def simulate_seperable(data_size):
"""Simulate learning a completely seperable data set."""
data = generate_sphere_data(10000, margin=0)
train_data, test_data = split_list(data, 0.75)
w = train(train_data, max_iter=500, r=0.01)
error = test(test_data, w)
status(train_data, test_data, error)
plot_data(data)
plot_w(data, w)
show()
def simulate_increasing(data_size, margin=0.3, max_iter=100, learning_rate=0.1,
steps=5, start=None, end=None):
"""Simulate learning an increasing training data set.
Generates an unseperable data set, and trains on an increasing training
set, then tests and plots.
start: Initial (first step) training data set size.
end: Final (last step) training data set size.
"""
data = generate_sphere_data(data_size, margin=margin)
train_data, test_data = split_list(data, 0.75)
# Initialize start/end sizes if not given.
start = len(train_data)/steps if start is None else start
end = len(train_data) if end is None else end
w_colors = ['b', 'c', 'm', 'y', 'k'] # w vector (line) graph color.
w_gs = [] # w plot graphs.
sizes = [] # Training data set sizes.
success = [] # Success rates according to training data set sizes.
for i in xrange(steps):
# Increase training data size according to iteration.
size = start + i*end/steps
current_train_data = train_data[:size]
w = train(current_train_data, max_iter=max_iter, r=learning_rate)
error = test(test_data, w)
status(current_train_data, test_data, error)
print
# Record size-success statistics.
sizes.append(size)
success.append(100 - error)
# Plot decision boundary.
w_color = w_colors[i] if i < len(w_colors) else w_colors[-1]
figure(0)
g, = plot_w(current_train_data, w, color=w_color)
w_gs.append(g)
figure(0).suptitle('Test data size: %d\nMaximum iterations: %d' % (len(test_data), max_iter))
plot_w_legend(w_gs, sizes)
plot_data(data)
figure(1).suptitle('Success rate according to training set size.')
plot_success_per_size(sizes, success)
show()