def create_synthetic_flip_transfer(file_dir="", dim=1):
n_target = 100
n_source = 100
n = n_target + n_source
sigma = 0.2
data = data_class.Data()
data.x = np.random.uniform(0, 1, (n, dim))
data.data_set_ids = np.zeros(n)
data.data_set_ids[n_target:] = 1
data.y = np.zeros(n)
data.y[(data.data_set_ids == 0) & (data.x[:, 0] >= 0.5)] = 2
data.y[(data.data_set_ids == 1) & (data.x[:, 0] >= 0.5)] = 1
data.y[(data.data_set_ids == 0) & (data.x[:, 0] <= 0.5)] = 1
data.y[(data.data_set_ids == 1) & (data.x[:, 0] <= 0.5)] = 2
data.y += np.random.normal(0, sigma, n)
data.set_train()
data.set_true_y()
data.is_regression = True
if dim == 1:
array_functions.plot_2d(data.x, data.y, data.data_set_ids)
s = synthetic_flip_file
if dim > 1:
s = synthetic_step_kd_transfer_file % dim
if file_dir != "":
s = file_dir + "/" + s
helper_functions.save_object(s, data)
def plot_g(self):
x = np.linspace(0,1)
x = array_functions.vec_to_2d(x)
g_orig = self.g_learner.predict_g(x)
g = 1 / (1+g_orig)
array_functions.plot_2d(x,g)
pass
def create_synthetic_piecewise_transfer(file_dir='', dim=1):
n_target = 150
n_source = 150
n = n_target + n_source
sigma = .2
data = data_class.Data()
data.x = np.random.uniform(0, 1, (n, dim))
data.data_set_ids = np.zeros(n)
data.data_set_ids[n_target:] = 1
data.y = np.zeros(n)
data.y[(data.data_set_ids == 0) & (data.x[:, 0] <= 1)] = 1
data.y[(data.data_set_ids == 0) & (data.x[:, 0] <= .66)] = 0
data.y[(data.data_set_ids == 0) & (data.x[:, 0] <= .33)] = 2
data.y[(data.data_set_ids == 1) & (data.x[:, 0] <= 1)] = 0
data.y[(data.data_set_ids == 1) & (data.x[:, 0] <= .66)] = 2
data.y[(data.data_set_ids == 1) & (data.x[:, 0] <= .33)] = 1
data.y += np.random.normal(0, sigma, n)
data.set_train()
data.set_true_y()
data.is_regression = True
if dim == 1:
array_functions.plot_2d(data.x, data.y, data.data_set_ids)
s = synthetic_piecewise_file
if dim > 1:
s = synthetic_piecewise_file % dim
if file_dir != '':
s = file_dir + '/' + s
helper_functions.save_object(s, data)
def create_synthetic_delta_linear_transfer():
slope = 5
target_fun = lambda x: slope * x
source_fun = lambda x: slope * x + 4
data = create_synthetic_regression_transfer(target_fun, source_fun)
array_functions.plot_2d(data.x, data.y, data.data_set_ids, title="Linear Delta Data Set")
s = synthetic_delta_linear_file
helper_functions.save_object(s, data)
def plot_target(self):
x = np.linspace(0,1)
x = array_functions.vec_to_2d(x)
d = data_lib.Data()
d.x = x
d.y = np.nan*np.ones(x.shape[0])
d.is_regression = True
o = self.target_learner.predict(d)
array_functions.plot_2d(x, o.y)
def create_digits():
digits_data = datasets.load_digits()
x = digits_data.data
y = digits_data.target
for i in range(x.shape[1]):
xi = array_functions.normalize(x[:, i])
yi = y
array_functions.plot_2d(xi, yi, alpha=.01)
pass
pass
def create_diabetes():
diabetes_data = datasets.load_diabetes()
x = diabetes_data.data
y = diabetes_data.target
for i in range(x.shape[1]):
xi = array_functions.normalize(x[:, i])
yi = array_functions.normalize(y)
array_functions.plot_2d(xi, yi)
pass
assert False
def create_digits():
digits_data = datasets.load_digits()
x = digits_data.data
y = digits_data.target
for i in range(x.shape[1]):
xi = array_functions.normalize(x[:, i])
yi = y
array_functions.plot_2d(xi, yi, alpha=0.01)
pass
pass
def create_diabetes():
diabetes_data = datasets.load_diabetes()
x = diabetes_data.data
y = diabetes_data.target
for i in range(x.shape[1]):
xi = array_functions.normalize(x[:, i])
yi = array_functions.normalize(y)
array_functions.plot_2d(xi, yi)
pass
assert False
def create_synthetic_delta_linear_transfer():
slope = 5
target_fun = lambda x: slope * x
source_fun = lambda x: slope * x + 4
data = create_synthetic_regression_transfer(target_fun, source_fun)
array_functions.plot_2d(data.x,
data.y,
data.data_set_ids,
title='Linear Delta Data Set')
s = synthetic_delta_linear_file
helper_functions.save_object(s, data)
def vis_data():
s = 'data_sets/' + data_file_dir + '/raw_data.pkl'
data = helper_functions.load_object(s)
x = data.x
y = data.y
for i in range(data.p):
xi = x[:, i]
title = 'Feature Names Missing'
if data.feature_names is not None:
title = data.feature_names[i]
array_functions.plot_2d(xi, y, data_set_ids=data.data_set_ids, title=title)
pass
pass
def create_synthetic_multitask_transfer():
slope_source = 8
target_slope1 = 4
target_slope2 = 4.5
source_func = lambda x: slope_source * x
target_funcs = [
lambda x: target_slope1 * x + 3, lambda x: target_slope2 * x + 8
]
data = create_synthetic_regression_transfer(target_funcs, source_func)
array_functions.plot_2d(data.x,
data.y,
data.data_set_ids,
title='Multitask Slant')
s = synthetic_slant_multitask
helper_functions.save_object(s, data)
def create_synthetic_classification(file_dir="", local=True):
dim = 1
n_target = 200
n_source = 200
n = n_target + n_source
data = data_class.Data()
data.x = np.random.uniform(0, 1, (n, dim))
data.data_set_ids = np.zeros(n)
data.data_set_ids[n_target:] = 1
data.y = np.zeros(n)
x, ids = data.x, data.data_set_ids
I = array_functions.in_range(x, 0, 0.25)
I2 = array_functions.in_range(x, 0.25, 0.5)
I3 = array_functions.in_range(x, 0.5, 0.75)
I4 = array_functions.in_range(x, 0.75, 1)
id0 = ids == 0
id1 = ids == 1
data.y[I & id0] = 1
data.y[I2 & id0] = 2
data.y[I3 & id0] = 1
data.y[I4 & id0] = 2
data.y[I & id1] = 3
data.y[I2 & id1] = 4
data.y[I3 & id1] = 3
data.y[I4 & id1] = 4
if local:
data.y[I3 & id1] = 4
data.y[I4 & id1] = 3
data.set_true_y()
data.set_train()
data.is_regression = False
noise_rate = 0
# data.add_noise(noise_rate)
data.add_noise(noise_rate, id0, np.asarray([1, 2]))
data.add_noise(noise_rate, id1, np.asarray([3, 4]))
s = synthetic_classification_file
if local:
s = synthetic_classification_local_file
i = id1
array_functions.plot_2d(data.x[i, :], data.y[i])
if file_dir != "":
s = file_dir + "/" + s
helper_functions.save_object(s, data)
def create_synthetic_classification(file_dir='', local=True):
dim = 1
n_target = 200
n_source = 200
n = n_target + n_source
data = data_class.Data()
data.x = np.random.uniform(0, 1, (n, dim))
data.data_set_ids = np.zeros(n)
data.data_set_ids[n_target:] = 1
data.y = np.zeros(n)
x, ids = data.x, data.data_set_ids
I = array_functions.in_range(x, 0, .25)
I2 = array_functions.in_range(x, .25, .5)
I3 = array_functions.in_range(x, .5, .75)
I4 = array_functions.in_range(x, .75, 1)
id0 = ids == 0
id1 = ids == 1
data.y[I & id0] = 1
data.y[I2 & id0] = 2
data.y[I3 & id0] = 1
data.y[I4 & id0] = 2
data.y[I & id1] = 3
data.y[I2 & id1] = 4
data.y[I3 & id1] = 3
data.y[I4 & id1] = 4
if local:
data.y[I3 & id1] = 4
data.y[I4 & id1] = 3
data.set_true_y()
data.set_train()
data.is_regression = False
noise_rate = 0
#data.add_noise(noise_rate)
data.add_noise(noise_rate, id0, np.asarray([1, 2]))
data.add_noise(noise_rate, id1, np.asarray([3, 4]))
s = synthetic_classification_file
if local:
s = synthetic_classification_local_file
i = id1
array_functions.plot_2d(data.x[i, :], data.y[i])
if file_dir != '':
s = file_dir + '/' + s
helper_functions.save_object(s, data)
def create_synthetic_step_linear_transfer(file_dir=""):
n_target = 100
n_source = 100
n = n_target + n_source
sigma = 0.5
data = data_class.Data()
data.x = np.random.uniform(0, 1, (n, 1))
data.data_set_ids = np.zeros(n)
data.data_set_ids[n_target:] = 1
data.y = np.reshape(data.x * 5, data.x.shape[0])
data.y[(data.data_set_ids == 1) & (data.x[:, 0] >= 0.5)] += 4
data.y += np.random.normal(0, sigma, n)
data.set_defaults()
data.is_regression = True
array_functions.plot_2d(data.x, data.y, data.data_set_ids, title="Linear Step Data Set")
s = synthetic_step_linear_transfer_file
if file_dir != "":
s = file_dir + "/" + s
helper_functions.save_object(s, data)
def create_synthetic_step_linear_transfer(file_dir=''):
n_target = 100
n_source = 100
n = n_target + n_source
sigma = .5
data = data_class.Data()
data.x = np.random.uniform(0, 1, (n, 1))
data.data_set_ids = np.zeros(n)
data.data_set_ids[n_target:] = 1
data.y = np.reshape(data.x * 5, data.x.shape[0])
data.y[(data.data_set_ids == 1) & (data.x[:, 0] >= .5)] += 4
data.y += np.random.normal(0, sigma, n)
data.set_defaults()
data.is_regression = True
array_functions.plot_2d(data.x,
data.y,
data.data_set_ids,
title='Linear Step Data Set')
s = synthetic_step_linear_transfer_file
if file_dir != '':
s = file_dir + '/' + s
helper_functions.save_object(s, data)
def train_g_nonparametric(self, target_data):
y_t, y_s, y_true = self.get_predictions(target_data)
is_labeled = target_data.is_labeled
labeled_inds = is_labeled.nonzero()[0]
n_labeled = len(labeled_inds)
g = cvx.Variable(n_labeled)
'''
L = array_functions.make_laplacian_uniform(target_data.x[labeled_inds,:],self.radius,metric) \
+ .0001*np.identity(n_labeled)
'''
L = array_functions.make_laplacian_kNN(target_data.x[labeled_inds,:],self.k,self.metric) \
+ .0001*np.identity(n_labeled)
if self.use_fused_lasso:
reg = cvx_functions.create_fused_lasso(-L, g)
else:
reg = cvx.quad_form(g,L)
loss = cvx.sum_entries(
cvx.power(
cvx.mul_elemwise(y_s[:,0], g) + cvx.mul_elemwise(y_t[:,0], (1-g)) - y_true[:,0],
2
)
)
constraints = [g >= 0, g <= .5]
#constraints += [g[0] == .5, g[-1] == 0]
obj = cvx.Minimize(loss + self.C*reg)
prob = cvx.Problem(obj,constraints)
assert prob.is_dcp()
try:
prob.solve()
g_value = np.reshape(np.asarray(g.value),n_labeled)
except:
k = 0
#assert prob.status is None
print 'CVX problem: setting g = ' + str(k)
print '\tsigma=' + str(self.sigma)
print '\tC=' + str(self.C)
print '\tradius=' + str(self.radius)
g_value = k*np.ones(n_labeled)
if self.should_plot_g and enable_plotting and target_data.x.shape[1] == 1:
array_functions.plot_2d(target_data.x[labeled_inds,:],g_value)
labeled_train_data = target_data.get_subset(labeled_inds)
assert labeled_train_data.y.shape == g_value.shape
g_nw = method.NadarayaWatsonMethod(copy.deepcopy(self.configs))
labeled_train_data.is_regression = True
labeled_train_data.y = g_value
labeled_train_data.true_y = g_value
g_nw.configs.loss_function = loss_function.MeanSquaredError()
g_nw.tune_loo(labeled_train_data)
g_nw.train(labeled_train_data)
'''
a =np.hstack((g_value[labeled_train_data.x.argsort(0)], np.sort(labeled_train_data.x,0)))
print str(a)
print 'g_nw sigma: ' + str(g_nw.sigma)
print 'C:' + str(self.C)
'''
target_data.is_regression = True
self.g = g_nw.predict(target_data).fu
self.g[labeled_inds] = g_value
assert not np.any(np.isnan(self.g))
def vis_data():
s = 'data_sets/' + data_file_dir + '/raw_data.pkl'
data = helper_functions.load_object(s)
x = data.x
y = data.y
titles = ['', '']
label_idx = [0, 1]
if plot_climate:
img_path = 'C:/PythonFramework/far_transfer/figures/climate-terrain.png'
image = imread(img_path)
label_idx = [0, 4]
if data_file_dir == 'climate-month':
titles = [
'Max Temperature Gradient: January',
'Max Temperature Gradient: April'
]
label_idx = [0, 4]
elif data_file_dir == 'irs-income':
titles = ['Income', 'Household Size']
elif data_file_dir == 'zillow-traffic':
titles = ['Morning Taxi Pickups', 'Housing Prices']
elif data_file_dir == 'kc-housing-spatial-floors':
titles = ['House Prices: 1 Floor', 'House Prices: 2 or More Floors']
if plot_features:
for i in range(data.p):
xi = x[:, i]
title = 'Feature Names Missing'
if data.feature_names is not None:
title = data.feature_names[i]
array_functions.plot_2d(xi,
y,
data_set_ids=data.data_set_ids,
title=title)
else:
for i, title in zip(label_idx, titles):
#plt.close()
I = data.data_set_ids == i
if plot_gradients or plot_values:
g, v = estimate_gradients(x, y, I)
if plot_values:
g = v
#g = np.log(g)
#g -= g.min()
#g += g.max()/10.0
#g /= g.max()
if data_file_dir == 'zillow-traffic':
if i == 0:
pass
g -= g.min()
g /= g.max()
#g **= .5
else:
pass
g -= g.min()
g /= g.max()
#g **= .5
else:
if i == 0:
g -= g.min()
g /= g.max()
g = np.sqrt(g)
else:
g -= g.min()
g /= g.max()
g **= 1
#array_functions.plot_heatmap(g, sizes=dot_sizes, fig=fig, title=title)
fig = plt.figure(i)
plt.title(title)
plt.axis('off')
plt.imshow(g)
array_functions.move_fig(fig, 750, 400)
#plt.show(block=False)
else:
fig = plt.figure(4)
array_functions.plot_heatmap(x[I, :],
y[I],
sizes=dot_sizes,
fig=fig,
title=title)
if plot_climate:
plt.imshow(image, zorder=0, extent=[-90, -78, 33.5, 38])
array_functions.move_fig(fig, 1400, 600)
plt.show(block=True)
pass