def class_2c_2f(thetas,threshold): """ Computes the classification map for a discrimination problem with 2 classes and 2 features. thetas is a dictionary of length = 1 (2 classes) and each independent theta vector is of length = 3 (2 features). """ from LR_functions import g pas = .01 x1 = np.arange(-1,1,pas) x2 = np.arange(-1,1,pas) x1, x2 = np.meshgrid(x1,x2) probas = g(thetas[1][0]+thetas[1][1]*x1+thetas[1][2]*x2) classi = probas.copy() classi[classi>=threshold[1]] = 1 classi[classi!=1] = 0 return x1, x2, probas, classi
def class_multi_2f(thetas):
"""
Computes the classification map for a discrimination problem
with more than 2 classes and only 2 features.
thetas is a dictionary of length = 3 (3 classes)
and each independent theta vector is of length = 3 (2 features).
"""
from LR_functions import g
pas = .01
x1 = np.arange(-1,1,pas)
x2 = np.arange(-1,1,pas)
x1, x2 = np.meshgrid(x1,x2)
probas = []
for i in sorted(thetas):
probas.append(g(thetas[i][0]+thetas[i][1]*x1+thetas[i][2]*x2))
probas = np.array(probas)
return x1, x2, np.max(probas,axis=0), np.argmax(probas,axis=0)
def class_multi_2f(thetas):
"""
Computes the classification map for a discrimination problem
with more than 2 classes and only 2 features.
thetas is a dictionary of length = 3 (3 classes)
and each independent theta vector is of length = 3 (2 features).
"""
from LR_functions import g
pas = .01
x1 = np.arange(-1, 1, pas)
x2 = np.arange(-1, 1, pas)
x1, x2 = np.meshgrid(x1, x2)
probas = []
for i in sorted(thetas):
probas.append(g(thetas[i][0] + thetas[i][1] * x1 + thetas[i][2] * x2))
probas = np.array(probas)
return x1, x2, np.max(probas, axis=0), np.argmax(probas, axis=0)
def class_2c_2f(thetas, threshold):
"""
Computes the classification map for a discrimination problem
with 2 classes and 2 features.
thetas is a dictionary of length = 1 (2 classes)
and each independent theta vector is of length = 3 (2 features).
"""
from LR_functions import g
pas = .01
x1 = np.arange(-1, 1, pas)
x2 = np.arange(-1, 1, pas)
x1, x2 = np.meshgrid(x1, x2)
probas = g(thetas[1][0] + thetas[1][1] * x1 + thetas[1][2] * x2)
classi = probas.copy()
classi[classi >= threshold[1]] = 1
classi[classi != 1] = 0
return x1, x2, probas, classi
def plot_hyp_func_1f(x,y,theta,method,threshold=None,x_ok=None,x_bad=None,th_comp=None,cmat_test=[],cmat_svm=[],cmat_train=[]):
num_t = np.unique(y.NumType.values.ravel())
num_t = map(int,list(num_t))
str_t = np.unique(y.Type.values.ravel())
str_t = map(str,list(str_t))
fig = plt.figure(figsize=(12,7))
fig.set_facecolor('white')
left, bottom = .1, .1
width, height = .5, .8
width_h, height_h = .25, .35
left_h, bottom_h = left+width+.05, bottom+height_h+.1
rect_1 = [left, bottom, width, height]
rect_2 = [left_h, bottom_h, width_h, height_h]
rect_3 = [left_h, bottom, width_h, height_h]
ax1 = plt.axes(rect_1)
ax2 = plt.axes(rect_2)
ax3 = plt.axes(rect_3)
x1 = x[y.NumType.values.ravel()==num_t[0]].values[:,0]
x2 = x[y.NumType.values.ravel()==num_t[1]].values[:,0]
nn,b,p = ax1.hist([x1,x2],25,normed=True,histtype='stepfilled',alpha=.2,color=('b','g'),label=str_t)
from LR_functions import g
# Plot the hypothesis function
syn = np.arange(-1,1,0.01)
hyp = g(theta[1][0]+theta[1][1]*syn)
norm = np.mean([np.max(nn[0]),np.max(nn[1])])
ax1.plot(syn,norm*hyp,'y-',lw=2,label='hypothesis')
if threshold:
thres = np.ones(len(hyp))*threshold[1]
imin = np.argmin(np.abs(thres-hyp))
t = syn[imin]
ax1.plot([t,t],[0,np.max(nn)],'orange',lw=3.,label='decision')
if th_comp:
hyp_svm = g(th_comp[1][0]+th_comp[1][1]*syn)
ax1.plot(syn,norm*hyp_svm,'magenta',lw=2.)
thres = np.ones(len(hyp_svm))*.5
imin = np.argmin(np.abs(thres-hyp_svm))
t = syn[imin]
ax1.plot([t,t],[0,np.max(nn)],'purple',lw=3.)
if x_ok and x_bad:
# plot well-classified events of the test set
#nn, b, p = ax1.hist([x_ok],25,normed=True,color=('k'),histtype='step',fill=False,ls='dashed',lw=2,label=['Test Set'])
# plot both well and badly-classified events of the test set
nn, b, p = ax1.hist([x_ok,x_bad],25,normed=True,color=('k','r'),histtype='step',fill=False,ls='dashed',lw=2,label=['Test Set'])
ax1.set_xlim([-1,1])
ax1.legend(prop={'size':12})
ax1.set_xlabel(x.columns[0])
ax1.set_title(x.columns[0])
from do_classification import plot_confusion_mat, dic_percent
s = 12
x_pos = .05
y_pos = .95
pas = .04
if list(cmat_test):
plot_confusion_mat(cmat_test,str_t,'Test',method,ax=ax3)
p_test = dic_percent(cmat_test,str_t)
ax1.text(x_pos,y_pos,method.upper(),size=s,transform=ax1.transAxes,color='orange')
ax1.text(x_pos,y_pos-pas,"Test : %.2f%%"%p_test['global'],size=s,transform=ax1.transAxes)
ax1.text(x_pos+.2,y_pos-pas,"(%.2f%%)"%(100-p_test['global']),size=s,color='red',transform=ax1.transAxes)
if list(cmat_train) and not list(cmat_svm):
plot_confusion_mat(cmat_train,str_t,'Training',method,ax=ax2)
p_train = dic_percent(cmat_train,str_t)
ax1.text(x_pos,y_pos-2*pas,"Training : %.2f%%"%p_train['global'],size=s,transform=ax1.transAxes)
elif list(cmat_train) and list(cmat_svm):
p_train = dic_percent(cmat_train,str_t)
ax1.text(x_pos,y_pos-2*pas,"Training : ",size=s,transform=ax1.transAxes)
ax1.text(x_pos+.15,y_pos-2*pas,"%s %s%%"%(str_t[0],p_train[(str_t[0],0)]),color='b',size=s,transform=ax1.transAxes)
ax1.text(x_pos+.15,y_pos-3*pas,"%s %s%%"%(str_t[1],p_train[(str_t[1],1)]),color='g',size=s,transform=ax1.transAxes)
plot_confusion_mat(cmat_svm,str_t,'Test','SVM',ax=ax2)
p_svm = dic_percent(cmat_svm,str_t)
ax1.text(x_pos,y_pos-4*pas,"SVM",size=s,transform=ax1.transAxes,color='purple')
ax1.text(x_pos,y_pos-5*pas,"Test : %.2f%%"%p_svm['global'],size=s,transform=ax1.transAxes)
elif not list(cmat_train) and list(cmat_svm):
plot_confusion_mat(cmat_svm,str_t,'Test','SVM',ax=ax2)
p_svm = dic_percent(cmat_svm,str_t)
ax1.text(x_pos,y_pos-3*pas,"SVM",size=s,transform=ax1.transAxes,color='purple')
ax1.text(x_pos,y_pos-4*pas,"Test : %.2f%%"%p_svm['global'],size=s,transform=ax1.transAxes)
plt.figtext(.1,.93,'(a)')
plt.figtext(.63,.93,'(b)')
plt.figtext(.63,.48,'(c)')
