def descStats(data):
"""
Compute descriptive statistics of data
"""
dataList = list(data)
logDataList = list(N.log10(dataList))
desc = dict()
if len(dataList) == 0:
desc['mean'] = 0
desc['median'] = 0
desc['logMean'] = 0
desc['logMedian'] = 0
elif len(dataList) < 2:
desc['mean'] = dataList[0]
desc['median'] = dataList[0]
desc['logMean'] = logDataList[0]
desc['logMedian'] = logDataList[0]
else:
desc['mean'] = mean(dataList)
desc['median'] = median(dataList)
desc['logMean'] = mean(logDataList)
desc['logMedian'] = median(logDataList)
if len(dataList) < 3:
desc['stdev'] = 0
desc['sterr'] = 0
desc['logStdev'] = 0
desc['logSterr'] = 0
else:
desc['stdev'] = std(dataList)
desc['sterr'] = stderr(dataList)
desc['logStdev'] = std(logDataList)
desc['logSterr'] = stderr(logDataList)
return desc
def gStats(self, missingValue=0.0):
"""dict of {geneID: (min,max,mean,median,std,stderr,
Shapiro-Wilk(w,p),normaltest_chisq (D'Agostino and Pearson),...}
"""
import scipy as S
import scipy.stats as SS
rv = {}
for k, v in self.items():
# print k,v
va = S.array(self.gValues(k, missingValue))
try:
normaltest = SS.normaltest(va)
except:
normaltest = None
try:
shapiro = SS.shapiro(va)
except:
shapiro = None
try:
rv[k] = (va.min(), va.max(), va.mean(), SS.median(va), SS.std(va), SS.stderr(va), normaltest, shapiro)
except:
print k, va
raise
return rv
def analyse(all_results):
analysis = {}
for parser_name, results in all_results.items():
scores = []
for result in results:
scores.append(result['score'])
analysis[parser_name] = (mean(scores), stderr(scores))
return analysis
def test_stderr(self):
"""
this is not in R, so used
sqrt(var(testcase))/sqrt(4)
"""
## y = stats.stderr(self.shoes[0])
## assert_approx_equal(y,0.775177399)
y = stats.stderr(self.testcase)
assert_approx_equal(y, 0.6454972244)
def test_stderr(self):
"""
this is not in R, so used
sqrt(var(testcase))/sqrt(4)
"""
## y = stats.stderr(self.shoes[0])
## assert_approx_equal(y,0.775177399)
y = stats.stderr(self.testcase)
assert_approx_equal(y,0.6454972244)
def mean_confidence_interval(data, confidence=0.90):
"""
T-distribution
"""
from scipy import stats
import numpy
data = np.array(data)
n = len(data)
m, se = np.mean(data), stats.stderr(data)
h = se * sp.stats.t._ppf((1+confidence)/2. , n-1)
return h*2
def eval_error_metrics(epoch, model, dataset, log_filename=None):
'''
Evaluate PSNR and SSIM over fixed evaluation dataset
'''
SSIM = []
PSNR = []
for data in dataset:
model.set_input(data)
model.test()
for j, b_path in enumerate(data['B_paths']):
sem_real = tensor2im(torch.unsqueeze(data['B'][j, ...], 0))
sem_pred = tensor2im(torch.unsqueeze(model.fake_B[j, ...], 0))
sem_real = np.array(Image.fromarray(sem_real).convert('L'))
sem_pred = np.array(Image.fromarray(sem_pred).convert('L'))
SSIM.append(compare_ssim(sem_pred, sem_real))
PSNR.append(compare_psnr(sem_pred, sem_real))
if b_path is not None:
save_image(sem_pred, b_path)
SSIM = np.array(SSIM)
mean_SSIM, std_err_SSIM = np.mean(SSIM), stderr(SSIM)
PSNR = np.array(PSNR)
mean_PSNR, std_err_PSNR = np.mean(PSNR), stderr(PSNR)
message = '(epoch: %d) PSNR=%s±%s, SSIM=%s±%s' % (
epoch, mean_PSNR, std_err_PSNR, mean_SSIM, std_err_SSIM)
print(message)
if log_filename is not None:
with open(log_filename, "a") as log_file:
log_file.write('%s\n' % message)
data['y'] = []
data['xci'] = []
data['yci'] = []
data['label'] = []
for run in sorted(valuemap.keys()):
rundata = valuemap[run]
row = {}
ci = {}
# Compute mean of all runs in all buckets
for bucket, bucketdata in rundata.iteritems():
bucketarray = np.array(bucketdata)
row[bucket] = bucketarray.mean()
# determine coinfidence interval if desired
if options.ci > 0:
ci[bucket] = stats.stderr(bucketarray) * stats.t._ppf((1+options.ci)/2., len(bucketarray))
if xParRegex.pattern not in row:
print "Error: x scalar \"" + args[0] + "\" not found, or all runs ignored due to the given include and exclude regexp!"
exit(-1)
if yParRegex.pattern not in row:
print "Error: y scalar \"" + args[1] + "\" not found, or all runs ignored due to the given include and exclude regexp!"
exit(-1)
data['x'].append(row[xParRegex.pattern])
data['y'].append(row[yParRegex.pattern])
if options.ci > 0:
data['xci'].append(ci[xParRegex.pattern])
data['yci'].append(ci[yParRegex.pattern])
else:
row={}
ci={}
if options.outfile:
if options.ci > 0:
outcol = np.zeros((3, len(bucketlist)))
else:
outcol = np.zeros((1, len(bucketlist)))
# Compute mean of all runs in all buckets
for bucket, bucketdata in rundata.iteritems():
bucketarray = np.array(bucketdata)
bucketmean = bucketarray.mean()
# determine coinfidence interval if desired
if options.ci > 0:
bucketci = stats.stderr(bucketarray) * stats.t._ppf((1+options.ci)/2., len(bucketarray)) * options.scale
ci[bucket] = bucketci
bucketmean*=options.scale
row[bucket] = bucketmean
if options.outfile:
outcol[0, bucketlist.index(bucket)] = bucketmean
if options.ci > 0:
outcol[1, bucketlist.index(bucket)] = bucketmean - bucketci
outcol[2, bucketlist.index(bucket)] = bucketmean + bucketci
if options.outfile:
outarray = np.vstack((outarray, outcol))
else:
# Plot row
if options.ci == 0:
row = {}
ci = {}
if options.outfile:
if options.ci > 0:
outcol = np.zeros((3, options.range[1] - options.range[0] + 1))
else:
outcol = np.zeros((1, options.range[1] - options.range[0] + 1))
# Compute mean of all runs in all buckets
for bucket, bucketdata in rundata.iteritems():
bucketarray = np.array(bucketdata)
bucketmean = bucketarray.mean()
# determine coinfidence interval if desired
if options.ci > 0:
bucketci = sp.stderr(bucketarray) * sp.t._ppf(
(1 + options.ci) / 2., len(bucketarray)) * options.scale
ci[bucket] = bucketci
bucketmean *= options.scale
row[bucket] = bucketmean
if options.outfile:
outcol[0, bucket - options.range[0]] = bucketmean
if options.ci > 0:
outcol[1, bucket - options.range[0]] = bucketmean - bucketci
outcol[2, bucket - options.range[0]] = bucketmean + bucketci
if options.outfile:
outarray = np.vstack((outarray, outcol))
else:
# Plot row
row = {}
ci = {}
if options.outfile:
if options.ci > 0:
outcol = np.zeros((3, len(bucketlist)))
else:
outcol = np.zeros((1, len(bucketlist)))
# Compute mean of all runs in all buckets
for bucket, bucketdata in rundata.iteritems():
bucketarray = np.array(bucketdata)
bucketmean = bucketarray.mean()
# determine coinfidence interval if desired
if options.ci > 0:
bucketci = stats.stderr(bucketarray) * stats.t._ppf(
(1 + options.ci) / 2., len(bucketarray)) * options.scale
ci[bucket] = bucketci
bucketmean *= options.scale
row[bucket] = bucketmean
if options.outfile:
outcol[0, bucketlist.index(bucket)] = bucketmean
if options.ci > 0:
outcol[1, bucketlist.index(bucket)] = bucketmean - bucketci
outcol[2, bucketlist.index(bucket)] = bucketmean + bucketci
if options.outfile:
outarray = np.vstack((outarray, outcol))
else:
# Plot row
row={}
ci={}
if options.outfile:
if options.ci > 0:
outcol = np.zeros((3, len(bucketlist)))
else:
outcol = np.zeros((1, len(bucketlist)))
# Compute mean of all runs in all buckets
for bucket, bucketdata in rundata.iteritems():
bucketarray = np.array(bucketdata)
bucketmean = bucketarray.mean()
# determine coinfidence interval if desired
if options.ci > 0:
bucketci = stats.stderr(bucketarray) * stats.t._ppf((1+options.ci)/2., len(bucketarray)) * options.scale
ci[bucket] = bucketci
bucketmean*=options.scale
row[bucket] = bucketmean
if options.outfile:
outcol[0, bucketlist.index(bucket)] = bucketmean
if options.ci > 0:
outcol[1, bucketlist.index(bucket)] = bucketmean - bucketci
outcol[2, bucketlist.index(bucket)] = bucketmean + bucketci
if options.outfile:
outarray = np.vstack((outarray, outcol))
else:
# Plot row
if options.ci == 0:
data['y'] = []
data['xci'] = []
data['yci'] = []
data['label'] = []
for run in sorted(valuemap.keys()):
rundata = valuemap[run]
row = {}
ci = {}
# Compute mean of all runs in all buckets
for bucket, bucketdata in rundata.iteritems():
bucketarray = np.array(bucketdata)
row[bucket] = bucketarray.mean()
# determine coinfidence interval if desired
if options.ci > 0:
ci[bucket] = stats.stderr(bucketarray) * stats.t._ppf(
(1 + options.ci) / 2., len(bucketarray))
if xParRegex.pattern not in row:
print "Error: x scalar \"" + args[
0] + "\" not found, or all runs ignored due to the given include and exclude regexp!"
exit(-1)
if yParRegex.pattern not in row:
print "Error: y scalar \"" + args[
1] + "\" not found, or all runs ignored due to the given include and exclude regexp!"
exit(-1)
data['x'].append(row[xParRegex.pattern])
data['y'].append(row[yParRegex.pattern])
if options.ci > 0:
