def plot_roc_curves_with_ps(Y,Y_cv,Xs,p0,p1,p,pathological,diagnoses=None):
if diagnoses is None:
diagnoses = pathological.columns.values
fig,ax = plt.subplots(len(diagnoses),4)
sns.set_context("notebook", font_scale=1,
rc={"lines.linewidth": 1.5,
'legend.fontsize': 12})
fig.set_size_inches(15,2*len(diagnoses))
for i in range(len(diagnoses)):
diagnosis = diagnoses[i]
ix = list(pathological.columns.values).index(diagnosis)
if diagnoses is not None and diagnosis not in diagnoses:
continue
for j,key in enumerate(['basic','total','all']):
X = Xs[key]
if len(p1[key][ix]) and len(p0[key][ix]):
ax[i,j].hist(p1[key][ix],bins=1000,range=(0,1),color='r',
normed=True,cumulative=True,histtype='step')
ax[i,j].hist(p0[key][ix],bins=1000,range=(0,1),color='k',
normed=True,cumulative=True,histtype='step')
ax[i,j].set_title(diagnosis)
if i==ax.shape[0]-1:
ax[i,j].set_xlabel('Predicted p(pathology)')
if j==0:
ax[i,j].set_ylabel('Cumulative fraction')
ax[i,j].set_xlim(0,1)
ax[i,j].set_ylim(0,1)
plot_roc_curves({key:Y_cv[key][ix] for key in Y_cv},
{key:p[key][ix] for key in p},
ax=ax[i,3],label='sparse')
fig.tight_layout()
def plot_roc_curves_with_ps(Y,Y_cv,Xs,p0,p1,p,pathological,diagnoses=None):
if diagnoses is None:
diagnoses = pathological.columns.values
fig,ax = plt.subplots(len(diagnoses),4)
sns.set_context("notebook", font_scale=1,
rc={"lines.linewidth": 1.5,
'legend.fontsize': 12})
fig.set_size_inches(15,2*len(diagnoses))
for i in range(len(diagnoses)):
diagnosis = diagnoses[i]
ix = list(pathological.columns.values).index(diagnosis)
if diagnoses is not None and diagnosis not in diagnoses:
continue
for j,key in enumerate(['basic','total','all']):
X = Xs[key]
if len(p1[key][ix]) and len(p0[key][ix]):
ax[i,j].hist(p1[key][ix],bins=1000,range=(0,1),color='r',
normed=True,cumulative=True,histtype='step')
ax[i,j].hist(p0[key][ix],bins=1000,range=(0,1),color='k',
normed=True,cumulative=True,histtype='step')
ax[i,j].set_title(diagnosis)
if i==ax.shape[0]-1:
ax[i,j].set_xlabel('Predicted p(pathology)')
if j==0:
ax[i,j].set_ylabel('Cumulative fraction')
ax[i,j].set_xlim(0,1)
ax[i,j].set_ylim(0,1)
plot_roc_curves({key:Y_cv[key][ix] for key in Y_cv},
{key:p[key][ix] for key in p},
ax=ax[i,3],label='sparse')
fig.tight_layout()
def plot_roc_curves(Y,p,ax=None,label='full',title='AUC',color=None):
if ax is None:
fig,ax = plt.subplots(1,1)
colors = {'basic':'gray','total':'pink','all':'red'}
aucs = {}
for key in p:
fpr,tpr,auc = get_roc_curve(Y[key],p[key])
if color is None:
color = 'red' if key not in colors else colors[key]
ax.plot(fpr, tpr, lw=2, color=color,
label={'full':'AUC using\n%s = %0.2f' % (key,auc),
'sparse':'%s = %.2f' % (title,auc)}[label])
aucs[key] = auc
ax.set_xlim(-0.01,1.01)
ax.set_ylim(-0.01,1.01)
ax.set_xlabel('False Positive Rate')#, fontsize='large', fontweight='bold')
ax.set_ylabel('True Positive Rate')#, fontsize='large', fontweight='bold')
ax.set_title(title)#, fontsize='large', fontweight='bold')
# Set the tick labels font
for label in (ax.get_xticklabels() + ax.get_yticklabels()):
pass
#label.set_fontsize('large')
#label.set_fontweight('bold')
ax.legend(loc="lower right")
return aucs
def plot_just_rocs(Y_clean,ps,ys,imps):
fig,axes = plt.subplots(4,4,sharex=True,sharey=True,figsize=(10,10))
y = {}
p = {}
for i in range(len(list(Y_clean))):
ax = axes.flat[i]
for imp in imps:
pimp = ps[imp][i,:,:].ravel() # Unravel all predictions of test data over all cv splits.
yimp = ys[imp][i,:,:].ravel() # Unravel all test data ground truth over all cv splits.
pimp = pimp[np.isnan(yimp)==0] # Remove NaNs (no ground truth)
yimp = yimp[np.isnan(yimp)==0] # Remove NaNs (no ground truth)
p[imp] = pimp[yimp.mask==False]#.compressed()
y[imp] = yimp[yimp.mask==False]#.compressed()
plot_roc_curves(y,p,ax=ax) # Plot on the given axes.
ax.set_title(list(Y_clean)[i]) # Set the title.
for i in range(i+1,len(axes.flat)):
ax = axes.flat[i]
ax.set_axis_off()
plt.tight_layout()
def plot_just_rocs(props,ps,ys,imps,plot=True,axes=None,m=None,n=None,color='k',title='AUC'):
if plot:
if axes is None:
if m is None:
m = max(1,1+int(len(props)/4))
if n is None:
n = min(4,len(props))
fig,axes = plt.subplots(m,n,sharex=True,sharey=True,
squeeze=False,figsize=(m*3.5,n*3.5))
y = {}
p = {}
aucs = {}
for i,prop in enumerate(props):
if plot:
ax = axes.flat[i]
for imp in imps:
pimp = ps[imp][i,:,:].ravel() # Unravel all predictions of test data over all cv splits.
yimp = ys[imp][i,:,:].ravel() # Unravel all test data ground truth over all cv splits.
pimp = pimp[np.isnan(yimp)==0] # Remove NaNs (no ground truth)
yimp = yimp[np.isnan(yimp)==0] # Remove NaNs (no ground truth)
p[imp] = pimp[yimp.mask==False]#.compressed()
y[imp] = yimp[yimp.mask==False]#.compressed()
if plot:
aucs[prop] = plot_roc_curves(y,p,ax=ax,label='sparse',title=title,color=color) # Plot on the given axes.
else:
aucsi = {}
for imp in p:
fpr,tpr,auc = get_roc_curve(y[imp],p[imp])
aucsi[imp] = auc
aucs[prop] = aucsi
if plot:
ax.set_title(props[i].replace('Clinpath ','').replace('Nos','NOS')) # Set the title.
if plot:
for i in range(i+1,len(axes.flat)):
ax = axes.flat[i]
ax.set_axis_off()
plt.tight_layout()
return aucs,axes