def train_with_sketch(self, feature_pos, features, label):
logit = 0
min_logit = float("inf")
max_logit = float("-inf")
for i in range(len(feature_pos)):
val = self.top_k.get_value_for_key(feature_pos[i]) * features[i]
if val > max_logit:
max_logit = val
if val < min_logit:
min_logit = val
# calculating wTx
logit += val
if max_logit - min_logit == 0:
max_logit = 1
min_logit = 0
normalized_weights = (logit - min_logit) / (max_logit - min_logit)
sigm_val = self.sigmoid(normalized_weights)
gradient = (label - sigm_val)
loss = self.loss(y=label, p=sigm_val)
self.loss_val += loss
if gradient != 0:
for i in range(len(feature_pos)):
updated_val = self.learning_rate * gradient * features[i]
value = self.cms.update(feature_pos[i], updated_val)
self.top_k.push(Node(feature_pos[i], value))
return loss
def train_with_sketch(self, feature_pos, features, label):
logit = 0
min_logit = float("inf")
max_logit = float("-inf")
for i in range(len(feature_pos)):
# print("top k at pos {} value {}".format(feature_pos[i], self.top_k.get_item(feature_pos[i])))
val = self.top_k.get_value_for_key(feature_pos[i]) * features[i]
if val > max_logit:
max_logit = val
if val < min_logit:
min_logit = val
logit += val
if max_logit - min_logit == 0:
max_logit = 1
min_logit = 0
normalized_weights = (logit - min_logit) / (max_logit - min_logit)
print("normalized weights {}".format(normalized_weights))
sigm_val = self.sigmoid(normalized_weights)
if sigm_val == 1.0:
sigm_val = sigm_val - (1e-5)
print("label {} sigmoid {}".format(label, sigm_val))
loss = self.loss(y=label, p=sigm_val)
gradient = (label - sigm_val)
if gradient != 0:
for i in range(len(feature_pos)):
# updating the change only on previous values
updated_val = self.learning_rate * gradient * features[i]
value = self.cms.update(feature_pos[i], updated_val)
self.top_k.push(Node(feature_pos[i], value))
return loss
def train_batch(self, examples_batch):
gradient = 0
loss = 0
feature_values = [0] * self.features
for example_index in examples_batch:
example, label = self.samples[example_index], self.true_labels[
example_index]
logit = 0
for i in range(len(example)):
val = self.top_k.get_value_for_key(i + 1) * example[i]
feature_values[i] += example[i]
logit += val
sigm_val = self.sigmoid(logit)
loss = self.loss(y=label, p=sigm_val)
diff_label = (label - sigm_val) # difference in label
if diff_label != 0:
for i in range(len(example)):
# updating the change only on previous values
if example[i] != 0:
grad_update = self.learning_rate * diff_label * example[
i]
if i + 1 in self.top_k_dict.keys():
self.top_k_dict[i + 1].append(grad_update)
value = self.cms.update(i, grad_update)
# self.recovered_weight[i] = value
self.top_k.push(Node(i + 1, value))
return loss
def train(self, example, label):
val = 0
for i in range(len(example)):
val += self.top_k.get_value_for_key(i) * example[i]
sigm_val = self.sigmoid(val)
# print("label {} sigmoid {}".format(label, sigm_val))
loss = self.loss(y=label, p=sigm_val)
diff_label = (label - sigm_val) # difference in label
if diff_label != 0:
for i in range(len(example)):
# updating the change only on previous values
grad_update = self.learning_rate * diff_label * example[i]
self.recovered_weight[i] += grad_update
self.top_k.push(Node(i, self.recovered_weight[i]))
return loss
def train_with_sketch(self, feature_pos, features, label):
logit = 0
for i in range(len(feature_pos)):
val = self.top_k.get_value_for_key(feature_pos[i]) * features[i]
# calculating wTx
logit += val
# print("label {} wx {}".format(label, logit))
gradient = (label - logit)
print("loss {}".format(gradient))
if gradient != 0:
for i in range(len(feature_pos)):
updated_val = 2 * self.learning_rate * gradient * features[i]
value = self.cms.update(feature_pos[i], updated_val)
self.top_k.push(Node(feature_pos[i], value))
return gradient
def train(self, feature_pos, features, label):
val = 0
for i in range(len(feature_pos)):
val += self.top_k.get_value_for_key(feature_pos[i]) * features[i]
sigm_val = self.sigmoid(val)
print("label {} sigmoid {}".format(label, sigm_val))
loss = self.loss(y=label, p=sigm_val)
diff_label = (label - sigm_val) # difference in label
if diff_label != 0:
for i in range(len(feature_pos)):
# updating the change only on previous values
grad_update = self.learning_rate * diff_label * features[i]
self.gradient_updates_dict[feature_pos[i]].append(grad_update)
self.gradients[feature_pos[i]] += grad_update
self.top_k.push(
Node(feature_pos[i], self.gradients[feature_pos[i]]))
return loss
def train_with_sketch(self, feature_pos, features, label):
logit = 0
for i in range(len(feature_pos)):
# print("top k at pos {} value {}".format(feature_pos[i], self.top_k.get_item(feature_pos[i])))
val = self.top_k.get_value_for_key(feature_pos[i]) * features[i]
logit += val
sigm_val = self.sigmoid(logit)
loss = self.loss(y=label, p=sigm_val)
diff_label = (label - sigm_val) # difference in label
if diff_label != 0:
for i in range(len(feature_pos)):
# updating the change only on previous values
grad_update = self.learning_rate * diff_label * features[i]
if feature_pos[i] in self.top_k_dict.keys():
self.top_k_dict[feature_pos[i]].append(grad_update)
value = self.cms.update(feature_pos[i], grad_update)
self.top_k.push(Node(feature_pos[i], value))
return loss
def train_with_sketch(self, example, label):
logit = 0
for i in range(len(example)):
val = self.top_k.get_value_for_key(i + 1) * example[i]
logit += val
sigm_val = self.sigmoid(logit)
loss = self.loss(y=label, p=sigm_val)
diff_label = (label - sigm_val) # difference in label
if diff_label != 0:
for i in range(len(example)):
# updating the change only on previous values
if example[i] != 0:
grad_update = self.learning_rate * diff_label * example[i]
if i + 1 in self.top_k_dict.keys():
self.top_k_dict[i + 1].append(grad_update)
value = self.cms.update(i, grad_update)
# self.recovered_weight[i] = value
self.top_k.push(Node(i + 1, value))
return loss
def store_in_topk(self, non_zero_index, non_zero_values):
for i in range(len(non_zero_index)):
self.top_k.push(Node(non_zero_index[i] + 1, non_zero_values[i]))