def plot_roc(predictions, fig_title, outdir):
plt.clf()
roc_labels = [str(k) for k in xrange(1, len(predictions) + 1)]
roc_labels[-1] = 'mean'
pyroc.plot_multiple_roc(predictions,
title=fig_title,
labels=roc_labels,
include_baseline=True)
for i in xrange(1, N_repeats + 1):
if not os.path.exists(outdir + "roc%d.pdf" % i):
plt.savefig(outdir + "roc%d.pdf" % i)
break
def ROCPlot(title, labels=None, *args):
'''
If the PyROC (https://github.com/marcelcaraciolo/PyROC)
module is installed, display the ROC curve for SVM/Logistic Regression classifiers.
Inputs:
=======
labels : Labels for the legend
args: Variable length arguments of the form : actual_1[], predicted_1[], actual_2[], predicted_2[], ....
'''
try:
from pyroc import random_mixture_model, ROCData, plot_multiple_roc
import pylab
except ImportError:
try:
from pyroc import random_mixture_model, ROCData, plot_multiple_roc
except ImportError:
print 'PyROC does not exist, skipping ROC demo. Install PyROC from : https://github.com/marcelcaraciolo/PyROC '
return
if (len(args) == 0):
x = random_mixture_model()
r1 = ROCData(x)
y = random_mixture_model()
r2 = ROCData(y)
lista = [r1, r2]
labels = ['Algorithm-1', 'Algorithm-2']
else:
lista = []
for i in range(0, len(args), 2):
x1 = args[i]
y1 = args[i + 1]
x1y1 = ((x1[k], y1[k]) for k in range(len(x1)))
r1 = ROCData(x1y1)
auc = '%.2f' % r1.auc()
if (labels):
labels[i / 2] = labels[i / 2] + ', AUC: {0} '.format(auc)
lista.append(r1)
plot_multiple_roc(lista, title, include_baseline=True, labels=labels)
pylab.close()
def ROCPlot(title, labels=None,*args):
'''
If the PyROC (https://github.com/marcelcaraciolo/PyROC)
module is installed, display the ROC curve for SVM/Logistic Regression classifiers.
Inputs:
=======
labels : Labels for the legend
args: Variable length arguments of the form : actual_1[], predicted_1[], actual_2[], predicted_2[], ....
'''
try:
from pyroc import random_mixture_model, ROCData, plot_multiple_roc
import pylab
except ImportError:
try:
from pyroc import random_mixture_model, ROCData, plot_multiple_roc
except ImportError:
print 'PyROC does not exist, skipping ROC demo. Install PyROC from : https://github.com/marcelcaraciolo/PyROC '
return
if(len(args)==0):
x = random_mixture_model()
r1 = ROCData(x)
y = random_mixture_model()
r2 = ROCData(y)
lista = [r1,r2]
labels = ['Algorithm-1','Algorithm-2']
else:
lista = []
for i in range(0,len(args),2):
x1 = args[i]
y1 = args[i+1]
x1y1 = ((x1[k],y1[k]) for k in range(len(x1)))
r1 = ROCData(x1y1)
auc = '%.2f'%r1.auc()
if(labels):
labels[i/2] = labels[i/2]+ ', AUC: {0} '.format(auc)
lista.append(r1)
plot_multiple_roc(lista,title,include_baseline=True,labels=labels)
pylab.close()
def main(argv):
parser = argparse.ArgumentParser(description="Run sentiment analysis using\
a positive and a negative input file")
parser.add_argument("-p", "--positive", help="input relative path of a \
positive data file", required=True)
parser.add_argument("-n", "--negative", help="input relative path of a \
negative data file", required=True)
parser.add_argument("-d", "--divisions", type=int, help="select the number \
of divisions created in input data: 1 out of d will \
be used for testing.", default=4)
parser.add_argument("-l", "--limit_features", type=int, help="number of best \
features to use", default="0")
parser.add_argument("-b", "--bigram", help="classify using bigram features.",
action="store_true")
parser.add_argument("-s", "--stopwords", help="filter out stop words before \
training.", action="store_true")
# parser.add_argument("-t", "--tag_negated_words", help="tag negated words with \
# word_not to capture more meaning.", action="store_true")
parser.add_argument("-r", "--randomize", help="randomize training data to \
reduce clumping while training.", action="store_true")
parser.add_argument("-a", "--average", help="train and test over each \
possible set of divisions and average the results for \
more smoothing.", action="store_true")
parser.add_argument("-g", "--graph", help="graphs the resulting ROC curves \
against eachother", action="store_true")
args = parser.parse_args()
# Set up ROC graphing data and import pyroc as needed
ROC_data = 0
if args.graph:
from pyroc import plot_multiple_roc
ROC_data = [[],[]]
# Set up stopword set
stopset = []
if args.stopwords:
from nltk.corpus import stopwords
print "Stop words are being filtered out."
stopset = set(stopwords.words('english'))
# Finds word scores if not using bigrams
word_scores = []
if not args.bigram:
word_scores = create_word_scores(args.positive, args.negative)
# Check to see what mode of testing is being used; input feature limit:
if args.limit_features:
limit = args.limit_features
if (args.bigram):
print '\nEvaluating the best %d bigram word features\n' % (limit)
evaluate_features(bigram_word_features, args.positive, args.negative, \
args.divisions, args.average, limit, args.randomize, \
args.stopwords, stopset, word_scores, ROC_data)
else:
print '\nEvaluating the best %d word features\n' % (limit)
evaluate_features(best_word_features, args.positive, args.negative, \
args.divisions, args.average, limit, args.randomize, \
args.stopwords, stopset, word_scores, ROC_data)
# Or iteration through default array of feature numbers
else:
for limit in NUM_FEATURES_TO_TEST:
if (args.bigram):
print '\nEvaluating the best %d bigram word features\n' % (limit)
evaluate_features(bigram_word_features, args.positive, args.negative, \
args.divisions, args.average, limit, args.randomize, \
args.stopwords, stopset, word_scores, ROC_data)
else:
print '\nEvaluating the best %d word features\n' % (limit)
evaluate_features(best_word_features, args.positive, args.negative, \
args.divisions, args.average, limit, args.randomize, \
args.stopwords, stopset, word_scores, ROC_data)
if args.graph:
plot_multiple_roc(ROC_data[0],'ROC Curves', labels = ROC_data[1])
truth = open("Paper/pred_results/truth.csv")
true_vals = {}
for line in truth:
if line.strip() == '':
continue
vals = line.strip().split("\t")
true_vals[vals[1]] = int(vals[0])
roclist = []
labels = []
for fname in sys.argv[1:]:
infile = open(fname)
data = []
for line in infile:
if line.strip() == '':
continue
vals = line.strip().split(",")
if not vals[0] in true_vals:
continue
data.append((true_vals[vals[0]], float(vals[1])))
roc = pyroc.ROCData(data)
print fname + ": " + str(roc.auc())
roclist.append(roc)
labels.append(";".join(
[x for x in fname.split("/")[-1].split("_")[0:-1] if x != ""]))
pyroc.plot_multiple_roc(roclist, labels=labels)
from gusPyCode.defs.HMMsplice_utils import * import pyroc as roc def toROCdata(scoredDict): rocData = [] for k in scoredDict: rocData.append((scoredDict[k][1],scoredDict[k][0],k)) return rocData anno = '/Users/biggus/Documents/James/Data/genomes/AaegL1/aaegypti.Tx-Ensembl.bed' mult = '/Users/biggus/Documents/James/Data/Solexa/aedes/hmmSplicer/finalResults/Lx_unfiltered/LX.gtag.collapsed.multi.bed' sngl = '/Users/biggus/Documents/James/Data/Solexa/aedes/hmmSplicer/finalResults/Lx_unfiltered/LX.gtag.collapsed.sngl.bed' mult = toROCdata(generateROCcurve(mult, anno, stepSize=10, wiggle=3)[1]) sngl = toROCdata(generateROCcurve(sngl, anno, stepSize=10, wiggle=3)[1]) mROC = roc.ROCData(mult,linestyle='r-') sROC = roc.ROCData(sngl,linestyle='b-') roc.plot_multiple_roc([mROC,sROC],title='Multiples vs Singles',labels=['Multiples','Singles'], include_baseline=1, equal_aspect=True) mROC.auc() sROC.auc()
def PlotMultipleROC(rocs, title='', labels=None, include_baseline=True):
import pyroc
pyroc.plot_multiple_roc(rocs, title, labels, include_baseline)