def plot_prediction(
learner=None,
data=None,
minx=0,
maxx=5,
step_size=0.01, # for plotting
label="function"):
plt.ion()
plt.xlabel("x")
plt.ylabel("y")
if data is None:
data = Data_from_file('data/simp_regr.csv',
prob_test=0,
boolean_features=False,
target_index=-1)
if learner is None:
learner = Linear_learner(data, squashed=False)
learner.learning_rate = 0.001
learner.learn(100)
learner.learning_rate = 0.0001
learner.learn(1000)
learner.learning_rate = 0.00001
learner.learn(10000)
learner.display(
1, "function learned is", learner.predictor_string(), "error=",
data.evaluate_dataset(data.train, learner.predictor, "sum-of-squares"))
plt.plot([e[0] for e in data.train], [e[-1] for e in data.train],
"bo",
label="data")
plt.plot(list(arange(minx, maxx, step_size)),
[learner.predictor([x]) for x in arange(minx, maxx, step_size)],
label=label)
plt.legend()
plt.draw()
def plot_polynomials(data=None,
learner_class=Linear_learner,
max_degree=5,
minx=0,
maxx=5,
num_iter=100000,
learning_rate=0.0001,
step_size=0.01, # for plotting
):
plt.ion()
plt.xlabel("x")
plt.ylabel("y")
if data is None:
data = Data_from_file('data/simp_regr.csv', prob_test=0,
boolean_features=False, target_index=-1)
plt.plot([e[0] for e in data.train], [e[-1] for e in data.train], "ko", label="data")
x_values = list(arange(minx, maxx, step_size))
line_styles = ['-', '--', '-.', ':']
colors = ['0.5', 'k', 'k', 'k', 'k']
for degree in range(max_degree):
data_aug = Data_set_augmented(data, [power_feat(n) for n in range(1, degree + 1)],
include_orig=False)
learner = learner_class(data_aug, squashed=False)
learner.learning_rate = learning_rate
learner.learn(num_iter)
learner.display(1, "For degree", degree,
"function learned is", learner.predictor_string(),
"error=", data.evaluate_dataset(data.train, learner.predictor, "sum-of-squares"))
ls = line_styles[degree % len(line_styles)]
col = colors[degree % len(colors)]
plt.plot(x_values, [learner.predictor([x]) for x in x_values], linestyle=ls, color=col,
label="degree=" + str(degree))
plt.legend(loc='upper left')
plt.draw()
def plot_steps(learner=None,
data=None,
criterion="sum-of-squares",
step=1,
num_steps=1000,
log_scale=True,
label=""):
"""
plots the training and test error for a learner.
data is the
learner_class is the class of the learning algorithm
criterion gives the evaluation criterion plotted on the y-axis
step specifies how many steps are run for each point on the plot
num_steps is the number of points to plot
"""
plt.ion()
plt.xlabel("step")
plt.ylabel("Average " + criterion + " error")
if log_scale:
plt.xscale('log') #plt.semilogx() #Makes a log scale
else:
plt.xscale('linear')
if data is None:
data = Data_from_file('data/holiday.csv',
num_train=19,
target_index=-1)
#data = Data_from_file('data/SPECT.csv', target_index=0)
# data = Data_from_file('data/mail_reading.csv', target_index=-1)
# data = Data_from_file('data/carbool.csv', target_index=-1)
random.seed(None) # reset seed
if learner is None:
learner = Linear_learner(data)
train_errors = []
test_errors = []
for i in range(1, num_steps + 1, step):
test_errors.append(
data.evaluate_dataset(data.test, learner.predictor, criterion))
train_errors.append(
data.evaluate_dataset(data.train, learner.predictor, criterion))
learner.display(2, "Train error:", train_errors[-1], "Test error:",
test_errors[-1])
learner.learn(num_iter=step)
plt.plot(range(1, num_steps + 1, step),
train_errors,
ls='-',
c='k',
label="training errors")
plt.plot(range(1, num_steps + 1, step),
test_errors,
ls='--',
c='k',
label="test errors")
plt.legend()
plt.pause(0.001)
plt.show(block=True)
learner.display(1, "Train error:", train_errors[-1], "Test error:",
test_errors[-1])
def test(**args):
data = Data_from_file('data/SPECT.csv', target_index=0)
# data = Data_from_file('data/mail_reading.csv', target_index=-1)
# data = Data_from_file('data/carbool.csv', target_index=-1)
learner = Linear_learner(data, **args)
learner.learn()
print("function learned is", learner.predictor_string())
for ecrit in Data_set.evaluation_criteria:
test_error = data.evaluate_dataset(data.test, learner.predictor, ecrit)
print(" Average", ecrit, "error is", test_error)
def testMain():
data = Data_from_file('data/mail_reading.csv', target_index=-1)
#data = Data_from_file('data/mail_reading_consis.csv', target_index=-1)
#data = Data_from_file('data/SPECT.csv', prob_test=0.5, target_index=0)
#data = Data_from_file('data/holiday.csv', target_index=-1) #, num_train=19)
nn1 = NN(data)
nn1.add_layer(Linear_complete_layer(nn1, 3))
nn1.add_layer(Sigmoid_layer(nn1)) # comment this or the next
# nn1.add_layer(ReLU_layer(nn1))
nn1.add_layer(Linear_complete_layer(nn1, 1))
nn1.add_layer(Sigmoid_layer(nn1))
nn1.learning_rate = 0.1
#nn1.learn(100)
from learnLinear import plot_steps
import time
start_time = time.perf_counter()
plot_steps(learner=nn1, data=data, num_steps=10000)
for eg in data.train:
print(eg, nn1.predictor(eg))
end_time = time.perf_counter()
print("Time:", end_time - start_time)
train_errors.append( sum(self.distance(self.class_of_eg(eg),eg)
for eg in self.dataset.train)
/len(self.dataset.train))
if self.dataset.test:
test_errors.append( sum(self.distance(self.class_of_eg(eg),eg)
for eg in self.dataset.test)
/len(self.dataset.test))
plt.plot(range(1,maxstep+1),train_errors,
label=str(self.num_classes)+" classes. Training set")
if self.dataset.test:
plt.plot(range(1,maxstep+1),test_errors,
label=str(self.num_classes)+" classes. Test set")
plt.legend()
plt.draw()
%data = Data_from_file('data/emdata1.csv', num_train=10, target_index=2000) % trivial example
data = Data_from_file('data/emdata2.csv', num_train=10, target_index=2000)
%data = Data_from_file('data/emdata0.csv', num_train=14, target_index=2000) % example from textbook
kml = K_means_learner(data,2)
num_iter=4
print("Class assignment after",num_iter,"iterations:")
kml.learn(num_iter); kml.show_classes()
# Plot the error
# km2=K_means_learner(data,2); km2.plot_error(20) # 2 classes
# km3=K_means_learner(data,3); km3.plot_error(20) # 3 classes
# km13=K_means_learner(data,13); km13.plot_error(20) # 13 classes
# data = Data_from_file('data/carbool.csv', target_index=2000,boolean_features=True)
# kml = K_means_learner(data,3)
# kml.learn(20); kml.show_classes()
# Testing
from learnDT import DT_learner
from learnProblem import Data_set, Data_from_file
def sp_DT_learner(min_prop=0.9):
def make_learner(dataset):
mne = len(dataset.train) * min_prop
return DT_learner(dataset, min_number_examples=mne)
return make_learner
data = Data_from_file('data/carbool.csv', target_index=-1)
#data = Data_from_file('data/SPECT.csv', target_index=0)
#data = Data_from_file('data/mail_reading.csv', target_index=-1)
#data = Data_from_file('data/holiday.csv', num_train=19, target_index=-1)
learner9 = Boosting_learner(data, sp_DT_learner(0.9))
#learner7 = Boosting_learner(data, sp_DT_learner(0.7))
#learner5 = Boosting_learner(data, sp_DT_learner(0.5))
predictor9 = learner9.learn(10)
for i in learner9.offsets:
print(i.__doc__)
import matplotlib.pyplot as plt
def plot_boosting(data,
steps=10,
thresholds=[0.5, 0.1, 0.01, 0.001],
error_example(predictor(example), dataset.target(example), to_optimize)
for example in data_subset)
return error
from learnProblem import Data_set, Data_from_file
def test(data):
"""Prints errors and the trees for various evaluation criteria and ways to select leaves.
"""
for crit in Data_set.evaluation_criteria:
for leaf in selections:
tree = DT_learner(data, to_optimize=crit,
leaf_selection=leaf).learn()
print("For", crit, "using", leaf, "at leaves, tree built is:",
tree.__doc__)
if data.test:
for ecrit in Data_set.evaluation_criteria:
test_error = data.evaluate_dataset(data.test, tree, ecrit)
print(" Average error for", ecrit, "using", leaf,
"at leaves is", test_error)
if __name__ == "__main__":
#print("carbool.csv"); test(data = Data_from_file('data/carbool.csv', target_index=-1))
# print("SPECT.csv"); test(data = Data_from_file('data/SPECT.csv', target_index=0))
print("mail_reading.csv")
test(data=Data_from_file('data/mail_reading.csv', target_index=-1))
# print("holiday.csv"); test(data = Data_from_file('data/holiday.csv', num_train=19, target_index=-1))
def plot_fig_7_15(): # different runs produce different plots
data = Data_from_file('data/SPECT.csv', target_index=0)
# data = Data_from_file('data/carbool.csv', target_index=-1)
plot_error(data)
def prod(L):
"""returns the product of the elements of L"""
res = 1
for e in L:
res *= e
return res
def random_dist(k):
"""generate k random numbers that sum to 1"""
res = [random.random() for i in range(k)]
s = sum(res)
return [v / s for v in res]
data = Data_from_file('data/emdata2.csv', num_train=10, target_index=2000)
eml = EM_learner(data, 2)
num_iter = 2
print("Class assignment after", num_iter, "iterations:")
eml.learn(num_iter)
eml.show_class(0)
# Plot the error
# em2=EM_learner(data,2); em2.plot_error(40) # 2 classes
# em3=EM_learner(data,3); em3.plot_error(40) # 3 classes
# em13=EM_learner(data,13); em13.plot_error(40) # 13 classes
# data = Data_from_file('data/carbool.csv', target_index=2000,boolean_features=False)
# [f.frange for f in data.input_features]
# eml = EM_learner(data,3)
# eml.learn(20); eml.show_class(0)
error_example(predictor(example), dataset.target(example), to_optimize)
for example in data_subset)
return error
from learnProblem import Data_set, Data_from_file
def test(data):
"""Prints errors and the trees for various evaluation criteria and ways to select leaves.
"""
for crit in Data_set.evaluation_criteria:
for leaf in ("mean", "median"):
tree = DT_learner(data, to_optimize=crit,
leaf_selection=leaf).learn()
print("For", crit, "using", leaf, "at leaves, tree built is:",
tree.__doc__)
if data.test:
for ecrit in Data_set.evaluation_criteria:
test_error = data.evaluate_dataset(data.test, tree, ecrit)
print(" Average error for", ecrit, "using", leaf,
"at leaves is", test_error)
if __name__ == "__main__":
# print("carbool.csv"); test(data = Data_from_file('data/carbool.csv', target_index=-1))
print("SPECT.csv")
test(data=Data_from_file('data/pima.txt', target_index=0))
# print("mail_reading.csv"); test(data = Data_from_file('data/mail_reading.csv', target_index=-1))
# print("holiday.csv"); test(data = Data_from_file('data/holiday.csv', num_train=19, target_index=-1))
# Artificial Intelligence: Foundations of Computational Agents
# http://artint.info
# Copyright David L Poole and Alan K Mackworth 2017.
# This work is licensed under a Creative Commons
# Attribution-NonCommercial-ShareAlike 4.0 International License.
# See: http://creativecommons.org/licenses/by-nc-sa/4.0/deed.en
from learnDT import DT_learner
from learnProblem import Data_set, Data_from_file
def test(data):
"""Prints errors and the trees for various evaluation criteria and ways to select leaves.
"""
for crit in Data_set.evaluation_criteria:
for leaf in ("mean", "median"):
tree = DT_learner(data, to_optimize=crit,
leaf_selection=leaf).learn()
print("For", crit, "using", leaf, "at leaves, tree built is:",
tree.__doc__)
if data.test:
for ecrit in Data_set.evaluation_criteria:
test_error = data.evaluate_dataset(data.test, tree, ecrit)
print(" Average error for", ecrit, "using", leaf,
"at leaves is", test_error)
if __name__ == "__main__":
test(data=Data_from_file('pima.txt', target_index=8))
def characteristic_error(target, prediction):
return [
1 - prediction[i] if target == i else -prediction[i]
for i in range(len(prediction))
]
def sum_squares_error(observed, predicted):
"""Returns the errors for each of the target features.
"""
return [obsd - pred for obsd, pred in zip(observed, predicted)]
data = Data_from_file('data/training.txt', target_index=-1)
#data = Data_from_file('data/mail_reading_consis.csv', target_index=-1)
#data = Data_from_file('data/SPECT.csv', prob_test=0.5, target_index=0)
# data = Data_from_file('data/holiday.csv', target_index=-1) #, num_train=19)
nn1 = NN(data)
nn1.add_layer(Linear_complete_layer(nn1, 50))
nn1.add_layer(Sigmoid_layer(nn1)) # comment this or the next
# nn1.add_layer(ReLU_layer(nn1))
nn1.add_layer(Linear_complete_layer(nn1, 10))
nn1.add_layer(Sigmoid_layer(nn1))
nn1.learning_rate = 0.1
# nn1.learn(100)
start_time = time.perf_counter()
plot_steps(learner=nn1,
data=data,