def gen_img_type(shape,imagetype, SNRval,filename,islowres=False):
glob_const=globalconstants.GlobalConstants();
init_set(glob_const);
glob_const.particle_shape=shape;
glob_const.image_type=imagetype;
glob_const.savefile();
img_gen=image_generate.GenerateImage(SNRval);
if islowres==False:
img_gen.generate_img();
full_img_matrix=img_gen.noisy_imgarr;
part_img_matrix=full_img_matrix[0:150,0:130];
else:
img_gen.lowres_image();
full_img_matrix=img_gen.noisy_lowres;
part_img_matrix=full_img_matrix[0:37,0:32];
#filename='scripts/Paper_Figures/Figure1/A_round_SNR2.png';
plt.imshow(part_img_matrix,cmap=cm.Greys_r,interpolation='none');
plt.axis('off')
#plt.show()
#also tried scipy.misc.imsave but this provides more
#flexibility and more options
plt.savefig(filename,dpi=300);
plt.show();
return;
def training_round_script():
glob_const=globalconstants.GlobalConstants();
init_set(glob_const);
SNRval=0.5;
#training_round_list=[];
training_round_list=range(2,15);
training_acc=[];
testing_acc=[];
rounds_tillnow=[];
pos_tr_eg=glob_const.no_pos_training_eg;
neg_tr_eg=glob_const.no_neg_training_eg;
pos_te_eg=glob_const.no_pos_testing_eg;
neg_te_eg=glob_const.no_neg_testing_eg;
for training_round in training_round_list:
boost=adaboost_train_test.AdaBoostTrainTest(SNRval,training_round,\
pos_tr_eg,neg_tr_eg,pos_te_eg,neg_te_eg);
boost.train_adaboost();
boost.test_adaboost();
training_acc.append(boost.training_accuracy);
testing_acc.append(boost.testing_accuracy);
rounds_tillnow.append(training_round);
#this now saves the file at every iteration
#if you want to quit the program before it finishes
vid_utils.savefile(rounds_tillnow,training_acc,'data/adaboost_training_acc.txt')
vid_utils.savefile(rounds_tillnow,testing_acc,'data/adaboost_testing_acc.txt')
plt.plot(training_round_list,training_acc,'bo-',label='Training Accuracy');
plt.plot(training_round_list,testing_acc,'ro-',label='Testing Accuracy');
plt.legend();
plt.show();
def training_round_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
SNRval = 2.
#overfitting_e_list=[];
overfitting_e_list = [0.1 * x for x in range(0, 5)]
training_acc = []
testing_acc = []
list_till_now = []
for overfitting_e in overfitting_e_list:
glob_const.overfitting_e = overfitting_e
glob_const.savefile()
traintest = train_and_test.TrainAndTest(SNRval)
traintest.train_tree()
traintest.test_tree()
list_till_now.append(overfitting_e)
training_acc.append(traintest.training_accuracy)
testing_acc.append(traintest.testing_accuracy)
vid_utils.savefile(list_till_now, training_acc,
'data/overfittng_e_training_acc.txt')
vid_utils.savefile(list_till_now, testing_acc,
'data/overfitting_e_testing_acc.txt')
print "Processed e: " + str(overfitting_e) + "...."
plt.plot(overfitting_e_list,
training_acc,
'bo-',
label='Training Accuracy')
plt.plot(overfitting_e_list, testing_acc, 'ro-', label='Testing Accuracy')
plt.legend()
plt.show()
pass
def training_round_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
SNRval = 0.5
#training_round_list=[];
training_round_list = range(2, 11)
training_acc = []
testing_acc = []
for training_round in training_round_list:
glob_const.training_rounds = training_round
glob_const.savefile()
traintest = train_and_test.TrainAndTest(SNRval)
traintest.train_tree()
traintest.test_tree()
training_acc.append(traintest.training_accuracy)
testing_acc.append(traintest.testing_accuracy)
vid_utils.savefile(training_round_list, training_acc,
'data/training_acc.txt')
vid_utils.savefile(training_round_list, testing_acc,
'data/testing_acc.txt')
plt.plot(training_round_list,
training_acc,
'bo-',
label='Training Accuracy')
plt.plot(training_round_list, testing_acc, 'ro-', label='Testing Accuracy')
plt.legend()
plt.show()
pass
def __init__(self):
glob_const=globalconstants.GlobalConstants()
self.height=glob_const.global_window_height;
self.width=glob_const.global_window_width;
#image_generate.GenerateImage(11,11);
self.gray_level=50;
self.firstrect=None;
self.secondrect=None;
def __init__(self, SNR, no_pos_eg, no_neg_eg):
example_images.ExampleImages.__init__(self, SNR, no_pos_eg, no_neg_eg)
glob_const = globalconstants.GlobalConstants()
self._lowres_width = glob_const.lowres_width
self._lowres_height = glob_const.lowres_height
self.img_width = 512
self.img_height = 512
self._ratio = None
self._set_ratio()
def __init__(self):
glob_const = globalconstants.GlobalConstants()
self.global_window_width = glob_const.global_window_width
self.global_window_height = glob_const.global_window_height
self.rectlist = None
#defined by subclass
#Other variables, to be defined during optimization
#should we initialize them here?
self.threshold = None
self.toggle = None
self.error = None
self.margin = None
self.response_vector = None
def depth_script():
#important: line 199 in function train_node.py
#which is self._print_histogram in create_children()
#function is uncommented for this script to run
#comment it back and reinstall after the script runs
#this also depends on information from self._print_loginfo
#in tree_node.py
glob_const=globalconstants.GlobalConstants();
init_set(glob_const);
SNRval=0.5;
traintest=train_and_test.TrainAndTest(SNRval);
traintest.train_tree();
def exampleno_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
SNRval = 0.5
#exampleno_list=[];
exampleno_list = range(50, 1000, 50)
training_acc = []
testing_acc = []
list_tillnow = []
time_list = []
training_rounds = 15
pos_te_eg = glob_const.no_pos_testing_eg
neg_te_eg = glob_const.no_neg_testing_eg
for exampleno in exampleno_list:
starttime = time.time()
glob_const.no_pos_training_eg = exampleno
glob_const.no_neg_training_eg = exampleno
glob_const.savefile()
boost=adaboost_train_test.AdaBoostTrainTest(SNRval,training_rounds,\
exampleno,exampleno,pos_te_eg,neg_te_eg)
boost.train_adaboost()
boost.test_adaboost()
elapsed = time.time() - starttime
#append values
training_acc.append(boost.training_accuracy)
testing_acc.append(boost.testing_accuracy)
list_tillnow.append(exampleno)
time_list.append(elapsed)
vid_utils.savefile(list_tillnow,training_acc,\
'data/training_acc_eg.txt')
vid_utils.savefile(list_tillnow,testing_acc,\
'data/testing_acc_eg.txt')
vid_utils.savefile(list_tillnow,time_list,\
'data/time_vals.txt')
plt.plot(exampleno_list, training_acc, 'bo-', label='Training Accuracy')
plt.plot(exampleno_list, testing_acc, 'ro-', label='Testing Accuracy')
plt.legend()
plt.show()
def compare_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
#SNRlist=[];
SNRlist = [0.25 * x for x in range(8, 9)]
tree_acc = []
boost_acc = []
training_rounds = 7
pos_tr_eg = glob_const.no_pos_training_eg
neg_tr_eg = glob_const.no_neg_training_eg
pos_te_eg = glob_const.no_pos_testing_eg
neg_te_eg = glob_const.no_neg_testing_eg
list_till_now = []
for SNRval in SNRlist:
boost=adaboost_train_test.AdaBoostTrainTest(SNRval,training_rounds,\
pos_tr_eg,neg_tr_eg,pos_te_eg,neg_te_eg)
traintest = train_and_test.TrainAndTest(SNRval)
#Train
boost.train_adaboost()
traintest.train_tree()
#Test
boost.test_adaboost()
traintest.test_tree()
#traintest.post_prune_tree();
#traintest.test_tree();
#get accuracies
boost_acc.append(boost.testing_accuracy)
tree_acc.append(traintest.testing_accuracy)
list_till_now.append(SNRval)
vid_utils.savefile(list_till_now, tree_acc, "data/Tree_accuracies.txt")
vid_utils.savefile(list_till_now, boost_acc,
"data/Boost_accuracies.txt")
print "Processed SNR: " + str(SNRval) + "..."
plt.plot(SNRlist, tree_acc, 'ro', label="Tree")
plt.plot(SNRlist, boost_acc, 'bo', label="Adaboost")
plt.xlim(0, 2.5)
plt.legend(loc=2)
plt.show()
def training_round_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
SNRval = 2.
#exampleno_list=[];
round_list = range(5, 30, 5)
training_acc = []
testing_acc = []
list_tillnow = []
time_list = []
pos_te_eg = glob_const.no_pos_testing_eg
neg_te_eg = glob_const.no_neg_testing_eg
for training_rounds in round_list:
starttime = time.time()
pos_training_eg = glob_const.no_pos_training_eg
neg_training_eg = glob_const.no_neg_training_eg
glob_const.savefile()
boost=adaboost_train_test.AdaBoostTrainTest(SNRval,training_rounds,\
pos_training_eg,neg_training_eg,pos_te_eg,neg_te_eg)
boost.train_adaboost()
boost.test_adaboost()
elapsed = time.time() - starttime
#append values
training_acc.append(boost.training_accuracy)
testing_acc.append(boost.testing_accuracy)
list_tillnow.append(training_rounds)
time_list.append(elapsed)
vid_utils.savefile(list_tillnow,training_acc,\
'data/training_acc_trainingrounds.txt')
vid_utils.savefile(list_tillnow,testing_acc,\
'data/testing_acc_trainingrounds.txt')
vid_utils.savefile(list_tillnow,time_list,\
'data/time_vals.txt')
plt.plot(round_list, training_acc, 'bo-', label='Training Accuracy')
plt.plot(round_list, testing_acc, 'ro-', label='Testing Accuracy')
plt.legend(loc=2)
plt.show()
def training_round_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
examples = glob_const.no_pos_training_eg
init_t = time.time()
SNRval = 0.5
traintest = train_and_test.TrainAndTest(SNRval)
traintest.train_tree()
print "Trained: " + str(time.time() - init_t)
init_t = time.time()
traintest.test_tree()
print "Tested: " + str(time.time() - init_t)
init_t = time.time()
traintest.post_prune_tree()
print "Post Pruned: " + str(time.time() - init_t)
return
def __init__(self,SNR,pos_eg_no=10,neg_eg_no=10):
glob_const=globalconstants.GlobalConstants();
self.global_window_width=glob_const.global_window_width;
self.global_window_height=glob_const.global_window_height;
self.SNR=SNR;#we should be able to change SNR from the highest point in the program
self._img_gen=image_generate.GenerateImage(self.SNR);
self._img_width=self._img_gen.width;
self._img_height=self._img_gen.height;
self.num_particles=self._img_gen.part_rows*self._img_gen.part_columns;
#training examples
self.training_examples=None;
if pos_eg_no ==0 and neg_eg_no==0:
raise ValueError("Cannot create instance with 0 positive and 0 negative examples")
if pos_eg_no>=0:
self._num_pos_eg=pos_eg_no;#public variable: num_pos_eg
else:
raise ValueError("number of positive examples cannot be negative")
if neg_eg_no>=0:
self._num_neg_eg=neg_eg_no;#public variable: num_neg_eg
else:
raise ValueError("number of positive examples cannot be negative")
#example weights:
self._pos_example_weights=None;
self._neg_example_weights=None;
self._set_example_weights();
self.num_neg_eg_img=self.num_particles;#this is the no. of negative examples generated per image
self._eg_filename='_examples.pickle';
#random number generator
random.seed();
#to be defined later
self.test_img=None;
def roc_script(SNRval,particle_shape,image_type):
glob_const=globalconstants.GlobalConstants();
init_set(glob_const);
glob_const.image_type=image_type;
glob_const.particle_shape=particle_shape;
glob_const.savefile();
traintest=train_and_test.TrainAndTest(SNRval);
traintest.train_tree();
traintest.test_tree();
[precision,recall]=traintest.get_precision_recall_curve();
precision_filename="data/precision_"+str(SNRval)+"_"+str(image_type)+\
"_"+str(particle_shape)+".txt";
vid_utils.savefile(precision,recall,precision_filename)
[tpr,fpr]=traintest.get_roc_curve();
roc_filename="data/roc_"+str(SNRval)+"_"+str(image_type)+\
"_"+str(particle_shape)+".txt";
vid_utils.savefile(tpr,fpr,roc_filename)
def exampleno_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
SNRval = 0.5
#exampleno_list=[];
exampleno_list = range(50, 250, 50)
training_acc = []
testing_acc = []
list_tillnow = []
time_list = []
for exampleno in exampleno_list:
starttime = time.time()
glob_const.no_pos_training_eg = exampleno
glob_const.no_neg_training_eg = exampleno
glob_const.savefile()
traintest = train_and_test.TrainAndTest(SNRval)
traintest.train_tree()
traintest.test_tree()
elapsed = time.time() - starttime
#append values
training_acc.append(traintest.training_accuracy)
testing_acc.append(traintest.testing_accuracy)
list_tillnow.append(exampleno)
time_list.append(elapsed)
vid_utils.savefile(list_tillnow,training_acc,\
'data/training_acc_eg.txt')
vid_utils.savefile(list_tillnow,testing_acc,\
'data/testing_acc_eg.txt')
vid_utils.savefile(list_tillnow,time_list,\
'data/time_vals.txt')
plt.plot(exampleno_list, training_acc, 'bo-', label='Training Accuracy')
plt.plot(exampleno_list, testing_acc, 'ro-', label='Testing Accuracy')
plt.legend()
plt.show()
def gamma_script():
glob_const = globalconstants.GlobalConstants()
init_set(glob_const)
SNRval = 2.
#maxdepth_list=[];
#gamma_list=range(1,7);
gamma_list = [0]
training_acc = []
testing_acc = []
list_tillnow = []
time_list = []
for gamma in gamma_list:
starttime = time.time()
glob_const.gamma = gamma
glob_const.savefile()
traintest = train_and_test.TrainAndTest(SNRval)
traintest.train_tree()
traintest.test_tree()
elapsed = time.time() - starttime
#append values
training_acc.append(traintest.training_accuracy)
testing_acc.append(traintest.testing_accuracy)
list_tillnow.append(gamma)
time_list.append(elapsed)
vid_utils.savefile(list_tillnow,training_acc,\
'data/training_acc_gamma.txt')
vid_utils.savefile(list_tillnow,testing_acc,\
'data/testing_acc_gamma.txt')
vid_utils.savefile(list_tillnow,time_list,\
'data/time_vals_depth.txt')
plt.plot(gamma_list, training_acc, 'bo-', label='Training Accuracy')
plt.plot(gamma_list, testing_acc, 'ro-', label='Testing Accuracy')
plt.legend()
plt.show()
def __init__(self, training_rounds=10, conv_theta=0.45):
self.training_rounds = training_rounds
#list of features;
glob_const = globalconstants.GlobalConstants()
self.global_window_width = glob_const.global_window_width
self.global_window_height = glob_const.global_window_height
#logging.basicConfig(filename='logfile.txt',level=logging.DEBUG)
self.conv_theta = conv_theta
#saving stuff
self._feature_filename = 'data/feature_list.pickle'
self._response_matrix_filename = 'data/response_matrix.pickle'
#to be defined later
self._features = None
self._training_examples = None
self._no_pos_training_eg = None
self._no_neg_training_eg = None
self.adaboost_rule = None
self.log_loss = None
self.training_error = None
self._int_image_list = None
import pbtspot
from pbtspot import set_default,train_and_test, globalconstants
import matplotlib.pyplot as plt
set_default.set_default_constants();
#set constants
glob_const=globalconstants.GlobalConstants();
glob_const.no_pos_training_eg=50;
glob_const.no_neg_training_eg=50;
glob_const.no_pos_testing_eg=1000;
glob_const.no_neg_testing_eg=1000;
glob_const.savefile();
#main script
SNR=0.5;
train_test=train_and_test.TrainAndTest(SNR);
train_test.load_saved_tree();
#train_test.train_tree();
#train_test.save_tree();
train_test.test_tree();
[tpr,fpr]=train_test.get_roc_curve();
plt.plot(fpr,tpr,'ro-');
plt.xlim(-0.05, 1.05)
plt.ylim(-0.05,1.05)
plt.xlabel("False Positive Rate")
plt.ylabel("True Positive Rate")
print "tpr: "+str(tpr)
print "fpr: "+str(fpr)
plt.show();
def __init__(self, SNR, width=512, height=512):
"""
Generates a synthetic image of particles
@param: SNR: float, signal to noise ratio
@param width:integer, image width
@param height: integer, image height
public methods:
def self.generate_img()
def self.reset_image()
def save_image(self,filename='A.jpg',highres=True):
def lowres_image(self)
public variables:
self.noisy_imgarr
self.particle_loc
self.scaled_noisy_imgarr
self.noisy_lowres
self.scaled_noisy_lowres
"""
if (height <= 0):
raise ValueError("Height must be positive:%d" % height)
if (width <= 0):
raise ValueError("Width must be positive:%d" % width)
#from global values
glob_const = globalconstants.GlobalConstants()
self.global_window_width = glob_const.global_window_width
self.global_window_height = glob_const.global_window_height
self.particle_shape = glob_const.particle_shape
#round or elongated
self.image_type = glob_const.image_type
#image type, determining the type of noise
#To be defined later
self.part_arr = None
#image array containing only particle Gaussians
self.imgarr = None
#image array with particles and background
self.noisy_imgarr = None
# noisy array, after adding Poisson noise to self.imgarr
self.scaled_noisy_imgarr = None
#after scaling noisy_imgarr, so that mean=0, sttdev=1
self.particle_loc = None
#array of particle coordinates
self.bg_noise = None
#Only when required to generate low resolution images
self.noisy_lowres = None
self.scaled_noisy_lowres = None
#consider putting these to global constants as well
self.lowres_height = 128
self.lowres_width = 128
#from paper, resolution of camera is:
#delta_x=50nm, delta_y=50nm. Thus all values are multiples of
#50nm. So, sigma_x=2 implies sigma_x=100nm, since 1 pixel=50nm
#no. of particles: rows and columns
self.part_rows = 16
#by default
self.part_columns = 16
#std. dev for the Gaussian, in pixels
self.sigma_sym = 2.
self.sigma_major = 4.
if self.particle_shape == 'round':
self.sigma_x = self.sigma_sym
self.sigma_y = self.sigma_sym
elif self.particle_shape == 'elongated':
self.sigma_x = self.sigma_major
self.sigma_y = self.sigma_sym
self.height = height
self.width = width
#Image parameters
self.SNR = SNR
self.background_level = 10
#Always run this script before other scripts to give default values to global variables #ignore at your own peril #do not change default values here #change values in specific scripts #these are default values and should NOT be touched import pbtspot from pbtspot import globalconstants default_const=globalconstants.GlobalConstants(); #write all default values default_const.particle_shape='round' default_const.global_window_width=10; default_const.global_window_height=10; default_const.image_type='A' #default values end here default_const.savefile();
