def sliding_MWU(values):
"""
RETURN
:param values:
:return:
"""
# ADD MEDIAN TO EITHER SIDE OF values
prefix = [np.median(values[: i + weight_radius]) for i in range(weight_radius)]
suffix = [
np.median(values[-i - weight_radius:])
for i in reversed(range(weight_radius))
]
combined = np.array(prefix + list(values) + suffix)
b = combined.itemsize
window = as_strided(
combined, shape=(len(values), weight_radius * 2), strides=(b, b)
)
med = (len(median_weight) + 1) / 2
m_score = np.array(
[
stats.mannwhitneyu(
w[:weight_radius],
w[-weight_radius:],
use_continuity=True,
alternative="two-sided",
)
for v in window
for r in [rankdata(v)]
for w in [(r - med) * median_weight]
]
)
return m_score
def jitter_MWU(values, start, mid, end):
"""
RETURN A BETTER MIDPOINT< ACCOUNTING FOR t-test RESULTS
"""
# ADD SOME CONSTRAINTS TO THE RANGE OF VALUES TESTED
m_start = min(mid, max(start + MIN_POINTS, mid - JITTER))
m_end = max(mid, min(mid + JITTER, end - MIN_POINTS))
if m_start == m_end:
return no_good_edge, no_good_edge, mid
mids = np.array(range(m_start, m_end))
# MWU SCORES
try:
m_score = np.array([
stats.mannwhitneyu(
values[max(start, m - MAX_POINTS):m],
values[m:min(end, m + MAX_POINTS)],
use_continuity=True,
alternative="two-sided",
) for m in mids
])
t_score = np.array([
stats.ttest_ind(
values[max(start, m - MAX_POINTS):m],
values[m:min(end, m + MAX_POINTS)],
equal_var=False,
) for m in mids
])
except Exception as e:
e = Except.wrap(e)
if "All numbers are identical" in e:
return no_good_edge, no_good_edge, mids[0]
raise e
# TOTAL SUM-OF-SQUARES
# DO NOT KNOW WHAT THIS WAS DOING
# if m_start - start == 0:
# # WE CAN NOT OFFSET BY ONE, SO WE ADD A DUMMY VALUE
# v_prefix = np.array([np.nan] + list(not_right(cumSS(values[start:m_end]), 1)))
# else:
# # OFFSET BY ONE, WE WANT cumSS OF ALL **PREVIOUS** VALUES
# v_prefix = not_right(
# not_left(cumSS(values[start:m_end]), m_start - start - 1), 1
# )
# v_suffix = not_right(cumSS(values[m_start:end][::-1])[::-1], end - m_end)
# v_score = v_prefix + v_suffix
# pvalue = np.sqrt(m_score[:, 1] * v_score) # GOEMEAN OF SCORES
# PICK LOWEST
pvalue = np.sqrt(m_score[:, 1] * t_score[:, 1])
best = np.argmin(pvalue)
return Data(pvalue=m_score[best, 1]), Data(pvalue=t_score[best,
1]), mids[best]
def square_vs_diag(codon_permutation_f, outdir):
# Replicates the analysis presented in Fig. 5B and creates plot
ret = Utils.Load(codon_permutation_f)
npr = np.array(ret['n+_risk'])
codes = ret['code']
squares = np.array([issquare(i) for i in codes[1:]])
diags = np.array([isdiag(i) for i in codes[1:]])
print('n diags: {}'.format(sum(diags)))
print('n squares: {}'.format(sum(squares)))
_, ax = plt.subplots(1, figsize=(4.7, 5.2), dpi=144)
grps = [npr[1:][squares], npr[1:][~squares & ~diags], npr[1:][diags]]
ax.boxplot(grps,
showfliers=False,
whis=(5, 95),
flierprops={
'color': 'k',
'marker': 'x',
'markersize': 2
},
boxprops={
'color': 'k',
'lw': 0.6
},
capprops={
'color': 'k',
'lw': 0.6
},
whiskerprops={
'color': 'k',
'lw': 0.6
},
medianprops={
'color': '',
'lw': 1.2
})
ax.set_ylim(0.15, 0.31)
ax.set_yticks([0.15, 0.2, 0.25, 0.3])
print('squares vs all: {}'.format(mannwhitneyu(grps[0], grps[1])))
print('squares vs diags: {}'.format(mannwhitneyu(grps[0], grps[2])))
print('diags vs all: {}'.format(mannwhitneyu(grps[2], grps[1])))
plt.savefig(join(outdir, 'Squares_diags.png'), dpi=144)
def jitter_MWU(values, start, mid, end):
# ADD SOME CONSTRAINTS TO THE RANGE OF VALUES TESTED
m_start = min(mid, max(start + MIN_POINTS, mid - JITTER))
m_end = max(mid, min(mid + JITTER, end - MIN_POINTS))
if m_start == m_end:
return no_good_edge, no_good_edge, mid
mids = np.array(range(m_start, m_end))
# MWU SCORES
m_score = np.array(
[
stats.mannwhitneyu(
values[max(start, m - MAX_POINTS) : m],
values[m : min(end, m + MAX_POINTS)],
use_continuity=True,
alternative="two-sided",
)
for m in mids
]
)
t_score = np.array(
[
stats.ttest_ind(
values[max(start, m - MAX_POINTS) : m],
values[m : min(end, m + MAX_POINTS)],
equal_var=False,
)
for m in mids
]
)
# TOTAL SUM-OF-SQUARES
if m_start - start == 0:
# WE CAN NOT OFFSET BY ONE, SO WE ADD A DUMMY VALUE
v_prefix = np.array([np.nan] + list(not_right(cumSS(values[start:m_end]), 1)))
else:
# OFFSET BY ONE, WE WANT cumSS OF ALL **PREVIOUS** VALUES
v_prefix = not_right(
not_left(cumSS(values[start:m_end]), m_start - start - 1), 1
)
v_suffix = not_right(cumSS(values[m_start:end][::-1])[::-1], end - m_end)
v_score = v_prefix + v_suffix
# PICK LOWEST
pvalue = np.sqrt(m_score[:, 1] * v_score) # GOEMEAN OF SCORES
best = np.argmin(pvalue)
return Data(pvalue=m_score[best, 1]), Data(pvalue=t_score[best, 1]), mids[best]
def mannwhitneyu_test(approaches, accuracy_values, save_path):
# Compute the Mann-Whitney rank test on samples x and y.
mannwhitneyu_test_frame = pd.DataFrame()
for i in range(len(approaches)):
for j in range(i, len(approaches), 1):
# iterate through approaches
approach_i = approaches[i]
approach_j = approaches[j]
values_i = accuracy_values.loc[:, approach_i]
values_j = accuracy_values.loc[:, approach_j]
t_statistic, two_tailed_p_test = stats.mannwhitneyu(values_i, values_j)
mannwhitneyu_test_frame.at[approach_i, approach_j] = two_tailed_p_test
save_path.mkdir(parents=True, exist_ok=True)
fig = plt.figure(figsize=(4, 2))
ax = fig.subplots()
ax = sns.heatmap(mannwhitneyu_test_frame, ax=ax, annot=True, fmt="0.3f", cmap="autumn", vmin=0, vmax=0.05)
plt.xticks(rotation=45)
fig.canvas.start_event_loop(sys.float_info.min)
path = save_path / 'mannwhitneyu.png'
fig.savefig(path, bbox_inches='tight', dpi=100)
plt.close(fig)
def sliding_MWU(values):
"""
RETURN
:param values:
:return:
"""
# ADD MEDIAN TO EITHER SIDE OF values
prefix = [
np.median(values[:i + weight_radius]) for i in range(weight_radius)
]
suffix = [
np.median(values[-i - weight_radius:])
for i in reversed(range(weight_radius))
]
combined = np.array(prefix + list(values) + suffix)
b = combined.itemsize
window = as_strided(combined,
shape=(len(values), weight_radius * 2),
strides=(b, b))
med = (len(median_weight) + 1) / 2
try:
m_score = np.array([
stats.mannwhitneyu(
w[:weight_radius],
w[-weight_radius:],
use_continuity=True,
alternative="two-sided",
) for v in window for r in [rankdata(v)]
for w in [(r - med) * median_weight]
])
return m_score
except Exception as cause:
cause = Except.wrap(cause)
if "All numbers are identical" in cause:
return np.ones((window.shape[0], 2))
raise cause
def plotExpBox_Main(inputFiles,headers,valcols,outputFile,sep,startRow,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,plotPvalueCluster,outputClusterPrefix,methodCluster,xlegendrotation,xlabe,ylabe,figsz,titl,showSampleSizes,trimToMinSize,relabels,logb,plotHistogramToFile,plotMedianForGroups,botta,showViolin,showBox,firstColAnnot,plotTrend,showLegend,makePzfxFile,makeBinMatrix,writeDataSummaryStat,summaryStatRange,minuslog10pvalue,minNDataToKeep,vfacecolor,valpha,outXYZPvalues,dividePlots):
#if plotPvalueCluster:
#if pvalue cluster is needed:
# from Bio.Cluster.cluster import *
# from Bio.Cluster import *
#endif
#the real deal!
plotData=[]
xtickLabels=[]
trendData={}
annot={}
minSize=-1
for inputFile,header,cols in zip(inputFiles,headers,valcols):
fin=generic_istream(inputFile)
startIdx=len(plotData)
if firstColAnnot:
colAnnot=cols[0]
cols=cols[1:]
annotThisFile=[]
annot[startIdx]=annotThisFile
else:
colAnnot=-1
annotThisFile=None
for col in cols:
plotData.append([])
xtickLabels.append(header[col])
colIndices=range(startIdx,startIdx+len(cols))
if plotTrend:
#print >> stderr,"plotTrend"
trendDataThisFile=[]
trendData[startIdx]=trendDataThisFile
else:
trendDataThisFile=None
lino=0
for lin in fin:
lino+=1
if lino<startRow:
continue
fields=lin.rstrip("\r\n").split(sep)
if plotTrend:
#print >> stderr,"a"
trendDataThisLine=[]
else:
trendDataThisLine=None
allDataOKThisLine=True
if colAnnot>=0:
annotThisFile.append(fields[colAnnot])
for idx,col in zip(colIndices,cols):
try:
value=float(fields[col])
if logb!=0:
if value==0.0:
raise ValueError
value=log(value)/logb
plotData[idx].append(value)
if plotTrend:
trendDataThisLine.append(value)
#print >> stderr,"value:",value
except:
allDataOKThisLine=False
if plotTrend:
if allDataOKThisLine:
trendDataThisFile.append(trendDataThisLine)
else:
trendDataThisFile.append(None)
fin.close()
if minSize==-1:
minSize=len(plotData[idx]) #or startIDX?
else:
minSize=min([minSize,len(plotData[idx])])
if trimToMinSize:
print >> stderr,"trimming to min size =",minSize
trimData(plotData,minSize)
if len(relabels)>0:
#if len(relabels)!=len(xtickLabels):
# print >> stderr,"relabels doesn't have the same length as original label vectors",xtickLabels,"=>",relabels
# exit()
print >> stderr,xtickLabels
print >> stderr,relabels
for i,relabel in zip(range(0,len(relabels)),relabels):
xtickLabels[i]=relabel
for i in range(0,len(plotMedianForGroups)):
plotMedianForGroups[i]=getCol0ListFromCol1ListStringAdv(xtickLabels,plotMedianForGroups[i])
#drawing medians:
medianToDraw=[]
for mediangrouper in plotMedianForGroups:
curD=[]
for c in mediangrouper:
curD.extend(plotData[c])
medianToDraw.append(median(curD))
for c in range(len(plotData)-1,-1,-1):
if len(plotData[c])<minNDataToKeep:
print >> stderr,xtickLabels[c],"discarded because has only",len(plotData[c]),"data points <",minNDataToKeep
del plotData[c]
del xtickLabels[c]
if not skipStat:
print >> stdout,"student t-test (1 sample; mean=0)"
print >> stdout,"sample","mean","p-val","median"
if writeDataSummaryStat:
fDSS=open(writeDataSummaryStat,"w")
print >> fDSS,"sample\tmean\tvar\tsd\tmin\tmax\tN\tNInRange["+str(summaryStatRange[0])+","+str(summaryStatRange[1])+"]\t%NInRange\tNbelowRange\t%Nbelow\tNAboveRange\t%NAbove"
for x in range(0,len(plotData)):
#print >> stderr, len(plotData[x])
try:
print >> stdout, xtickLabels[x],mean(plotData[x]),ttest_1samp(plotData[x],0)[1],median(plotData[x])
except:
print >> stdout, xtickLabels[x],mean(plotData[x]),"NA",median(plotData[x])
if writeDataSummaryStat:
sumData,N,NIN,NBelow,NAbove=filterDataInRangeInclusive(plotData[x],summaryStatRange[0],summaryStatRange[1])
if NIN>1:
#print >> stderr,"sumData=",sumData
#print >> stderr,mean
mea=mean2(sumData)
DDOF=1
sd=std(sumData,ddof=DDOF)
var=sd*sd
mi=min(sumData)
ma=max(sumData)
else:
mea="NA"
sd="NA"
var="NA"
mi="NA"
ma="NA"
print >> fDSS,xtickLabels[x]+"\t"+str(mea)+"\t"+str(var)+"\t"+str(sd)+"\t"+str(mi)+"\t"+str(ma)+"\t"+str(N)+"\t"+str(NIN)+"\t"+str(float(NIN)*100/N)+"\t"+str(NBelow)+"\t"+str(float(NBelow)*100/N)+"\t"+str(NAbove)+"\t"+str(float(NAbove)*100/N)
pvalueM=[]
if writeDataSummaryStat:
fDSS.close()
print >> stdout,""
print >> stdout,"student t-test (2 samples)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
try:
pvalue=ttest_ind(plotData[x],plotData[y])[1]
except:
pvalue=1.0
if minuslog10pvalue and str(pvalue)!="NA":
try:
pvalue=-1*log(pvalue,10)
except:
pvalue=-1000.0
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
print >> stdout,""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_t_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_t",xtickLabels,pvalueM,methodCluster)
pvalueM=[]
print >> stdout,"welch t-test"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
try:
pvalue=welchs_approximate_ttest_arr(plotData[x],plotData[y])[3]
except:
pvalue=1.0
if minuslog10pvalue and str(pvalue)!="NA":
try:
pvalue=-1*log(pvalue,10)
except:
pvalue=-1000.0
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
if outXYZPvalues:
writeXYZPvalues(outXYZPvalues+"_Welch.xyz",xtickLabels,pvalueM)
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_Welch_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_Welch",xtickLabels,pvalueM,methodCluster)
print >> stdout,""
print >> stdout,"non-parametric (Mann-Whitney U)" #"non-parametric (Mann-Whitney U if larger n<=20 else Wilcoxon)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue=mannwhitneyu(plotData[x],plotData[y])[1]*2
except:
pvalue=1.0
if minuslog10pvalue and str(pvalue)!="NA":
try:
pvalue=-1*log(pvalue,10)
except:
pvalue=-1000.0
print >> stdout,pvalue, #mann-whiteney need to mul by 2 (one tail to two tail)
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if outXYZPvalues:
writeXYZPvalues(outXYZPvalues+"_U.xyz",xtickLabels,pvalueM)
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_U_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_U",xtickLabels,pvalueM,methodCluster)
#####now the variance tests
print >> stdout,""
print >> stdout,"Ansari-Bradley Two-sample Test for difference in scale parameters "
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue=ansari(plotData[x],plotData[y])[1]
except:
pvalue="NA"
if minuslog10pvalue and str(pvalue)!="NA":
try:
pvalue=-1*log(pvalue,10)
except:
pvalue=-1000.0
#pvalue=1.0
print >> stdout,pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_Ansari_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_Ansari",xtickLabels,pvalueM,methodCluster)
#####
#####now the variance tests
print >> stdout,""
print >> stdout,"Fligner's Two-sample Test for equal variance (non-parametrics)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue=fligner(plotData[x],plotData[y])[1]
except:
pvalue="NA"
#pvalue=1.0
if minuslog10pvalue and str(pvalue)!="NA":
try:
pvalue=-1*log(pvalue,10)
except:
pvalue=-1000.0
print >> stdout,pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_fligner_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_fligner",xtickLabels,pvalueM,methodCluster)
#####
#####now the variance tests
print >> stdout,""
print >> stdout,"Levene's Two-sample Test for equal variance"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue=levene(plotData[x],plotData[y])[1]
except:
pvalue="NA"
#pvalue=1.0
if minuslog10pvalue and str(pvalue)!="NA":
try:
pvalue=-1*log(pvalue,10)
except:
pvalue=-1000.0
print >> stdout,pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_levene_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_levene",xtickLabels,pvalueM,methodCluster)
#####
#####now the variance tests
print >> stdout,""
print >> stdout,"Bartlett's Two-sample Test for equal variance (for normal distributions)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue=bartlett(plotData[x],plotData[y])[1]
except:
pvalue="NA"
#pvalue=1.0
if minuslog10pvalue and str(pvalue)!="NA":
try:
pvalue=-1*log(pvalue,10)
except:
pvalue=-1000.0
print >> stdout,pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_bartlett_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_bartlett",xtickLabels,pvalueM,methodCluster)
#####
figure(figsize=figsz)
subplots_adjust(top=0.9, bottom=botta, left=0.2, right=0.8)
if len(titl)==0:
titl=outputFile
plotExpBox(plotData,xtickLabels,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,xlegendrotation,xlabe,ylabe,titl,showSampleSizes,showViolin,showBox,annot,trendData,showLegend,makePzfxFile,makeBinMatrix,dividePlots)
#ylim([0,200])
for m in medianToDraw:
axhline(y=m,linestyle=':',color='gray')
savefig(outputFile,bbox_inches="tight")
if len(plotHistogramToFile)>0:
drawHistogram(plotHistogramToFile,plotData,xtickLabels)
drawDensigram(plotHistogramToFile+".density.png",plotData,xtickLabels)
def plotExpBox_Main(inputFile,header,cols,outputFile,sep,startRow,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,plotPvalueCluster,outputClusterPrefix,methodCluster,xlegendrotation,xlabe,ylabe,figsz,titl,showSampleSizes):
#if plotPvalueCluster:
#if pvalue cluster is needed:
# from Bio.Cluster.cluster import *
# from Bio.Cluster import *
#endif
#the real deal!
fin=generic_istream(inputFile)
plotData=[]
xtickLabels=[]
for col in cols:
plotData.append([])
xtickLabels.append(header[col])
colIndices=range(0,len(cols))
lino=0
for lin in fin:
lino+=1
if lino<startRow:
continue
fields=lin.rstrip("\r\n").split(sep)
for idx,col in zip(colIndices,cols):
try:
value=float(fields[col])
plotData[idx].append(value)
except:
pass
fin.close()
print >> stdout,"student t-test (1 sample; mean=0)"
print >> stdout,"sample","mean","p-val"
for x in range(0,len(plotData)):
print >> stdout, xtickLabels[x],mean(plotData[x]),ttest_1samp(plotData[x],0)[1]
pvalueM=[]
print >> stdout,""
print >> stdout,"student t-test (2 samples)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
pvalue=ttest_ind(plotData[x],plotData[y])[1]
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
print >> stdout,""
makePValueRawPlot(outputClusterPrefix+"_t_raw",xtickLabels,pvalueM)
if plotPvalueCluster:
makePValueClusterPlot(outputClusterPrefix+"_t",xtickLabels,pvalueM,methodCluster)
pvalueM=[]
print >> stdout,"welch t-test"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
pvalue=welchs_approximate_ttest_arr(plotData[x],plotData[y])[3]
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
makePValueRawPlot(outputClusterPrefix+"_Welch_raw",xtickLabels,pvalueM)
if plotPvalueCluster:
makePValueClusterPlot(outputClusterPrefix+"_Welch",xtickLabels,pvalueM,methodCluster)
print >> stdout,""
print >> stdout,"non-parametric (Mann-Whitney U)" #"non-parametric (Mann-Whitney U if larger n<=20 else Wilcoxon)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
pvalue=mannwhitneyu(plotData[x],plotData[y])[1]*2
print >> stdout,pvalue, #mann-whiteney need to mul by 2 (one tail to two tail)
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
makePValueRawPlot(outputClusterPrefix+"_U_raw",xtickLabels,pvalueM)
if plotPvalueCluster:
makePValueClusterPlot(outputClusterPrefix+"_U",xtickLabels,pvalueM,methodCluster)
figure(figsize=figsz)
if len(titl)==0:
titl=outputFile
plotExpBox(plotData,xtickLabels,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,xlegendrotation,xlabe,ylabe,titl,showSampleSizes)
#ylim([0,200])
savefig(outputFile,bbox_inches="tight")
def compute_auc(train, test, Y, y):
"""Computes measures of accuracy for train and test data.
Computes the ROC curve and the area under that curve, as a
measure of classification accuracy. The threshold corresponding
to the point on the ROC curve farthest from `y=x` line is selected
and fraction of correct predictions corresponding to that
threshold is returned.
in
Arguments
train : float array
Array of predictions of the model on training data where rows
correspond to labels and columns correspond to samples.
test : float array
Array of predictions of the model on testing data where rows
correspond to labels and columns correspond to samples.
Y : float array
Training labels where rows correspond to labels
and columns to samples. This is an array of {1,-1}.
y : float array
Testing labels where rows correspond to labels
and columns to samples. This is an array of {1,-1}.
Returns
performance : float array
Array containing the AUC for training data, classification
accuracy for training data, AUC for testing data and
classification accuracy for testing data, in that order.
.. note::
* For binary-class classification, AUC is proportional to the Mann-Whitney U test statistic which computes a measure of the separation between values of positive labels and negative labels.
* For multi-class classification, this formula for computing classifier AUC is one of many. A more principled way would involve computing the Volume under an ROC surface.
"""
# BINARY CASE
if (Y.shape[0] == 2):
# computing train AUC
NP = (Y[0, :] == 1).sum()
NM = (Y[0, :] == -1).sum()
try:
U = stats.mannwhitneyu(train[0, (Y[0, :] == 1)], train[0, (Y[0, :] == -1)])
train_auc = 1. - U[0] / (NP * NM)
except ValueError:
# if all function outputs are equal, AUC = 0.5
train_auc = 0.5
# computing test AUC
NP = (y[0, :] == 1).sum()
NM = (y[0, :] == -1).sum()
try:
U = stats.mannwhitneyu(test[0, (y[0, :] == 1)], test[0, (y[0, :] == -1)])
test_auc = 1. - U[0] / (NP * NM)
except ValueError:
# if all function outputs are equal, AUC = 0.5
test_auc = 0.5
# MULTICLASS CASE
else:
# computing train AUC
NP = (Y == 1).sum()
NM = (Y == -1).sum()
try:
U = stats.mannwhitneyu(train[(Y == 1)], train[(Y == -1)])
train_auc = 1. - U[0] / (NP * NM)
except ValueError:
# if all function outputs are equal, AUC = 0.5
train_auc = 0.5
# computing test AUC
NP = (y == 1).sum()
NM = (y == -1).sum()
try:
U = stats.mannwhitneyu(test[(y == 1)], test[(y == -1)])
test_auc = 1. - U[0] / (NP * NM)
except ValueError:
# if all function outputs are equal, AUC = 0.5
test_auc = 0.5
# accuracy = number of examples where argmax of prediction
# equals true label
# train accuracy
train_accuracy = (train.argmax(0) - Y.argmax(0) == 0).sum() / float(Y.shape[1])
# test accuracy
test_accuracy = (test.argmax(0) - y.argmax(0) == 0).sum() / float(y.shape[1])
# train exploss
train_loss = np.sum(np.exp(-1 * train * Y))
# test exploss
test_loss = np.sum(np.exp(-1 * test * y))
return np.array([train_loss, train_auc, train_accuracy, test_loss, test_auc, test_accuracy])
def plotExpBox_Main(inputFiles,headers,valcols,outputFile,sep,startRow,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,plotPvalueCluster,outputClusterPrefix,methodCluster,xlegendrotation,xlabe,ylabe,figsz,titl,showSampleSizes,trimToMinSize,relabels,logb,plotHistogramToFile,plotMedianForGroups,botta):
#if plotPvalueCluster:
#if pvalue cluster is needed:
# from Bio.Cluster.cluster import *
# from Bio.Cluster import *
#endif
#the real deal!
plotData=[]
xtickLabels=[]
minSize=-1
for inputFile,header,cols in zip(inputFiles,headers,valcols):
fin=generic_istream(inputFile)
startIdx=len(plotData)
for col in cols:
plotData.append([])
xtickLabels.append(header[col])
colIndices=range(startIdx,startIdx+len(cols))
lino=0
for lin in fin:
lino+=1
if lino<startRow:
continue
fields=lin.rstrip("\r\n").split(sep)
for idx,col in zip(colIndices,cols):
try:
value=float(fields[col])
if logb!=0:
if value==0.0:
raise ValueError
value=log(value)/logb
plotData[idx].append(value)
except:
pass
fin.close()
if minSize==-1:
minSize=len(plotData[idx]) #or startIDX?
else:
minSize=min([minSize,len(plotData[idx])])
if trimToMinSize:
print >> stderr,"trimming to min size =",minSize
trimData(plotData,minSize)
if len(relabels)>0:
#if len(relabels)!=len(xtickLabels):
# print >> stderr,"relabels doesn't have the same length as original label vectors",xtickLabels,"=>",relabels
# exit()
print >> stderr,xtickLabels
print >> stderr,relabels
for i,relabel in zip(range(0,len(relabels)),relabels):
xtickLabels[i]=relabel
for i in range(0,len(plotMedianForGroups)):
plotMedianForGroups[i]=getCol0ListFromCol1ListStringAdv(xtickLabels,plotMedianForGroups[i])
#drawing medians:
medianToDraw=[]
for mediangrouper in plotMedianForGroups:
curD=[]
for c in mediangrouper:
curD.extend(plotData[c])
medianToDraw.append(median(curD))
for c in range(len(plotData)-1,-1,-1):
if len(plotData[c])==0:
print >> stderr,xtickLabels[c],"discarded"
del plotData[c]
del xtickLabels[c]
print >> stdout,"student t-test (1 sample; mean=0)"
print >> stdout,"sample","mean","p-val"
for x in range(0,len(plotData)):
#print >> stderr, len(plotData[x])
try:
print >> stdout, xtickLabels[x],mean(plotData[x]),ttest_1samp(plotData[x],0)[1]
except:
print >> stdout, xtickLabels[x],"NA","NA"
pvalueM=[]
print >> stdout,""
print >> stdout,"student t-test (2 samples)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
try:
pvalue=ttest_ind(plotData[x],plotData[y])[1]
except:
pvalue=1.0
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
print >> stdout,""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_t_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_t",xtickLabels,pvalueM,methodCluster)
pvalueM=[]
print >> stdout,"welch t-test"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
pvalue=welchs_approximate_ttest_arr(plotData[x],plotData[y])[3]
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_Welch_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_Welch",xtickLabels,pvalueM,methodCluster)
print >> stdout,""
print >> stdout,"non-parametric (Mann-Whitney U)" #"non-parametric (Mann-Whitney U if larger n<=20 else Wilcoxon)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue=mannwhitneyu(plotData[x],plotData[y])[1]*2
except:
pvalue=1.0
print >> stdout,pvalue, #mann-whiteney need to mul by 2 (one tail to two tail)
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_U_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_U",xtickLabels,pvalueM,methodCluster)
figure(figsize=figsz)
subplots_adjust(top=0.9, bottom=botta, left=0.2, right=0.8)
if len(titl)==0:
titl=outputFile
plotExpBox(plotData,xtickLabels,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,xlegendrotation,xlabe,ylabe,titl,showSampleSizes)
#ylim([0,200])
for m in medianToDraw:
axhline(y=m,linestyle=':',color='gray')
savefig(outputFile,bbox_inches="tight")
if len(plotHistogramToFile)>0:
drawHistogram(plotHistogramToFile,plotData,xtickLabels)
def plotExpBox_Main(inputFiles,headers,valcols,outputFile,sep,startRow,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,plotPvalueCluster,outputClusterPrefix,methodCluster,xlegendrotation,xlabe,ylabe,figsz,titl,showSampleSizes,trimToMinSize,relabels,logb,plotHistogramToFile,plotMedianForGroups,botta):
#if plotPvalueCluster:
#if pvalue cluster is needed:
# from Bio.Cluster.cluster import *
# from Bio.Cluster import *
#endif
#the real deal!
plotData=[]
xtickLabels=[]
minSize=-1
for inputFile,header,cols in zip(inputFiles,headers,valcols):
fin=generic_istream(inputFile)
startIdx=len(plotData)
for col in cols:
plotData.append([])
xtickLabels.append(header[col])
colIndices=range(startIdx,startIdx+len(cols))
lino=0
for lin in fin:
lino+=1
if lino<startRow:
continue
fields=lin.rstrip("\r\n").split(sep)
for idx,col in zip(colIndices,cols):
try:
value=float(fields[col])
if logb!=0:
value=log(value)/logb
if value<-100000:
raise ValueError
plotData[idx].append(value)
except:
pass
fin.close()
if minSize==-1:
minSize=len(plotData[idx]) #or startIDX?
else:
minSize=min([minSize,len(plotData[idx])])
if trimToMinSize:
print >> stderr,"trimming to min size =",minSize
trimData(plotData,minSize)
if len(relabels)>0:
#if len(relabels)!=len(xtickLabels):
# print >> stderr,"relabels doesn't have the same length as original label vectors",xtickLabels,"=>",relabels
# exit()
for i,relabel in zip(range(0,len(relabels)),relabels):
xtickLabels[i]=relabel
for i in range(0,len(plotMedianForGroups)):
plotMedianForGroups[i]=getCol0ListFromCol1ListStringAdv(xtickLabels,plotMedianForGroups[i])
#drawing medians:
medianToDraw=[]
for mediangrouper in plotMedianForGroups:
curD=[]
for c in mediangrouper:
curD.extend(plotData[c])
medianToDraw.append(median(curD))
for c in range(len(plotData)-1,-1,-1):
if len(plotData[c])==0:
print >> stderr,xtickLabels[c],"discarded"
del plotData[c]
del xtickLabels[c]
print >> stdout,"student t-test (1 sample; mean=0)"
print >> stdout,"sample","mean","p-val"
for x in range(0,len(plotData)):
#print >> stderr, len(plotData[x])
try:
print >> stdout, xtickLabels[x],mean(plotData[x]),ttest_1samp(plotData[x],0)[1]
except:
print >> stdout, xtickLabels[x],"NA","NA"
pvalueM=[]
print >> stdout,""
print >> stdout,"student t-test (2 samples)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
try:
pvalue=ttest_ind(plotData[x],plotData[y])[1]
except:
pvalue=1.0
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
print >> stdout,""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_t_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_t",xtickLabels,pvalueM,methodCluster)
pvalueM=[]
print >> stdout,"welch t-test"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
pvalue=welchs_approximate_ttest_arr(plotData[x],plotData[y])[3]
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_Welch_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_Welch",xtickLabels,pvalueM,methodCluster)
print >> stdout,""
print >> stdout,"non-parametric (Mann-Whitney U)" #"non-parametric (Mann-Whitney U if larger n<=20 else Wilcoxon)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue=mannwhitneyu(plotData[x],plotData[y])[1]*2
except:
pvalue=1.0
print >> stdout,pvalue, #mann-whiteney need to mul by 2 (one tail to two tail)
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_U_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_U",xtickLabels,pvalueM,methodCluster)
figure(figsize=figsz)
subplots_adjust(top=0.9, bottom=botta, left=0.2, right=0.8)
if len(titl)==0:
titl=outputFile
plotExpBox(plotData,xtickLabels,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,xlegendrotation,xlabe,ylabe,titl,showSampleSizes)
#ylim([0,200])
for m in medianToDraw:
axhline(y=m,linestyle=':',color='gray')
savefig(outputFile,bbox_inches="tight")
if len(plotHistogramToFile)>0:
drawHistogram(plotHistogramToFile,plotData,xtickLabels)
def plotExpBox_Main(inputFiles,headers,valcols,outputFile,sep,startRow,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,plotPvalueCluster,outputClusterPrefix,methodCluster,xlegendrotation,xlabe,ylabe,figsz,titl,showSampleSizes,trimToMinSize,relabels):
#if plotPvalueCluster:
#if pvalue cluster is needed:
# from Bio.Cluster.cluster import *
# from Bio.Cluster import *
#endif
#the real deal!
plotData=[]
xtickLabels=[]
minSize=-1
for inputFile,header,cols in zip(inputFiles,headers,valcols):
fin=generic_istream(inputFile)
startIdx=len(plotData)
for col in cols:
plotData.append([])
xtickLabels.append(header[col])
colIndices=range(startIdx,startIdx+len(cols))
lino=0
for lin in fin:
lino+=1
if lino<startRow:
continue
fields=lin.rstrip("\r\n").split(sep)
for idx,col in zip(colIndices,cols):
try:
value=float(fields[col])
plotData[idx].append(value)
except:
pass
fin.close()
if minSize==-1:
minSize=len(plotData[idx]) #or startIDX?
else:
minSize=min([minSize,len(plotData[idx])])
if trimToMinSize:
print >> stderr,"trimming to min size =",minSize
trimData(plotData,minSize)
if len(relabels)>0:
if len(relabels)!=len(xtickLabels):
print >> stderr,"relabels doesn't have the same length as original label vectors",xtickLabels,"=>",relabels
exit()
xtickLabels=relabels
print >> stdout,"student t-test (1 sample; mean=0)"
print >> stdout,"sample","mean","p-val"
for x in range(0,len(plotData)):
#print >> stderr, len(plotData[x])
print >> stdout, xtickLabels[x],mean(plotData[x]),ttest_1samp(plotData[x],0)[1]
pvalueM=[]
print >> stdout,""
print >> stdout,"student t-test (2 samples)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
pvalue=ttest_ind(plotData[x],plotData[y])[1]
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
print >> stdout,""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_t_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_t",xtickLabels,pvalueM,methodCluster)
pvalueM=[]
print >> stdout,"welch t-test"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
pvalue=welchs_approximate_ttest_arr(plotData[x],plotData[y])[3]
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_Welch_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_Welch",xtickLabels,pvalueM,methodCluster)
print >> stdout,""
print >> stdout,"non-parametric (Mann-Whitney U)" #"non-parametric (Mann-Whitney U if larger n<=20 else Wilcoxon)"
print >> stdout,"p-val",
for x in range(0,len(plotData)):
print >> stdout,xtickLabels[x],
pvalueM=[]
print >> stdout,""
for x in range(0,len(plotData)):
pvalueRow=[]
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0,len(plotData)):
if y<=x:
print >> stdout, "",
if x==y:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
pvalue=mannwhitneyu(plotData[x],plotData[y])[1]*2
print >> stdout,pvalue, #mann-whiteney need to mul by 2 (one tail to two tail)
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout,"";
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix+"_U_raw",xtickLabels,pvalueM)
makePValueClusterPlot(outputClusterPrefix+"_U",xtickLabels,pvalueM,methodCluster)
figure(figsize=figsz)
if len(titl)==0:
titl=outputFile
plotExpBox(plotData,xtickLabels,showIndPoints,mark,markMean,showMean,notch,whisker,outliers,xlegendrotation,xlabe,ylabe,titl,showSampleSizes)
#ylim([0,200])
savefig(outputFile,bbox_inches="tight")
import numpy
import pandas
import datetime
import scipy.stats.stats as st
# import matplotlib.pyplot as plt
format = "%Y_%m_%d"
current_date=datetime.datetime.today()
print 'dataset path: /Volumes/daelsaid/inquisit/INQUISIT_RERUN/all_launches_dir/scored/clean_copies/scored_data_2017_07_01_comb_mturk.csv'
print ' '
print current_date
scored_data=pandas.read_csv('/Volumes/daelsaid/inquisit/INQUISIT_RERUN/all_launches_dir/scored/clean_copies/scored_data_2017_07_01_newageranges_comb_mturk.csv')
y=['trial1', 'trial2', 'trial3', 'trial4', 'trial5', 'listb', 'trial6', 'trial7']
test_df=pandas.DataFrame(data=scored_data.set_index(['gender', 'age_range']).loc[:,'trial1':'trial7'])
for col in y:
print ' '
print '*******' + col + '*******'
for idx,val in test_df.groupby(level=1):
if len(val) >3:
u,p= st.mannwhitneyu(val.loc['female'][col],val.loc['male'][col], use_continuity=True,alternative='two-sided')
if p < 0.05:
print 'age_range ' + list(set(val.loc['female'][col].index.tolist()))[0] + ':', u, p, '**'
else:
print 'age_range ' + list(set(val.loc['female'][col].index.tolist()))[0] + ':', u, p, 'ns'
else:
print 'age_range ' + list(set(val.loc['female'][col].index.tolist()))[0]+ ': n is less than 3'
)
ax['circ'].scatter(df_py.jitter, df_py.vector_strength, edgecolors='w', lw=.5,
color=sns.xkcd_rgb['dark fuchsia'], \
label='pyramidal', s=point_size
)
ax['vs_freq_beat'].scatter(df_pu_b.frequency, df_pu_b.vector_strength, edgecolors='w', lw=.5,
color=sns.xkcd_rgb['azure'], \
label='p-units', s=point_size)
ax['vs_freq_beat'].scatter(df_py_b.frequency, df_py_b.vector_strength, edgecolors='w', lw=.5,
color=sns.xkcd_rgb['dark fuchsia'], \
label='pyramidal', s=point_size)
# --- circular variance scatter plots
ax['circ_beat'].scatter(df_pu_b.jitter, df_pu_b.vector_strength, edgecolors='w', lw=.5,
color=sns.xkcd_rgb['azure'], \
label='p-units', s=point_size
)
ax['circ_beat'].scatter(df_py_b.jitter, df_py_b.vector_strength, edgecolors='w', lw=.5,
color=sns.xkcd_rgb['dark fuchsia'], \
label='pyramidal', s=point_size
)
print(stats.mannwhitneyu(df_pu.vector_strength, df_py.vector_strength, use_continuity=True, alternative=None))
print(stats.mannwhitneyu(df_pu_b.vector_strength, df_py_b.vector_strength, use_continuity=True, alternative=None))
print(np.median(df_pu.vector_strength))
print(np.median(df_py.vector_strength))
print(np.median(df_pu_b.vector_strength))
print(np.median(df_py_b.vector_strength))
def stat_test(df,df_num):
for i in df_num.columns:
df_1=df[df['Response']==1][i]
df_0=df[df['Response']==0][i]
tsats,pval=stats.ttest_ind(df_1,df_0)
tstas,pval=stats.mannwhitneyu(df_1,df_0)
def compute_auc(train, test, Y, y):
"""Computes measures of accuracy for train and test data.
Computes the ROC curve and the area under that curve, as a
measure of classification accuracy. The threshold corresponding
to the point on the ROC curve farthest from `y=x` line is selected
and fraction of correct predictions corresponding to that
threshold is returned.
Arguments
train : float array
Array of predictions of the model on training data where rows
correspond to labels and columns correspond to samples.
test : float array
Array of predictions of the model on testing data where rows
correspond to labels and columns correspond to samples.
Y : float array
Training labels where rows correspond to labels
and columns to samples. This is an array of {1,-1}.
y : float array
Testing labels where rows correspond to labels
and columns to samples. This is an array of {1,-1}.
Returns
performance : float array
Array containing the AUC for training data, classification
accuracy for training data, AUC for testing data and
classification accuracy for testing data, in that order.
.. note::
* For binary-class classification, AUC is proportional to the Mann-Whitney U test statistic which computes a measure of the separation between values of positive labels and negative labels.
* For multi-class classification, this formula for computing classifier AUC is one of many. A more principled way would involve computing the Volume under an ROC surface.
"""
# computing train AUC
NP = (Y == 1).sum()
NM = (Y == -1).sum()
try:
U = stats.mannwhitneyu(train[(Y == 1)], train[(Y == -1)])
train_auc = 1. - U[0] / (NP * NM)
except ValueError:
train_auc = 0.5
# computing test AUC
NP = (y == 1).sum()
NM = (y == -1).sum()
try:
U = stats.mannwhitneyu(test[(y == 1)], test[(y == -1)])
test_auc = 1. - U[0] / (NP * NM)
except ValueError:
test_auc = 0.5
# accuracy = number of examples where argmax of prediction
# equals true label
# train accuracy
train_accuracy = (train.argmax(0) - Y.argmax(0) == 0).sum() / float(
Y.shape[1])
# test accuracy
test_accuracy = (test.argmax(0) - y.argmax(0) == 0).sum() / float(
y.shape[1])
return np.array([train_auc, train_accuracy, test_auc, test_accuracy])
def plotExpBox_Main(inputFiles, headers, valcols, outputFile, sep, startRow,
showIndPoints, mark, markMean, showMean, notch, whisker,
outliers, plotPvalueCluster, outputClusterPrefix,
methodCluster, xlegendrotation, xlabe, ylabe, figsz, titl,
showSampleSizes, trimToMinSize, relabels, logb,
plotHistogramToFile, plotMedianForGroups, botta,
showViolin, showBox, firstColAnnot, plotTrend, showLegend,
makePzfxFile, makeBinMatrix, writeDataSummaryStat,
summaryStatRange, minuslog10pvalue, minNDataToKeep,
vfacecolor, valpha, outXYZPvalues, dividePlots):
#if plotPvalueCluster:
#if pvalue cluster is needed:
# from Bio.Cluster.cluster import *
# from Bio.Cluster import *
#endif
#the real deal!
plotData = []
xtickLabels = []
trendData = {}
annot = {}
minSize = -1
for inputFile, header, cols in zip(inputFiles, headers, valcols):
fin = generic_istream(inputFile)
startIdx = len(plotData)
if firstColAnnot:
colAnnot = cols[0]
cols = cols[1:]
annotThisFile = []
annot[startIdx] = annotThisFile
else:
colAnnot = -1
annotThisFile = None
for col in cols:
plotData.append([])
xtickLabels.append(header[col])
colIndices = range(startIdx, startIdx + len(cols))
if plotTrend:
#print >> stderr,"plotTrend"
trendDataThisFile = []
trendData[startIdx] = trendDataThisFile
else:
trendDataThisFile = None
lino = 0
for lin in fin:
lino += 1
if lino < startRow:
continue
fields = lin.rstrip("\r\n").split(sep)
if plotTrend:
#print >> stderr,"a"
trendDataThisLine = []
else:
trendDataThisLine = None
allDataOKThisLine = True
if colAnnot >= 0:
annotThisFile.append(fields[colAnnot])
for idx, col in zip(colIndices, cols):
try:
value = float(fields[col])
if logb != 0:
if value == 0.0:
raise ValueError
value = log(value) / logb
plotData[idx].append(value)
if plotTrend:
trendDataThisLine.append(value)
#print >> stderr,"value:",value
except:
allDataOKThisLine = False
if plotTrend:
if allDataOKThisLine:
trendDataThisFile.append(trendDataThisLine)
else:
trendDataThisFile.append(None)
fin.close()
if minSize == -1:
minSize = len(plotData[idx]) #or startIDX?
else:
minSize = min([minSize, len(plotData[idx])])
if trimToMinSize:
print >> stderr, "trimming to min size =", minSize
trimData(plotData, minSize)
if len(relabels) > 0:
#if len(relabels)!=len(xtickLabels):
# print >> stderr,"relabels doesn't have the same length as original label vectors",xtickLabels,"=>",relabels
# exit()
print >> stderr, xtickLabels
print >> stderr, relabels
for i, relabel in zip(range(0, len(relabels)), relabels):
xtickLabels[i] = relabel
for i in range(0, len(plotMedianForGroups)):
plotMedianForGroups[i] = getCol0ListFromCol1ListStringAdv(
xtickLabels, plotMedianForGroups[i])
#drawing medians:
medianToDraw = []
for mediangrouper in plotMedianForGroups:
curD = []
for c in mediangrouper:
curD.extend(plotData[c])
medianToDraw.append(median(curD))
for c in range(len(plotData) - 1, -1, -1):
if len(plotData[c]) < minNDataToKeep:
print >> stderr, xtickLabels[c], "discarded because has only", len(
plotData[c]), "data points <", minNDataToKeep
del plotData[c]
del xtickLabels[c]
if not skipStat:
print >> stdout, "student t-test (1 sample; mean=0)"
print >> stdout, "sample", "mean", "p-val", "median"
if writeDataSummaryStat:
fDSS = open(writeDataSummaryStat, "w")
print >> fDSS, "sample\tmean\tvar\tsd\tmin\tmax\tN\tNInRange[" + str(
summaryStatRange[0]) + "," + str(
summaryStatRange[1]
) + "]\t%NInRange\tNbelowRange\t%Nbelow\tNAboveRange\t%NAbove"
for x in range(0, len(plotData)):
#print >> stderr, len(plotData[x])
try:
print >> stdout, xtickLabels[x], mean(
plotData[x]), ttest_1samp(plotData[x],
0)[1], median(plotData[x])
except:
print >> stdout, xtickLabels[x], mean(
plotData[x]), "NA", median(plotData[x])
if writeDataSummaryStat:
sumData, N, NIN, NBelow, NAbove = filterDataInRangeInclusive(
plotData[x], summaryStatRange[0], summaryStatRange[1])
if NIN > 1:
#print >> stderr,"sumData=",sumData
#print >> stderr,mean
mea = mean2(sumData)
DDOF = 1
sd = std(sumData, ddof=DDOF)
var = sd * sd
mi = min(sumData)
ma = max(sumData)
else:
mea = "NA"
sd = "NA"
var = "NA"
mi = "NA"
ma = "NA"
print >> fDSS, xtickLabels[x] + "\t" + str(mea) + "\t" + str(
var) + "\t" + str(sd) + "\t" + str(mi) + "\t" + str(
ma) + "\t" + str(N) + "\t" + str(NIN) + "\t" + str(
float(NIN) * 100 /
N) + "\t" + str(NBelow) + "\t" + str(
float(NBelow) * 100 /
N) + "\t" + str(NAbove) + "\t" + str(
float(NAbove) * 100 / N)
pvalueM = []
if writeDataSummaryStat:
fDSS.close()
print >> stdout, ""
print >> stdout, "student t-test (2 samples)"
print >> stdout, "p-val",
for x in range(0, len(plotData)):
print >> stdout, xtickLabels[x],
print >> stdout, ""
for x in range(0, len(plotData)):
pvalueRow = []
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0, len(plotData)):
if y <= x:
print >> stdout, "",
if x == y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
try:
pvalue = ttest_ind(plotData[x], plotData[y])[1]
except:
pvalue = 1.0
if minuslog10pvalue and str(pvalue) != "NA":
try:
pvalue = -1 * log(pvalue, 10)
except:
pvalue = -1000.0
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout, ""
print >> stdout, ""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix + "_t_raw", xtickLabels,
pvalueM)
makePValueClusterPlot(outputClusterPrefix + "_t", xtickLabels,
pvalueM, methodCluster)
pvalueM = []
print >> stdout, "welch t-test"
print >> stdout, "p-val",
for x in range(0, len(plotData)):
print >> stdout, xtickLabels[x],
print >> stdout, ""
for x in range(0, len(plotData)):
pvalueRow = []
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0, len(plotData)):
if y <= x:
print >> stdout, "",
if x == y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
try:
pvalue = welchs_approximate_ttest_arr(
plotData[x], plotData[y])[3]
except:
pvalue = 1.0
if minuslog10pvalue and str(pvalue) != "NA":
try:
pvalue = -1 * log(pvalue, 10)
except:
pvalue = -1000.0
print >> stdout, str(pvalue),
pvalueRow.append(pvalue)
print >> stdout, ""
if outXYZPvalues:
writeXYZPvalues(outXYZPvalues + "_Welch.xyz", xtickLabels, pvalueM)
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix + "_Welch_raw", xtickLabels,
pvalueM)
makePValueClusterPlot(outputClusterPrefix + "_Welch", xtickLabels,
pvalueM, methodCluster)
print >> stdout, ""
print >> stdout, "non-parametric (Mann-Whitney U)" #"non-parametric (Mann-Whitney U if larger n<=20 else Wilcoxon)"
print >> stdout, "p-val",
for x in range(0, len(plotData)):
print >> stdout, xtickLabels[x],
pvalueM = []
print >> stdout, ""
for x in range(0, len(plotData)):
pvalueRow = []
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0, len(plotData)):
if y <= x:
print >> stdout, "",
if x == y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue = mannwhitneyu(plotData[x], plotData[y])[1] * 2
except:
pvalue = 1.0
if minuslog10pvalue and str(pvalue) != "NA":
try:
pvalue = -1 * log(pvalue, 10)
except:
pvalue = -1000.0
print >> stdout, pvalue, #mann-whiteney need to mul by 2 (one tail to two tail)
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout, ""
if outXYZPvalues:
writeXYZPvalues(outXYZPvalues + "_U.xyz", xtickLabels, pvalueM)
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix + "_U_raw", xtickLabels,
pvalueM)
makePValueClusterPlot(outputClusterPrefix + "_U", xtickLabels,
pvalueM, methodCluster)
#####now the variance tests
print >> stdout, ""
print >> stdout, "Ansari-Bradley Two-sample Test for difference in scale parameters "
print >> stdout, "p-val",
for x in range(0, len(plotData)):
print >> stdout, xtickLabels[x],
pvalueM = []
print >> stdout, ""
for x in range(0, len(plotData)):
pvalueRow = []
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0, len(plotData)):
if y <= x:
print >> stdout, "",
if x == y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue = ansari(plotData[x], plotData[y])[1]
except:
pvalue = "NA"
if minuslog10pvalue and str(pvalue) != "NA":
try:
pvalue = -1 * log(pvalue, 10)
except:
pvalue = -1000.0
#pvalue=1.0
print >> stdout, pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout, ""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix + "_Ansari_raw", xtickLabels,
pvalueM)
makePValueClusterPlot(outputClusterPrefix + "_Ansari", xtickLabels,
pvalueM, methodCluster)
#####
#####now the variance tests
print >> stdout, ""
print >> stdout, "Fligner's Two-sample Test for equal variance (non-parametrics)"
print >> stdout, "p-val",
for x in range(0, len(plotData)):
print >> stdout, xtickLabels[x],
pvalueM = []
print >> stdout, ""
for x in range(0, len(plotData)):
pvalueRow = []
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0, len(plotData)):
if y <= x:
print >> stdout, "",
if x == y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue = fligner(plotData[x], plotData[y])[1]
except:
pvalue = "NA"
#pvalue=1.0
if minuslog10pvalue and str(pvalue) != "NA":
try:
pvalue = -1 * log(pvalue, 10)
except:
pvalue = -1000.0
print >> stdout, pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout, ""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix + "_fligner_raw",
xtickLabels, pvalueM)
makePValueClusterPlot(outputClusterPrefix + "_fligner",
xtickLabels, pvalueM, methodCluster)
#####
#####now the variance tests
print >> stdout, ""
print >> stdout, "Levene's Two-sample Test for equal variance"
print >> stdout, "p-val",
for x in range(0, len(plotData)):
print >> stdout, xtickLabels[x],
pvalueM = []
print >> stdout, ""
for x in range(0, len(plotData)):
pvalueRow = []
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0, len(plotData)):
if y <= x:
print >> stdout, "",
if x == y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue = levene(plotData[x], plotData[y])[1]
except:
pvalue = "NA"
#pvalue=1.0
if minuslog10pvalue and str(pvalue) != "NA":
try:
pvalue = -1 * log(pvalue, 10)
except:
pvalue = -1000.0
print >> stdout, pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout, ""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix + "_levene_raw", xtickLabels,
pvalueM)
makePValueClusterPlot(outputClusterPrefix + "_levene", xtickLabels,
pvalueM, methodCluster)
#####
#####now the variance tests
print >> stdout, ""
print >> stdout, "Bartlett's Two-sample Test for equal variance (for normal distributions)"
print >> stdout, "p-val",
for x in range(0, len(plotData)):
print >> stdout, xtickLabels[x],
pvalueM = []
print >> stdout, ""
for x in range(0, len(plotData)):
pvalueRow = []
pvalueM.append(pvalueRow)
print >> stdout, xtickLabels[x],
for y in range(0, len(plotData)):
if y <= x:
print >> stdout, "",
if x == y:
if minuslog10pvalue:
pvalueRow.append(0.0)
else:
pvalueRow.append(1.0)
else:
pvalueRow.append(pvalueM[y][x])
else:
#if max(len(plotData[x]),len(plotData[y]))<=20:
try:
pvalue = bartlett(plotData[x], plotData[y])[1]
except:
pvalue = "NA"
#pvalue=1.0
if minuslog10pvalue and str(pvalue) != "NA":
try:
pvalue = -1 * log(pvalue, 10)
except:
pvalue = -1000.0
print >> stdout, pvalue,
pvalueRow.append(pvalue)
#else:
# print >> stdout,wilcoxon(plotData[x],plotData[y])[1], # this is two-tailed already stdout, "", #
print >> stdout, ""
if plotPvalueCluster:
makePValueRawPlot(outputClusterPrefix + "_bartlett_raw",
xtickLabels, pvalueM)
makePValueClusterPlot(outputClusterPrefix + "_bartlett",
xtickLabels, pvalueM, methodCluster)
#####
figure(figsize=figsz)
subplots_adjust(top=0.9, bottom=botta, left=0.2, right=0.8)
if len(titl) == 0:
titl = outputFile
plotExpBox(plotData, xtickLabels, showIndPoints, mark, markMean, showMean,
notch, whisker, outliers, xlegendrotation, xlabe, ylabe, titl,
showSampleSizes, showViolin, showBox, annot, trendData,
showLegend, makePzfxFile, makeBinMatrix, dividePlots)
#ylim([0,200])
for m in medianToDraw:
axhline(y=m, linestyle=':', color='gray')
savefig(outputFile, bbox_inches="tight")
if len(plotHistogramToFile) > 0:
drawHistogram(plotHistogramToFile, plotData, xtickLabels)
drawDensigram(plotHistogramToFile + ".density.png", plotData,
xtickLabels)
