1
alpha = 1.0/(t*lamb)
tmp = y[j] * s * dotProduct(X[j], w)
g = l_de(tmp)
s *= (1 - alpha * lamb)
w = increment(w, -(alpha*y[j]*g/s), X[j])
print "epoch "+str(i)
return increment({},s,w)
def show_incorrect_case():
train_data,test_data = read_files()
X_train, y_train = data_label(train_data)
X_test, y_test = data_label(test_data)
lamb = 0.1
w_o= pickle.load(open("weight",'rb'))
test_len = len(X_test)
example = 0
for i in range(test_len):
w = w_o
tmp = dotProduct(X_test[i],w)
if (np.sign(tmp)!=y_test[i]):
example += 1
print "predicted_score: ",np.sign(tmp)
print "true vote: ",y_test[i]
dict_tmp = dotProduct_vector(X_test[i],w)
sorted_dict = sorted(dict_tmp.items(),key=lambda x: abs(x[1]),reverse=True)
print_list_wx = sorted_dict[:8]
print_list_abs_wx = [(a,abs(b)) for a,b in print_list_wx]
print_list_x = [X_test[i][x] for (x,k) in print_list_wx ]
print_list_w = [w[x] for (x,k) in print_list_wx ]
print "wx:"
print print_list_wx
print "\n"
print "abs_wx"
print print_list_abs_wx
print "\n"
print "x"
print print_list_x
print "\n"
print "w"
print print_list_w
if example == 3:
break
def score_interval():
train_data,test_data = read_files()
X_train, y_train = data_label(train_data)
X_test, y_test = data_label(test_data)
lamb = 1e-1
w = pegasos_SGD(X_train,y_train,lamb,30)
a = []
test_len = len(X_test)
for i in range(test_len):
a.append(dotProduct(X_test[i],w))
a.sort()
print min(a)
print max(a)
plt.plot(a)
plt.show()
def score_confidence():
train_data,test_data = read_files()
X_train, y_train = data_label(train_data)
X_test, y_test = data_label(test_data)
lamb = 1e-1
# w = pegasos_SGD(X_train,y_train,lamb,30)
#
# pickle.dump(w,open("weight",'wb'))
w = pickle.load(open("weight",'rb'))
a = [0]*18 #correct count
b = [0]*18 #incorrect count
c = [0]*18 #total number
test_len = len(X_test)
for i in range(test_len):
tmp = dotProduct(X_test[i],w)
c[int(tmp+9)] += 1
if (np.sign(tmp)==y_test[i]):
a[int(tmp+9)] += 1
else:
b[int(tmp+9)] += 1
# for j in range(18):
# if b[j]==0:
# print "interval [%s,%s] has %s points and %s are correct, the ratio is %s "%(j-9,j-8,b[j],a[j],0)
# else:
# print "interval [%s,%s] has %s points and %s are correct, the ratio is %4.2f "%(j-9,j-8,b[j],a[j],float(a[j])/b[j])
wide = 0.35
p1 = plt.bar(np.arange(18),a,width=wide, color='g')
p2 = plt.bar(np.arange(18),b,width=wide, color='r', bottom=a)
plt.ylabel("Frequency")
plt.xlabel("Score Intervals")
# plt.title("Score Confidence")
X_ticks = ["[%s,%s]"%(k-9,k-8) for k in np.arange(18)]
plt.xticks(np.arange(18)+wide/2,X_ticks,rotation=45)
plt.legend((p1[0],p2[0]),("Correct","Incorrect"))
plt.savefig("t45_score_confidence.png")
def guessEval(examples, weights):
correct = 0
for i in range(len(examples)):
prompt = examples[i][0][0]
maxScore = 0
maxResponse = examples[0][0][1]
for j in range(len(examples)):
response = examples[j][0][1]
if prompt[0].caller == response[0].caller:
continue
guess = (prompt, response)
phi = swda_feature_extractor(guess)
score = dotProduct(weights, phi)
if score > maxScore:
maxScore = score
maxResponse = response
if maxResponse == examples[i][0][1]:
correct = correct + 1
return 1.0 * correct / len(examples)
def fit(self, X):
self.w_ = dict()
t = 0
for j in range(len(X)):
t += 1
step_size = 1 / (t * self.lambda_reg)
w_dot_x = util.dotProduct(self.w_, X[j])
y = 1 if 1 in X[j] else -1
if y * w_dot_x < 1:
util.increment(self.w_, (1 - 1 / t), self.w_)
util.increment(self.w_, step_size * y, X[j])
else:
util.increment(self.w_, (1 - 1 / t), self.w_)
return self.w_
def expectimax_value(self, game, action_list, depth=1):
#Do all of the actions in action_list
for action in action_list:
game = game.players[self.turn_num].do_move(game, action)
turn_num = (self.turn_num+1) % 4
total_action_list = [] #Stores all of the actions made so that you can undo them
while depth > 0:
opp = game.players[turn_num]
opp_action_list = self.guess_opp_move(opp, game)
#Add opponent actions with the opponent object so we can undo them
if opp_action_list:
for opp_action in opp_action_list:
game = opp.do_move(game, opp_action)
total_action_list.append((opp.turn_num, opp_action))
turn_num = (turn_num+1) % 4
if turn_num == self.turn_num: depth -= 1
#Find value of your estimated future state
expected_features = self.feature_extractor(game)
expected_score = util.dotProduct(expected_features, self.weights)
for i in range(len(game.players)):
player = game.players[i]
print "cities and settlements: ", player.turn_num, player.cities_and_settlements
print "total actions list: ", total_action_list
#Undo moves the player made
for i in range(len(total_action_list)-1, -1, -1):
to_undo = total_action_list[i]
opp_num, opp_action = to_undo
print "Loop cities and settles: ",opp_num, game.players[opp_num].cities_and_settlements
game = game.players[opp_num].undo_move(game, opp_action)
#Undo moves the player made
for i in range(len(action_list)-1, -1, -1):
game = game.players[self.turn_num].undo_move(game, action_list[i])
return expected_score
def pegasos(x, y, l):
w = dict()
t = 2
temp_loss = 0
cnt = 0
flag = True
for i in range(2):
for j in range(len(x)):
t = t + 1
n = 1/(l*t)
if y[j]*(dotProduct(w, x[j])) < 1:
cnt = cnt +1
temp = x[j].copy()
increment(temp, (n*y[j]-1), temp)
increment(w,-n*l,w)
increment(w,1,temp)
else:
increment(w,-n*l,w)
return w
def updateWeights(self, game):
# print game
#Get current score
cur_features = self.feature_extractor(game)
target = util.dotProduct(cur_features, self.weights)
pred = self.prevScore
# print "f**k" , cur_features["Player "+ str(1) + " Settlements"]
#If there are no previous features you can't update
if self.prevFeatures:
#Update weights
diff = pred - target
for feature, val in self.prevFeatures.items():
# print diff, val
self.weights[feature] -= self.eta * diff * val
self.prevFeatures = cur_features
self.prevScore = target
def predict(weights, testSet, args):
correct = 0
incorrect = 0
total = 0
for data in testSet:
data = json.loads(data)
title = data['title']
subreddit = data['subreddit']
features = FeatureExtractor.extractFeatures(title, args)
maxScore = float('-inf')
prediction = ''
for key in weights.keys():
weightVector = weights[key]
score = util.dotProduct(weightVector, features)
if score > maxScore:
prediction = key
maxScore = score
if prediction == subreddit:
correct += 1
else:
if args.verbose:
try:
print title
print "predicted: " + prediction.encode('utf-8')
print features
printRelevantWeights(weights, features)
print "-----------------"
except UnicodeEncodeError:
print "error"
incorrect += 1
total += 1
print 'accuracy ' + str(float(correct) / total)
print 'wrong ' + str(float(incorrect) / total)
def learnPredictor(trainExamples, testExamples, featureExtractor):
'''
Given |trainExamples| and |testExamples| (each one is a list of (x,y)
pairs), a |featureExtractor| to apply to x, and the number of iterations to
train |numIters|, return the weight vector (sparse feature vector) learned.
You should implement stochastic gradient descent.
Note: call evaluatePredictor() on both trainExamples and testExamples
to see how you're doing as you learn after each iteration.
'''
weights = {} # feature => weight
stepSize = 1
numIters = 15
for it in range(0, numIters):
# iterate through every training example and extract the features of x
for x, y in trainExamples:
phi = featureExtractor(x)
# print phi
# if y * score < 1 (wrong prediction) then calculate gradient loss then update weight for each feature
margin = y*util.dotProduct(weights, phi)
if (1-margin) > 0:
indicator = 1
else:
indicator = 0
scale = stepSize*indicator*y
increment(weights, scale, phi)
# this uses the defined feature extractor to predict the classification of x
def predictor(x):
phi = featureExtractor(x)
# create thresholds for different scores
score = dotProduct(phi, weights)
# return 1 if (dotProduct(phi, weights) > 0) else -1
# Print out training and test error for every iteration:
# print 'TRAINING ERROR:', util.evaluatePredictor(trainExamples, predictor)
# print 'TEST ERROR:', util.evaluatePredictor(testExamples, predictor)
return weights
def main():
path = ""
image_name = ""
if len(sys.argv) > 1:
path = sys.argv[1]
if not os.path.exists(path):
print "The path provided does not exist."
return
directories = path.split('/')
file_name = directories[len(directories) - 1]
else:
print "Please supply a path to an image."
return
os.system("python segment.py " + path)
segments = []
for f in os.listdir(SEGMENTS_PATH):
if 'temp' in f and image_name in f: # Ways to identify segments of the given path
segments.append(os.path.join(SEGMENTS_PATH, f))
f = open('weights.out') # Read in weights
weights = eval(f.readline())
stop_sign_flag = False
for segment in segments:
score = util.dotProduct(weights,
seg_util.segmentFeatureExtractor(segment))
if score >= 0: # Stop sign found
stop_sign_flag = True
break
if stop_sign_flag:
print "Stop sign detected!"
else:
print "No stop sign detected"
def ErrorAnalysis(X_training, y_training, X_testing, y_testing, lambda_reg):
'''Question 5.1.
Prints information about the top incorrect reviews, ordered by the magnitude of their score.
'''
theta = Pegasos(X_training, y_training, lambda_reg)
scores = [util.dotProduct(theta, x) for x in X_testing]
# (index, score) pairs, sorted by the score's absolute value in descending order.
score_indexes = sorted(
enumerate(scores),
reverse=True,
key=lambda index_score_pair: abs(index_score_pair[1]))
num_incorrect_examples = 0
MAX_NUM_WRONG_EXAMPLES = 10
# Print out the information about all the incorrect examples, in order of largest score.
for row_index, score in score_indexes:
if num_incorrect_examples >= MAX_NUM_WRONG_EXAMPLES:
break
y_testing_val = y_testing[row_index]
if np.sign(score) != np.sign(y_testing_val):
x_testing_row = X_testing[row_index]
PrintReviewInfo(x_testing_row, y_testing_val, score, theta)
num_incorrect_examples += 1
def main():
path = ""
image_name = ""
if len(sys.argv) > 1:
path = sys.argv[1]
if not os.path.exists(path):
print "The path provided does not exist."
return
directories = path.split("/")
file_name = directories[len(directories) - 1]
else:
print "Please supply a path to an image."
return
os.system("python segment.py " + path)
segments = []
for f in os.listdir(SEGMENTS_PATH):
if "temp" in f and image_name in f: # Ways to identify segments of the given path
segments.append(os.path.join(SEGMENTS_PATH, f))
f = open("weights.out") # Read in weights
weights = eval(f.readline())
stop_sign_flag = False
for segment in segments:
score = util.dotProduct(weights, seg_util.segmentFeatureExtractor(segment))
if score >= 0: # Stop sign found
stop_sign_flag = True
break
if stop_sign_flag:
print "Stop sign detected!"
else:
print "No stop sign detected"
def hingeLoss(w, features, y): return max(0, 1 - dotProduct(w, features) * y)
def predictor(x):
return dotProduct(x, weights)
def trainAndTest():
"""Defines K-means clustering and perform clustered regression.
"""
# Import the training and test data as numpy arrays
train_array = util.csvAsArray('data/train_updated.csv')
test_array = util.csvAsArray('data/test.csv')
# Generate a list of (feature vector, value) tuples for the training data
feature_names = util.getCsvHeaders('data/train_updated.csv')
train_examples = []
k_examples = []
for i in range(len(train_array)):
feature_count = range(len(train_array[i]) - 1)
feature_values = [train_array[i][j] for j in feature_count]
feature_vector = util.featurize(feature_values, feature_names)
output = train_array[i][len(train_array[0]) - 1]
train_examples.append((feature_vector, output))
random.shuffle(train_examples)
for i in range(1, NUM_SPLITS, 2):
startTest = i * len(train_examples) / NUM_SPLITS
endTest = (i + 1) * len(train_examples) / NUM_SPLITS
currentTrain = train_examples[0:startTest] + train_examples[
endTest:len(train_examples)]
currentTest = train_examples[startTest:endTest]
# Cluster the data using k-means
(centroids, assign, loss, loss_list,
centroid_vals) = kmeans.kmeans(currentTrain, NUM_CENTROIDS, K_ITERS)
# Make clusters
cluster_list = [[] for _ in range(len(centroids))]
for j in range(len(currentTrain)):
cluster_list[assign[i]].append(currentTrain[j])
# Train a regression model on the training data (by cluster)
# and evaluate its mean squared error with the train data
regression_error = 0
predictor_list = []
pre_computed_centroid_dots = [
util.dotProduct(centroids[k], centroids[k])
for k in range(len(centroids))
]
for cluster_points in cluster_list:
boostedRegressionPredictor = boostedtree.learnBoostedRegression(
cluster_points, SGD_ITERS, ETA, 5, 0)
predictor_list.append(boostedRegressionPredictor)
def predictor(x):
centroid_ind = 0
minDist = float('inf')
for k in range(len(centroids)):
cur_dist = util.dotProduct(x, x) - 2 * util.dotProduct(
centroids[k], x) + pre_computed_centroid_dots[k]
min_dist = float('inf')
if cur_dist < min_dist:
assignment = k
min_dist = cur_dist
return predictor_list[i](x)
regression_error = boostedtree.evaluatePredictor(
predictor, currentTest)
#regression_error /= len(train_examples)
# Print the results
print ""
print "------------------"
print "CLUSTERED REGRESSION WITH BOOSTING"
print "------------------"
print "Leaving out segment: " + str(i)
print "Number of centroids: " + str(10)
print "Number of examples: " + str(len(train_examples))
print "Regression MSE: " + str(regression_error)
print ""
return predictor_list, centroids, regression_error
def test3c0():
weights = {"hello": 1, "world": 1}
data = submission.generateDataset(5, weights)
for datapt in data:
grader.requireIsEqual((util.dotProduct(datapt[0], weights) >= 0),
(datapt[1] == 1))
def test2c():
weights = {'hello': 1, 'world': 1}
data = submission.generateDataset(5, weights)
for datapt in data:
grader.requireIsEqual(util.dotProduct(datapt[0], weights) >= 0,
datapt[1] == 1)
def test3c_0():
weights = {"hello": 1, "world": 1}
data = submission.generateDataset(5, weights)
for datapt in data:
grader.requireIsEqual((util.dotProduct(datapt[0], weights) >= 0), (datapt[1] == 1))
#Compare data
# threeCards = {}
# for _ in range(0,3):
# thisChoice = random.choice(pCard.keys())
# threeCards[thisChoice] = util.extract(thisChoice,pCard)
# modelValue = util.dotProduct(weights,threeCards[thisChoice])
# oracleValue =
correctCount = 0
totalCount = 0
for key,value in testData.iteritems():
totalCount += 1
compareResults = []
myChoice = (0,0)
trueChoice = (0,0)
for i in range(0,3):
if value[i] > trueChoice[1]: trueChoice = (key[i],value[i])
thisValue = []
modelValue = util.dotProduct(weights,util.extract(key[i],pCard))
if modelValue > myChoice[1]: myChoice = (key[i],modelValue)
thisValue.append((key[i],value[i],modelValue))
compareResults.append(thisValue)
if trueChoice[0] == myChoice[0]:
compareResults.append((trueChoice,True))
correctCount += 1
else: compareResults.append(False)
print compareResults
print correctCount/float(totalCount)
def getTDScore(self, features):
return util.dotProduct(self.weights_, features)
def pegasos_loss(X,y,w,lamb):
ans = (lamb/2.0)*dotProduct(w,w)+max(0,1-y*dotProduct(w,X))
return ans
def predict(self, x):
print "Learned Score:" + str(util.dotProduct(self.weights, x))
return math.copysign(1.0, util.dotProduct(self.weights, x))
def main():
#loading the shuffled data
with open('data.pickle', 'rb') as f:
review = pickle.load(f)
#Splitting into training and test sets
train, test = split(review)
#Splitting x and y values and getting reasy for training
x_train = []
x_test = []
y_train = []
y_test = []
for i in train:
y_train.append(i.pop())
x_train.append(bag_of_words(i))
for i in test:
y_test.append(i.pop())
x_test.append(bag_of_words(i))
l = 0.5
print("Pegasos fast")
start_time = time.time()
w1 = pegasos_fast(x_train, y_train, l)
time1 = time.time() - start_time
print("--- %s seconds ---" % (time1) )
#print(len(w1))
#print("percentage error:", per_loss(x_test,y_test,w1))
#error analysis
pos = []
count = 0
for i in range(len(y_test)):
if np.sign(dotProduct(w1, x_test[i])) != np.sign(y_test[i]) and count<2:
count = count +1
pos.append(i)
wrong1 = x_test[pos[0]].copy()
new_wrong1 = wrong1.copy()
abs_wrong1 = wrong1.copy()
print("wrong 1, real:", y_test[pos[0]])
wrong2 = x_test[pos[1]].copy()
new_wrong2 = wrong2.copy()
abs_wrong2 = wrong2.copy()
print("wrong 2, real:", y_test[pos[1]])
#multiplying weight
for i in wrong2:
wt = w1.get(i,0)
abs_wrong2[i] = abs(abs_wrong2[i]*wt)
wrong2[i] = wrong2[i]*wt
keys = sorted(abs_wrong2)
absvals = sorted(abs_wrong2.values())
#print("name:"," absolute product"," product", " x"," w")
for i in range(len(keys)):
print(keys[-i],",", abs_wrong2[keys[-i]],",", wrong2[keys[-i]],",", new_wrong2[keys[-i]],",", w1.get(keys[-i],0),",")
'''
def loss(w, phi, y):
return max(1 - util.dotProduct(w, phi) * y, 0)
def printExamples(examples, weights, featureExtractor, SCORE_THRESHOLD):
# random.jumpahead(1)
random.shuffle(examples)
# SCORE_THRESHOLD = .5
print "Finding Interesting Examples..."
tpFound = fpFound = tnFound = fnFound = False
for example in examples:
prompt, response = example[0]
if not examineCollab(prompt, response) : continue
phi = featureExtractor(example[0])
score = dotProduct(weights, phi)
if not tpFound and score > SCORE_THRESHOLD and example[1] == 1:
print "FOUND: True Positive"
print "Prompt"
for utt in example[0][0]:
print utt.text_words();
print "Response"
for utt in example[0][1]:
print utt.text_words();
for key in phi:
if key in weights:
print "{0}: {1}".format(key, weights[key])
tpFound = True
if not fpFound and score > SCORE_THRESHOLD and example[1] == -1:
print "FOUND: False Positive"
print "Prompt"
for utt in example[0][0]:
print utt.text_words();
print "Response"
for utt in example[0][1]:
print utt.text_words();
for key in phi:
if key in weights:
print "{0}: {1}".format(key, weights[key])
fpFound = True
if not tnFound and score < -SCORE_THRESHOLD and example[1] == -1:
print "FOUND: True Negative"
print "Prompt"
for utt in example[0][0]:
print utt.text_words();
print "Response"
for utt in example[0][1]:
print utt.text_words();
for key in phi:
if key in weights:
print "{0}: {1}".format(key, weights[key])
tnFound = True
if not fnFound and score < -SCORE_THRESHOLD and example[1] == 1:
print "FOUND: False Negative"
print "Prompt"
for utt in example[0][0]:
print utt.text_words();
print "Response"
for utt in example[0][1]:
print utt.text_words();
for key in phi:
if key in weights:
print "{0}: {1}".format(key, weights[key])
fnFound = True
if tpFound and fpFound and tnFound and fnFound:
break
