def add_data_logistic(self, event):
if event.inaxes == self.ax[0]:
x0_coord = event.xdata
x1_coord = event.ydata
if event.key is None: #shift not pressed
self.ax[0].scatter(x0_coord,
x1_coord,
marker='x',
s=10,
c='red',
label="y=1")
self.y = np.append(self.y, 1)
else:
self.ax[0].scatter(x0_coord,
x1_coord,
marker='o',
s=10,
label="y=0",
facecolors='none',
edgecolors=dlc['dlblue'],
lw=3)
self.y = np.append(self.y, 0)
self.X = np.append(self.X,
np.array([[x0_coord, x1_coord]]),
axis=0)
self.fig.canvas.draw()
def mk_cost_lines(self, firsttime=False):
''' makes vertical cost lines'''
if not firsttime:
for artist in self.cost_items:
artist.remove()
self.cost_items = []
cstr = f"cost = (1/{self.m})*("
ctot = 0
label = 'cost for point'
addedbreak = False
for p in zip(self.x_train, self.y_train):
f_wb_p = sigmoid(self.w * p[0] + self.b)
c_p = compute_cost_matrix(p[0].reshape(-1, 1),
p[1],
np.array(self.w),
self.b,
logistic=True,
lambda_=0,
safe=True)
c_p_txt = c_p
a = self.ax.vlines(p[0],
p[1],
f_wb_p,
lw=3,
color=dlc["dlpurple"],
ls='dotted',
label=label)
label = '' #just one
cxy = [p[0], p[1] + (f_wb_p - p[1]) / 2]
b = self.ax.annotate(f'{c_p_txt:0.1f}',
xy=cxy,
xycoords='data',
color=dlc["dlpurple"],
xytext=(5, 0),
textcoords='offset points')
cstr += f"{c_p_txt:0.1f} +"
if len(cstr) > 38 and addedbreak is False:
cstr += "\n"
addedbreak = True
ctot += c_p
self.cost_items.extend((a, b))
ctot = ctot / (len(self.x_train))
cstr = cstr[:-1] + f") = {ctot:0.2f}"
## todo.. figure out how to get this textbox to extend to the width of the subplot
c = self.ax.text(0.05,
0.02,
cstr,
transform=self.ax.transAxes,
color=dlc["dlpurple"])
self.cost_items.append(c)
def update_contour_wb_lines(self, w, b, firsttime=False):
self.w = w
self.b = b
cst = compute_cost_matrix(self.x_train.reshape(-1, 1),
self.y_train,
np.array(self.w),
self.b,
logistic=True,
lambda_=0,
safe=True)
# remove lines and re-add on contour plot and 3d plot
if not firsttime:
for artist in self.dyn_items:
artist.remove()
a = self.axc.scatter(self.w,
self.b,
s=100,
color=dlc["dlblue"],
zorder=10,
label="cost with \ncurrent w,b")
b = self.axc.hlines(self.b,
self.axc.get_xlim()[0],
self.w,
lw=4,
color=dlc["dlpurple"],
ls='dotted')
c = self.axc.vlines(self.w,
self.axc.get_ylim()[0],
self.b,
lw=4,
color=dlc["dlpurple"],
ls='dotted')
d = self.axc.annotate(f"Cost: {cst:0.2f}",
xy=(self.w, self.b),
xytext=(4, 4),
textcoords='offset points',
bbox=dict(facecolor='white'),
size=10)
#Add point in 3D surface plot
e = self.axs.scatter3D(self.w, self.b, cst, marker='X', s=100)
self.dyn_items = [a, b, c, d, e]
def calc_logistic(self, event):
''' called on run gradient event '''
for it in [1, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096]:
w, self.b, J_hist = gradient_descent(self.x_train.reshape(-1, 1),
self.y_train.reshape(-1, 1),
np.array(self.w).reshape(
-1, 1),
self.b,
0.1,
it,
logistic=True,
lambda_=0,
verbose=False)
self.w = w[0, 0]
self.dplot.update(self.w, self.b)
self.con_plot.update_contour_wb_lines(self.w, self.b)
self.con_plot.path.add_path_item(self.w, self.b)
self.cplot.add_cost(J_hist)
time.sleep(0.3)
self.fig.canvas.draw()
def __init__(self, x_train, y_train, w_range, b_range):
# setup figure
fig = plt.figure(figsize=(10, 6))
fig.canvas.toolbar_visible = False
fig.canvas.header_visible = False
fig.canvas.footer_visible = False
fig.set_facecolor('#ffffff') #white
gs = GridSpec(2, 2, figure=fig)
ax0 = fig.add_subplot(gs[0, 0])
ax1 = fig.add_subplot(gs[0, 1])
ax2 = fig.add_subplot(gs[1, 0], projection='3d')
ax3 = fig.add_subplot(gs[1, 1])
pos = ax3.get_position().get_points() ##[[lb_x,lb_y], [rt_x, rt_y]]
h = 0.05
width = 0.2
axcalc = plt.axes([pos[1, 0] - width, pos[1, 1] - h, width,
h]) #lx,by,w,h
ax = np.array([ax0, ax1, ax2, ax3, axcalc])
self.fig = fig
self.ax = ax
self.x_train = x_train
self.y_train = y_train
self.w = 0. #initial point, non-array
self.b = 0.
# initialize subplots
self.dplot = data_plot(ax[0], x_train, y_train, self.w, self.b)
self.con_plot = contour_and_surface_plot(ax[1], ax[2], x_train,
y_train, w_range, b_range,
self.w, self.b)
self.cplot = cost_plot(ax[3])
# setup events
self.cid = fig.canvas.mpl_connect('button_press_event',
self.click_contour)
self.bcalc = Button(axcalc,
'Run Gradient Descent \nfrom current w,b (click)',
color=dlc["dlorange"])
self.bcalc.on_clicked(self.calc_logistic)
def plt_prob(ax, w_out, b_out):
""" plots a decision boundary but include shading to indicate the probability """
#setup useful ranges and common linspaces
x0_space = np.linspace(0, 4, 100)
x1_space = np.linspace(0, 4, 100)
# get probability for x0,x1 ranges
tmp_x0, tmp_x1 = np.meshgrid(x0_space, x1_space)
z = np.zeros_like(tmp_x0)
for i in range(tmp_x0.shape[0]):
for j in range(tmp_x1.shape[1]):
z[i, j] = sigmoid(
np.dot(w_out, np.array([tmp_x0[i, j], tmp_x1[i, j]])) + b_out)
cmap = plt.get_cmap('Blues')
new_cmap = truncate_colormap(cmap, 0.0, 0.5)
pcm = ax.pcolormesh(tmp_x0,
tmp_x1,
z,
norm=cm.colors.Normalize(vmin=0, vmax=1),
cmap=new_cmap,
shading='nearest',
alpha=0.9)
ax.figure.colorbar(pcm, ax=ax)