ind1 = np.random.choice(np.arange(y.size), 5000)
x = x[ind1,:]
y = y[ind1]
#
xtest = x_test[indices2,:]
ytest = t_test[indices2]
ytest = np.cast[int](ytest)
ytest[ytest==4]=-1
ytest[ytest==9]=1
ind2 = np.random.choice(np.arange(ytest.size),500)
xtest = xtest[ind2,:]
ytest = ytest[ind2]
'''
import mnist_subset
x, y, xtest, ytest = mnist_subset.init(4, 9, 2500, 500)
#### Useing data from class here
#mnist_49_3000 = sio.loadmat('mnist_49_3000.mat')
#X = mnist_49_3000['x']
#Y = mnist_49_3000['y']
#X = X.T
#Y = Y[0,:]
#
#numofx =1000
#x = X[:numofx,:] # X= x+xtest
#y = Y[:numofx]
#xtest = X[numofx:, :]
#ytest = Y[numofx:]
numoftrain = 10
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
#Author: Saibal De
#Class: EECS545 Machine Learning
#Title: LP with Random Queries
#Date: 12-12-2018
import numpy as np
import mnist_subset
import mnist_graph
from lp import LP
x_train, y_train, _, _ = mnist_subset.init(4, 9, 2500, 0)
graph = mnist_graph.init(x_train)
y_train = (1 + y_train) / 2
inds = np.random.choice(np.arange(2500), 150, replace=False)
accuracy = np.zeros(150, dtype=np.float)
num_query = np.arange(1, 151)
for i in range(150):
graph.set_label(i, y_train[i])
LP(graph)
accuracy[i] = graph.accuracy(y_train)
np.savez("test_lp_random.npz", num_query=num_query, accuracy=accuracy)
#Author: Saibal De
#Class: EECS545 Machine Learning
#Title: LP with VOpt Queries
#Date: 12-13-2018
import numpy as np
import mnist_subset
import mnist_graph
from lp import LP
from VM import VM
num_train = 2500
num_test = 0
x_train, y_train, _, _ = mnist_subset.init(4, 9, num_train, num_test)
graph = mnist_graph.init(x_train)
y_train = (1 + y_train) / 2
num_query = np.arange(10, 151, 10)
accuracy = np.zeros(num_queries.size, dtype=np.float)
for j in range(num_queries.size):
# Reset graph
graph.u = [i for i in range(num_train)]
graph.l = []
# Run VM
indices = VM(graph.laplacian, num_train, num_queries[j])
# Query labels
import matplotlib.pyplot as plt
import mnist_subset
import mnist_graph
from lp import LP
from TSA import query_with_TSA
# Parameters
digit1 = 4
digit2 = 9
num_train = 1000
num_test = 0
# Load data and generate graph
x_train, y_train, _, _ = mnist_subset.init(digit1, digit2, num_train, num_test)
y_train = (1 + y_train) / 2
graph = mnist_graph.init(x_train)
q = (y_train == 1).sum() / y_train.size
print(graph.weights.max())
# Randomly initialize some labels
np.random.seed(0)
num_init_labels = 25
init_labels = np.random.choice(np.arange(0, num_train),
num_init_labels,
replace=False)
for i in init_labels:
graph.set_label(i, y_train[i])