def knn():
train_data, train_labels = load_train()
#for validation
valid_data, valid_labels = load_valid()
#for test
#valid_data, valid_labels = load_test()
values = [1, 3, 5, 7, 9]
ratio = []
for k in values:
c = 0
prediction_labels = run_knn(k, train_data, train_labels, valid_data)
for i in range(len(valid_labels)):
if valid_labels[i] == prediction_labels[i]:
c += 1
ratio.append(float(c) / len(prediction_labels))
plt.plot(values, ratio)
#for validation
plt.axis([1, 9, 0.81, 0.87])
#for test
#plt.axis([1, 9, 0.87, 0.95])
plt.show()
def knn():
train_data, train_labels = load_train()
#for validation
valid_data, valid_labels = load_valid()
#for test
#valid_data, valid_labels = load_test()
values = [1, 3, 5, 7, 9]
ratio = []
for k in values:
c = 0
prediction_labels = run_knn(k, train_data, train_labels, valid_data)
for i in range(len(valid_labels)):
if valid_labels[i] == prediction_labels[i]:
c += 1
ratio.append(float(c) / len(prediction_labels))
plt.plot(values, ratio)
#for validation
plt.axis([1, 9, 0.81, 0.87])
#for test
#plt.axis([1, 9, 0.87, 0.95])
plt.show()
def run_logistic_regression():
train_inputs, train_targets = load_train()
valid_inputs, valid_targets = load_valid()
# TODO: initialize parameters
parameters = {
'learning_rate': 0.01 ,
'weight_regularization': 0 ,
'num_iterations': 10
}
# logistic regression weights
dimension = 28*28
z = np.ones([dimension+1, 1], int)
z = z/100.0
#weight = np.matrix('1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1')
for i in xrange(0,28*28):
if i%2 == 1:
z[i] = 0
weights = z
#weights = 1,1,2,1
# Verify that your logistic function produces the right gradient.
# diff should be very close to 0.
#run_check_grad(parameters)
# Begin learning with gradient descent
for t in xrange(parameters['num_iterations']):
# TODO: you will need to modify this loop to create plots, etc.
# find the negative log likelihood and derivatives w.r.t. weights
f, df, frac_correct_train = logistic(weights,
train_inputs,
train_targets,
parameters)
_, _, frac_correct_valid = logistic(weights,
valid_inputs,
valid_targets,
parameters)
if np.isnan(f) or np.isinf(f):
raise ValueError("nan/inf error")
# update parameters
for i in range(weights.shape[0]):
weights[i] = weights[i] + parameters['learning_rate'] * (df[i] - 0.001*(weights[i]))
# print some stats
print ("ITERATION:{:4d} LOGL:{:4.2f} "
"TRAIN FRAC:{:2.2f} VALID FRAC:{:2.2f}").format(t+1,
f,
frac_correct_train*100,
frac_correct_valid*100)
def run_logistic_regression(hyperparameters):
# TODO specify training data
train_inputs, train_targets = load_train()
valid_inputs, valid_targets = load_valid()
# N is number of examples; M is the number of features per example.
N, M = train_inputs.shape
# Logistic regression weights
# TODO:Initialize to random weights here.
#weights = np.random.normal(0, 0.2, (train_inputs.shape[1]+1,1))
weights = np.zeros(785).reshape((785, 1))
# Verify that your logistic function produces the right gradient.
# diff should be very close to 0.
run_check_grad(hyperparameters)
# Begin learning with gradient descent
logging = np.zeros((hyperparameters['num_iterations'], 5))
for t in xrange(hyperparameters['num_iterations']):
# Find the negative log likelihood and its derivatives w.r.t. the weights.
f, df, predictions = logistic(weights, train_inputs, train_targets,
hyperparameters)
# Evaluate the prediction.
cross_entropy_train, frac_correct_train = evaluate(
train_targets, predictions)
if np.isnan(f) or np.isinf(f):
raise ValueError("nan/inf error")
# update parameters
weights = weights - hyperparameters['learning_rate'] * df / N
# Make a prediction on the valid_inputs.
predictions_valid = logistic_predict(weights, valid_inputs)
# Evaluate the prediction.
cross_entropy_valid, frac_correct_valid = evaluate(
valid_targets, predictions_valid)
# print some stats
print(
"ITERATION:{:4d} TRAIN NLOGL:{:4.2f} TRAIN CE:{:.6f} "
"TRAIN FRAC:{:2.2f} VALID CE:{:.6f} VALID FRAC:{:2.2f}").format(
t + 1, f / N, cross_entropy_train, frac_correct_train * 100,
cross_entropy_valid, frac_correct_valid * 100)
logging[t] = [
f / N, cross_entropy_train, frac_correct_train * 100,
cross_entropy_valid, frac_correct_valid * 100
]
return logging
dist = l2_distance(valid_data.T, train_data.T)
nearest = np.argsort(dist, axis=1)[:,:k]
train_labels = train_labels.reshape(-1)
valid_labels = train_labels[nearest]
# note this only works for binary labels
valid_labels = (np.mean(valid_labels, axis=1) >= 0.5).astype(np.int)
valid_labels = valid_labels.reshape(-1,1)
return valid_labels
if __name__ == '__main__':
train_inputs, train_targets = utils.load_train()
valid_inputs, valid_targets = utils.load_valid()
test_inputs, test_targets = utils.load_test()
set_k = [1,3,5,7,9]
accuracy_valid_output = {}
accuracy_test_output = {}
length_valid = len(valid_inputs)
length_test = len(test_inputs)
for k in set_k:
valid_outputs = run_knn(k, train_inputs, train_targets, valid_inputs)
test_outputs = run_knn(k, train_inputs, train_targets, test_inputs)
count_valid = np.sum(valid_outputs == valid_targets)
def run_logistic_regression(hyperparameters):
# specify training data
xIn = False
while xIn == False:
x = raw_input('Training Set LARGE or SMALL? ')
print(x)
if x == 'LARGE':
print("HELLO")
train_inputs, train_targets = load_train()
xIn = True
elif x == 'SMALL':
print("hello")
train_inputs, train_targets = load_train_small()
xIn = True
else:
print("Please input LARGE or SMALL")
valid_inputs, valid_targets = load_valid()
test_inputs, test_targets = load_test()
# N is number of examples; M is the number of features per example.
N, M = train_inputs.shape
print("N:", N, " M:", M)
# Logistic regression weights
# Initialize to random weights here.
weights = np.random.normal(0, 0.001, (M+1, 1))
# Verify that your logistic function produces the right gradient.
# diff should be very close to 0.
run_check_grad(hyperparameters)
# Begin learning with gradient descent
logging = np.zeros((hyperparameters['num_iterations'], 5))
for t in xrange(hyperparameters['num_iterations']):
# Find the negative log likelihood and its derivatives w.r.t. the weights.
f, df, predictions = logistic(weights, train_inputs, train_targets, hyperparameters)
# Evaluate the prediction.
cross_entropy_train, frac_correct_train = evaluate(train_targets, predictions)
if np.isnan(f) or np.isinf(f):
raise ValueError("nan/inf error")
# update parameters
weights = weights - hyperparameters['learning_rate'] * df / N
# Make a prediction on the valid_inputs.
predictions_valid = logistic_predict(weights, valid_inputs)
# Evaluate the prediction.
cross_entropy_valid, frac_correct_valid = evaluate(valid_targets, predictions_valid)
# print some stats
print ("ITERATION:{:4d} TRAIN NLOGL:{:4.2f} TRAIN CE:{:.6f} "
"TRAIN FRAC:{:2.2f} VALID CE:{:.6f} VALID FRAC:{:2.2f}").format(
t+1, f / N, cross_entropy_train, frac_correct_train*100,
cross_entropy_valid, frac_correct_valid*100)
logging[t] = [f / N, cross_entropy_train, frac_correct_train*100, cross_entropy_valid, frac_correct_valid*100]
return logging
# -*- coding: utf-8 -*-
from utils import load_train, load_valid
from run_knn import run_knn
(train_inputs, train_targets) = load_train()
(valid_inputs, valid_targets) = load_valid()
for k in [1, 3, 5, 7, 9]:
print run_knn(k, train_inputs, train_targets, valid_inputs)
Created on Feb 26 2017
Author: Weiping Song
"""
import os, sys
import tensorflow as tf
import numpy as np
import argparse, random
from model import GRU4Rec
from utils import load_train, load_valid
unfold_max = 20
error_during_training = False
train_x, train_y, n_items = load_train(unfold_max)
valid_x, valid_y, _ = load_valid(unfold_max)
class Args():
is_training = True
layers = 1
rnn_size = 100
n_epochs = 10
batch_size = 50
keep_prob = 1
learning_rate = 0.001
decay = 0.98
decay_steps = 2 * 1e3
sigma = 0.0001
init_as_normal = False
grad_cap = 0
import utils
from run_knn import run_knn
import plot_digits
import numpy as np
import matplotlib.pyplot as plt
if __name__ == "__main__":
#loading the dataset
train_data, train_labels = utils.load_train()
#loading the validation set
valid_data,valid_labels = utils.load_valid()
# vector of each k
K = np.array([1,3,5,7,9])
#dictionnay result
results={}
for k in K:
#prediction
prediction = run_knn(k,train_data,train_labels,valid_data)
#computing the precision
results[k]= np.mean(prediction==valid_labels)
dist = l2_distance(valid_data.T, train_data.T)
nearest = np.argsort(dist, axis=1)[:, :k]
train_labels = train_labels.reshape(-1)
valid_labels = train_labels[nearest]
# note this only works for binary labels
valid_labels = (np.mean(valid_labels, axis=1) >= 0.5).astype(np.int)
valid_labels = valid_labels.reshape(-1, 1)
return valid_labels
if (__name__ == "__main__"):
train_x, train_y = utils.load_train()
valid_x, valid_y = utils.load_valid()
k = [1, 3, 5, 7, 9]
y1 = []
for i in k:
knn = run_knn(i, train_x, train_y, valid_x)
# print(knn)
# print(valid_y)
# print(valid_y == knn)
result = (np.sum(np.bitwise_xor(knn, valid_y)) /
float(valid_x.shape[0]))
# result = np.mean(valid_y == knn)
y1.append(result)
print("Error Rate " + str(result))
test_x, test_y = utils.load_test()
y2 = []
for j in k:
def train_RCNN(args):
train_x, train_y, n_items = load_train(args.max_len)
args.n_items = n_items
data = list(zip(train_x, train_y))
random.shuffle(data)
train_x, train_y = zip(*data)
num_batches = len(train_x) // args.batch_size
global valid_x
global valid_y
valid_x, valid_y, _ = load_valid(args.max_len)
print('#Items: {}'.format(n_items))
print('#Training Nums: {}'.format(len(train_x)))
gpu_config = tf.ConfigProto()
gpu_config.gpu_options.allow_growth = True
with tf.Session(config=gpu_config) as sess:
model = RCNN(args)
if args.is_store:
saver = tf.train.Saver(tf.global_variables())
ckpt = tf.train.get_checkpoint_state(args.checkpoint_dir)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
print('Restore model from {} successfully!'.format(
ckpt.model_checkpoint_path))
else:
print('Restore model from {} failed!'.format(
args.checkpoint_dir))
return
else:
sess.run(tf.global_variables_initializer())
best_epoch = -1
best_step = -1
best_loss = np.inf
best_HR = np.inf
max_stay, stay_cnt = 20, 0
losses = 0.0
for epoch in range(args.epochs):
for i in range(num_batches):
x = train_x[i * args.batch_size:(i + 1) * args.batch_size]
y = train_y[i * args.batch_size:(i + 1) * args.batch_size]
fetches = [
model.sum_loss, model.global_step, model.lr, model.train_op
]
feed_dict = {model.X: x, model.Y: y}
loss, step, lr, _ = sess.run(fetches, feed_dict)
losses += loss
if step % 50 == 0:
print('Epoch-{}\tstep-{}\tlr:{:.6f}\tloss: {:.6f}'.format(
epoch + 1, step, lr, losses / 50))
losses = 0.0
if step % 1000 == 0:
valid_loss, HR, NDCG, MRR = eval_validation(
model, sess, args.batch_size)
print(
'step-{}\teval_validation\tloss:{:6f}\tHR@{}:{:.6f}\tNDCG@{}:{:.6f}\tMRR@{}:{:.6f}'
.format(step, valid_loss, args.top_k, HR, args.top_k,
NDCG, args.top_k, MRR))
if HR > best_HR or (valid_loss < best_loss and HR > 0.0):
best_HR = HR
best_loss = valid_loss
best_epoch = epoch + 1
best_step = step
stay_cnt = 0
ckpt_path = args.checkpoint_dir + 'model.ckpt'
model.saver.save(sess, ckpt_path, global_step=step)
print("model saved to {}".format(ckpt_path))
else:
stay_cnt += 1
if stay_cnt >= max_stay:
break
if stay_cnt >= max_stay:
break
print("best model at:epoch-{}\tstep-{}\tloss:{:.6f}\tHR@{}:{:.6f}".
format(best_epoch, best_step, best_loss, args.top_k, best_HR))
from l2_distance import l2_distance
import numpy as np
from run_knn import run_knn
import matplotlib.pyplot as plt
"""
CSC 2515 - Assignment 1
Tausif Sharif
Notes:
- Runs the run_knn.py functions from here
- Will show and save relevant plots
"""
trainInputs, trainTargets = load_train()
smallInputs, smallTargets = load_train_small()
validInputs, validTargets = load_valid()
testInputs, testTargets = load_test()
kList = [1, 3, 5, 7, 9]
classRates = range(0, len(kList))
classRatesT = range(0, len(kList))
listCount = 0
for k in kList:
correctCount = 0
validLables = run_knn(k, trainInputs, trainTargets, validInputs)
for i in xrange(len(validLables)):
if validLables[i] == validTargets[i]:
correctCount += 1
classRates[listCount] = (correctCount / float(len(validLables)))
listCount += 1