def __init__(self, emulated, LEDGrided, LCDed, keyboardhacked, touch):
self.qOut= Queue.Queue(maxsize=0)
self.qIn = Queue.Queue(maxsize=0)
self.lowerBoarder = 7
self.res = 8
self.mapping = {8:7,4:3,2:1,1:0}
self.emulated = emulated
self.LEDGrided = LEDGrided
self.LCDed = LCDed
self.keyboardhacked = keyboardhacked
self.touch = touch
if self.LEDGrided or self.LCDed or self.keyboardhacked:
self.i2chandler = i2cHandler.handler(self.qOut, self.LEDGrided, self.LCDed, self.keyboardhacked, self.touch)
if self.emulated or self.keyboardhacked:
import em
print "Emulated"
self.qOutEmulated = Queue.Queue(maxsize=0)
self.emu = em.em(self.qOutEmulated, self.qIn)
if self.LEDGrided or self.keyboardhacked:
self.i2chandler.resetPacketCount()
if self.keyboardhacked:
print "Keyboardhacked"
self.clearScreen(5)
if self.LEDGrided:
self.clearScreen(width*length)
def em_experiment(clf, X_tr, y_tr, X_te, y_te, n_classes, multi_class=False):
mlb = MultiLabelBinarizer(classes=range(n_classes))
mlb.fit(np.expand_dims(np.hstack((y_tr, y_te)), 1))
y_tr_bin = mlb.transform(np.expand_dims(y_tr, 1))
y_te_bin = mlb.transform(np.expand_dims(y_te, 1))
train_priors = np.mean(y_tr_bin, 0)
test_priors = np.mean(y_te_bin, 0)
clf = init_classifiers(clf)
print("Fitting", clf)
clf.fit(X_tr, y_tr)
test_posteriors = clf.predict_proba(X_te)
posteriors_test_priors = np.mean(test_posteriors, axis=0)
print('train priors', train_priors, sep='\n')
print('test priors', test_priors, sep='\n')
print('posteriors mean', posteriors_test_priors, sep='\n')
print()
em_test_posteriors, em_test_priors, history = em(y_te,
test_posteriors,
train_priors,
multi_class=multi_class)
measures = get_measures_from_singlehist_measures(history)
print('Results')
print('prior from: train test post em')
for i, (a, b, c, d) in enumerate(
zip(train_priors, test_priors, posteriors_test_priors,
em_test_priors)):
print(f'{i:11d} - {a:3.3f} {b:3.3f} {c:3.3f} {d:3.3f}')
return measures
def cluster(X, K, divergence, debug=False):
if divergence == 'KL':
dist_cls = distributions.KL
if np.any(X <= 0): # for MFCCs...
X = X - X.min() + 1e-8
X = 5. * X / X.sum(1)[:,nax]
elif divergence == 'IS':
dist_cls = distributions.ItakuraSaito
if np.any(X <= 0): # for MFCCs...
X = X - X.min() + 1e-8
# X = X / X.sum(1)[:,nax]
elif divergence == 'EU':
dist_cls = distributions.SquareDistance
else:
print 'Wrong divergence'
sys.exit(0)
assignments, centroids, _ = kmeans.kmeans_best_of_n(X, K, n_trials=10,
dist_cls=dist_cls, debug=debug)
init_pi = np.ones(K) / K
init_obs_distr = centroids
tau_em, obs_distr, pi, em_ll_train, _ = em.em(X, centroids, n_iter=10)
# tau_hmm, A, obs_distr, pi, ll_train, _ = hmm.em_hmm(X, init_pi, init_obs_distr, n_iter=10)
# seq_hmm, _ = hmm.viterbi(X, pi, A, obs_distr)
Tracer()()
return {'kmeans': assignments,
'EM': np.argmax(tau_em, axis=1),
# 'hmm_smoothing': np.argmax(tau_hmm, axis=1),
# 'hmm_viterbi': seq_hmm,
}
def main(job_id, params):
num_runs = 20
obs_length = 100
num_states = 2
num_obs = 2
# readin hmm indx
t = 0
try:
with open(os.path.join('.', 'hmm_index.txt')) as hmm_index_file:
t = int(hmm_index_file.read())
sys.stderr.write("!!!!!!!!!!!!!!!!!!HMM INDEX: " + str(t) + " !!!!!!!!!!!!!!!\n")
except IOError:
t = 0
# generate HMM observations
np.random.seed(0x6b6c26b2)
seeds = np.random.randint(0x0fffffff, size=num_runs)
np.random.seed(seeds[t])
# random hmm
z_mat, t_mat = random_hmm(num_states, num_obs)
pi_vec = np.array([1.0 / num_states] * num_states)
hmm_test = HMM(z_mat, t_mat, pi_vec)
# random obs trajectory
obs = hmm_test.generate(obs_length)[np.newaxis, :]
# calculate log likelihood for input HMM parameters
z_mat_p_input = np.array([[params['z_mat_p_0'][0], params['z_mat_p_1'][0]]])
t_mat_p_input = np.array([[params['t_mat_p_0'][0], params['t_mat_p_1'][0]]])
# pi_vec_input = np.array([params['pi_0'], 1 - params['pi_0']])
hmm_estimate = make_parameterized_HMM(z_mat_p_input, t_mat_p_input, pi_vec)
hmm_loglikelihood = hmm_estimate.loglikelihood(obs[0])
# use the current suggest point and run EM to get a new point
hmm_em_est, _, _ = em.em(hmm_estimate, hmm_estimate.z_mat, hmm_estimate.t_mat, obs, 30, 0.1)
em_est_z_mat, em_est_t_mat = retrieve_parameterized_HMM(hmm_em_est)
em_est_ll = -hmm_em_est.loglikelihood(obs[0])
em_est_z_mat.reshape((em_est_z_mat.size,))
em_est_t_mat.reshape((em_est_t_mat.size,))
print em_est_t_mat
print em_est_z_mat
historical_points = [{'params': {}}]
# write z_mat
for i, v in enumerate(em_est_z_mat[0]):
historical_points[0]['params']['z_mat_p_' + str(i)] = {'values': np.array([v]), 'type': 'float'}
# write t_mat
for i, v in enumerate(em_est_t_mat[0]):
historical_points[0]['params']['t_mat_p_' + str(i)] = {'values': np.array([v]), 'type': 'float'}
historical_points[0]['value'] = em_est_ll
dump_new_history('.', historical_points)
return -hmm_loglikelihood
def em_random_restarts(num_restarts, num_clusters, parameters_df):
ll = None
clusters = None
for i in range(num_restarts):
# Initialize cluster statistics
new_clusters = [cl.Cluster() for j in range(num_clusters)]
cl.initialize_clusters(new_clusters, parameters_df)
# Run EM & Get LL Value
new_ll = em.em(parameters_df, new_clusters)
# Save Best LL Value and its clusters
if ll is None or new_ll > ll:
ll = new_ll
clusters = new_clusters
return ll, clusters
def main():
# Load image
im = Image.open(image_file).convert('RGB')
width, height = im.size
# Convenience function to build image band-by-band from array data
def image_from_array(dat):
bands = [Image.new('L', (width, height)) for n in range(3)]
for i in range(3):
bands[i].putdata(dat[:, i])
return Image.merge('RGB', bands)
# Resize image
width, height = int(width / image_rescale), int(height / image_rescale)
im = im.resize((width, height))
# Summary image
summary = Image.new('RGB', (width * 2 + 40, height * 2 + 60),
(255, 255, 255))
draw = ImageDraw.Draw(summary)
draw.text((5, height + 10), 'Original', fill=(0, 0, 0))
draw.text((width + 25, height + 10),
'Noise V = %.2f, C = %.2f' % (noise_var, noise_cov),
fill=(0, 0, 0))
draw.text((5, 2 * height + 40), 'Blocked Gamma', fill=(0, 0, 0))
draw.text((width + 25, 2 * height + 40), 'Dists', fill=(0, 0, 0))
del draw
summary.paste(im, (10, 10))
# Flatten to emissions
real_emissions = list(im.getdata())
num_data = len(real_emissions)
real_emissions = np.array(real_emissions)
# Block emissions
width_blocks = np.array_split(np.arange(width), block_splits)
height_blocks = np.array_split(np.arange(height), block_splits)
idx = np.arange(num_data)
idx.resize((height, width))
blocks = []
for hb in height_blocks:
for wb in width_blocks:
block = [idx[h, w] for h in hb for w in wb]
blocks.append(np.array(block))
# Generate noise
v, c = noise_var, noise_cov
cov = [[v, c, c], [c, v, c], [c, c, v]]
noise = np.random.multivariate_normal([0, 0, 0], cov, width * height)
noisy_emissions = real_emissions + noise
# Generate noisy image
noisy = image_from_array(noisy_emissions)
summary.paste(noisy, (30 + width, 10))
# Use K-means to initialize components
results = kmeans(noisy_emissions, num_comps)
init_gamma = results['best']
means = results['means']
# Analyze color space
if do_colormap:
col = {'R': 0, 'G': 1, 'B': 2}
plt.figure()
for i, (d, c1, c2) in enumerate([(real_emissions, 'R', 'G'),
(real_emissions, 'R', 'B'),
(real_emissions, 'G', 'B'),
(noisy_emissions, 'R', 'G'),
(noisy_emissions, 'R', 'B'),
(noisy_emissions, 'G', 'B')]):
plt.subplot(2, 3, i + 1)
plt.hexbin(d[:, col[c1]],
d[:, col[c2]],
gridsize=30,
extent=(0, 255, 0, 255))
plt.plot(means[:, col[c1]], means[:, col[c2]], '.k')
plt.xlabel(c1)
plt.ylabel(c2)
plt.axis([-20, 275, -20, 275])
plt.savefig('image_test_color_colormap.png')
plt.show()
# Do EM
results = em(noisy_emissions,
[MultivariateNormal() for n in range(num_comps)],
count_restart=count_restart,
blocks=blocks,
max_reps=100,
init_gamma=init_gamma,
trace=True,
pi_max=pi_max)
dists = results['dists']
dists_trace = results['dists_trace']
pi = results['pi']
print 'Iterations: %(reps)d' % results
gamma = np.transpose(results['gamma'])
means = np.array([d.mean() for d in dists])
covs = np.array([d.cov() for d in dists])
# Reconstruct with blocked gamma
rec_blocked_gamma = np.array(
[np.average(means, weights=g, axis=0) for g in gamma])
im_blocked_gamma = image_from_array(rec_blocked_gamma)
summary.paste(im_blocked_gamma, (10, 40 + height))
# Reconstruct from distributions alone
pi_opt = pi_maximize(noisy_emissions, dists)
phi = np.empty((num_data, num_comps))
for c in range(num_comps):
phi[:, c] = dists[c].density(noisy_emissions)
phi = np.matrix(phi)
for i, pi in enumerate(pi_opt):
phi[:, i] *= pi
gamma_dists = phi / np.sum(phi, axis=1)
rec_dists = np.array(np.dot(gamma_dists, means))
im_dists = image_from_array(rec_dists)
summary.paste(im_dists, (30 + width, 40 + height))
# Show summary image
if show_summary:
summary.show()
summary.save('image_test_color_reconstruction.png')
# Compare RMSE between reconstructions
def rmse(x):
return np.sqrt(np.mean((x - real_emissions)**2))
print 'Raw MSE: %.1f' % rmse(noisy_emissions)
print 'Blocked Gamma MSE: %.1f' % rmse(rec_blocked_gamma)
print 'Dists MSE: %.1f' % rmse(rec_dists)
# Visualize variance components
if do_variance_viz:
temp_files = []
col = {'R': 0, 'G': 1, 'B': 2}
fig = plt.figure()
for i, (d, c1, c2) in enumerate([(real_emissions, 'R', 'G'),
(real_emissions, 'R', 'B'),
(real_emissions, 'G', 'B'),
(noisy_emissions, 'R', 'G'),
(noisy_emissions, 'R', 'B'),
(noisy_emissions, 'G', 'B')]):
ax = fig.add_subplot(2, 3, i + 1)
plt.hexbin(d[:, col[c1]],
d[:, col[c2]],
gridsize=30,
extent=(0, 255, 0, 255))
plt.xlabel(c1)
plt.ylabel(c2)
plt.axis([-20, 275, -20, 275])
for idx, dists in enumerate(dists_trace):
ells = []
for i, (d, c1, c2) in enumerate([(real_emissions, 'R', 'G'),
(real_emissions, 'R', 'B'),
(real_emissions, 'G', 'B'),
(noisy_emissions, 'R', 'G'),
(noisy_emissions, 'R', 'B'),
(noisy_emissions, 'G', 'B')]):
for dist in dists:
m, c = dist.mean(), dist.cov()
cm = (c[[col[c1], col[c2]]])[:, [col[c1], col[c2]]]
e, v = la.eigh(cm)
ell = Ellipse(xy=[m[col[c1]], m[col[c2]]],
width=np.sqrt(e[0]),
height=np.sqrt(e[1]),
angle=(180.0 / np.pi) * np.arccos(v[0, 0]))
ells.append(ell)
ax = fig.add_subplot(2, 3, i + 1)
ax.add_artist(ell)
ell.set_clip_box(ax.bbox)
ell.set_alpha(0.9)
ell.set_facecolor(np.fmax(np.fmin(m / 255, 1), 0))
file_name = 'tmp_%03d.png' % idx
temp_files.append(file_name)
plt.savefig(file_name, dpi=100)
for ell in ells:
ell.remove()
command = ('mencoder', 'mf://tmp_*.png', '-mf',
'type=png:w=800:h=600:fps=5', '-ovc', 'lavc', '-lavcopts',
'vcodec=mpeg4', '-oac', 'copy', '-o',
'image_test_color_components.avi')
os.spawnvp(os.P_WAIT, 'mencoder', command)
for temp_file in temp_files:
os.unlink(temp_file)
# Find common variance components
print 'True noise:'
print cov
chols = [la.cholesky(c) for c in covs]
chol_recon = np.zeros((3, 3))
for i in range(3):
for j in range(3):
if j > i: continue
chol_recon[i, j] = np.Inf
for chol in chols:
if abs(chol[i, j]) < abs(chol_recon[i, j]):
chol_recon[i, j] = chol[i, j]
cov_recon = np.dot(chol_recon, np.transpose(chol_recon))
print 'Reconstructed noise:'
print cov_recon
print "ACCURACIES: " print "Histogram: ", mean(histogram_accuracies) print "Histogram Sensitivity", mean(histogram_sensitivities) print "Histogram Specificity", mean(histogram_specificities) print "Logreg: ", mean(logreg_accuracies) print "Logreg Sensitivity", mean(logreg_sensitivities) print "Logreg Specificity", mean(logreg_specificities) print "Gaussian Means: ", mean(gaussian_mean_accuracies) print "Gaussian Distinct: ", mean(gaussian_distinct_accuracies) print "Random: ", len(samples_healthy) / float(len(samples_healthy) + len(samples_unhealthy)) samples_unhealthy = list(samples_unhealthy) samples_unhealthy.extend(samples_s) samples_unhealthy = array(samples_unhealthy) sigma, mu, pi = em(samples_unhealthy, 2, 1000) results = test_em_for_2(sigma, mu, pi, samples_murmur, samples_s) print results """ sigma, mu, pi = em(samples_unhealthy, 13, 1000) labeled_diseases = [] results = test_em_for_13(sigma, mu, pi, labeled_diseases) print results """
# Initialize summary image
summary = Image.new('L', (28 * num_components + 65, 28 * len(num_blocks)), 255)
# Do inference for varying numbers of blocks
idxs = np.argsort(map(np.sum, emissions))
reps = []
for block_i, num_block in enumerate(num_blocks):
# Block data
blocks = np.array_split(idxs, num_block)
# Run EM
results = em(emissions, [
Product([Bernoulli() for i in range(28 * 28)])
for n in range(num_components)
],
count_restart=3.0,
blocks=blocks,
gamma_seed=137,
init_gamma=(init_to_labels and labels or None))
dists = results['dists']
print 'Reps: %d' % results['reps']
reps.append(results['reps'])
# Produce summary image
offset = 0
im = Image.new('L', (28 * len(dists), 28))
for d in results['dists']:
digit = Image.new('L', (28, 28))
digit.putdata(np.array(d.mean()) * 255)
im.paste(digit, (offset, 0))
offset += 28
for wb in width_blocks:
block = [idx[h, w] for h in hb for w in wb]
blocks.append(np.array(block))
# Generate noise
noise = np.random.normal(0, noise_sd, width * height)
noisy_emissions = real_emissions + noise
# Generate noisy image
noisy = Image.new('L', (width, height))
noisy.putdata(noisy_emissions)
summary.paste(noisy, (30 + width, 10))
# Do EM
results = em(noisy_emissions,
[NormalFixedMean(m, max_sigma=max_sigma) for m in range(256)],
count_restart=count_restart,
blocks=blocks)
dists = results['dists']
pi = results['pi']
print 'Iterations: %(reps)d' % results
gamma = np.transpose(results['gamma'])
means = np.array([d.mean() for d in dists])
sds = np.array([d.sd() for d in dists])
# Display summary figures
display_densities(real_emissions, dists)
# Reconstruct with argmax
im_argmax = Image.new('L', (width, height))
reconstruct_argmax = means[np.argmax(gamma, axis=1)]
Tracer()()
return seq, A, obs_distr, ll_test, monitor_vals
if __name__ == '__main__':
X = np.loadtxt('EMGaussian.data')
Xtest = np.loadtxt('EMGaussian.test')
K = 4
# Run simple EM (no HMM)
iterations = 40
assignments, centers, _ = kmeans.kmeans_best_of_n(X, K, n_trials=5)
new_centers = [distributions.Gaussian(c.mean, np.eye(2)) \
for c in centers]
tau, obs_distr, pi, gmm_ll_train, gmm_ll_test = \
em.em(X, new_centers, assignments, n_iter=iterations, Xtest=Xtest)
# example with fixed parameters
A = 1. / 6 * np.ones((K, K))
A[np.diag(np.ones(K)) == 1] = 0.5
lalpha, lbeta = alpha_beta(Xtest, pi, A, obs_distr)
log_p = smoothing(lalpha, lbeta)
p = np.exp(log_p)
def plot_traj(p):
plt.figure()
ind = np.arange(100)
for k in range(K):
plt.subplot(K, 1, k + 1)
plt.bar(ind, p[:100, k])
plt.imshow(data)
plt.show()
# Initialize with K-means
if init == 'kmeans':
init_gamma = kmeans(data.reshape((dim*dim,1)), 2)['best']
# Do (potentially adaptive) blocked EM, depending on strategy
for block_strategy in block_strategies:
# Only 'perfect' strategy uses the true states
blocks = block(data, block_strategy, true = x)
# Do EM
results = em(data.reshape((dim*dim,)),
model,
count_restart = count_restart,
blocks = blocks,
init_gamma = init_gamma,
pi_max = pi_max)
print 'Iterations: %d (%s)' % (results['reps'], block_strategy)
dists = results['dists']
pi = results['pi']
# Display results
if show_each:
for p, d in zip(np.transpose(pi), dists):
print '%s: %s' % (p, d.display())
print
if graphics:
display_densities(data.reshape((dim*dim,)), dists)
display_hist(data.reshape((dim*dim,)), dists)
t = time.time()
assignments, centroids, dists = \
kmeans.kmeans_best_of_n(X, K, n_trials=4, dist_cls=distributions.KL)
print 'K-means: {}s'.format(time.time() - t)
results[algos.kmeans] = {
'seq': assignments,
'centroids': centroids,
}
seqs[algos.kmeans] = assignments
# EM
if options.init == 'em' or algos.em in algs:
iterations = 10
t = time.time()
tau_em, obs_distr, pi, em_ll_train, _ = em.em(X,
centroids,
n_iter=options.n_iter)
print 'EM: {}s'.format(time.time() - t)
results[algos.em] = {
'seq': np.argmax(tau_em, axis=1),
'obs_distr': obs_distr,
'll_train': em_ll_train,
'tau': tau_em,
}
seqs[algos.em] = np.argmax(tau_em, axis=1)
# initialization
if options.init == 'em':
# initialize with EM
init_pi = pi
init_obs_distr = obs_distr
if init == 'true':
init_gamma = data_comp
data_mu = mu[data_comp]
data = np.random.normal(data_mu, 1)
blocks = np.array_split(np.arange(n), num_blocks)
# Initialize with K-means
if init == 'kmeans':
init_gamma = kmeans(data.reshape((n,1)), 2)['best']
# Do EM
results = em(data,
model,
count_restart = count_restart,
blocks = blocks,
init_gamma = init_gamma,
init_reps = em_steps,
max_reps = em_steps,
pi_max = pi_max,
trace = True)
if show_each:
print 'Iterations: %(reps)d' % results
dists, dists_trace = results['dists'], results['dists_trace']
pi, pi_trace = results['pi'], results['pi_trace']
# Display results
if show_each:
for p, d in zip(np.transpose(pi), dists):
print '%s: %s' % (p, d.display())
print
if graphics:
def main():
# Load image
im = Image.open(image_file).convert('RGB')
width, height = im.size
# Convenience function to build image band-by-band from array data
def image_from_array(dat):
bands = [Image.new('L', (width, height)) for n in range(3)]
for i in range(3):
bands[i].putdata(dat[:,i])
return Image.merge('RGB', bands)
# Resize image
width, height = int(width / image_rescale), int(height / image_rescale)
im = im.resize((width, height))
# Summary image
summary = Image.new('RGB', (width * 2 + 40, height * 2 + 60),
(255, 255, 255))
draw = ImageDraw.Draw(summary)
draw.text((5, height + 10), 'Original', fill = (0, 0, 0))
draw.text((width + 25, height + 10),
'Noise V = %.2f, C = %.2f' % (noise_var, noise_cov),
fill = (0, 0, 0))
draw.text((5, 2 * height + 40), 'Blocked Gamma', fill = (0, 0, 0))
draw.text((width + 25, 2 * height + 40), 'Dists', fill = (0, 0, 0))
del draw
summary.paste(im, (10, 10))
# Flatten to emissions
real_emissions = list(im.getdata())
num_data = len(real_emissions)
real_emissions = np.array(real_emissions)
# Block emissions
width_blocks = np.array_split(np.arange(width), block_splits)
height_blocks = np.array_split(np.arange(height), block_splits)
idx = np.arange(num_data)
idx.resize((height, width))
blocks = []
for hb in height_blocks:
for wb in width_blocks:
block = [idx[h, w] for h in hb for w in wb]
blocks.append(np.array(block))
# Generate noise
v, c = noise_var, noise_cov
cov = [[v, c, c], [c, v, c], [c, c, v]]
noise = np.random.multivariate_normal([0, 0, 0], cov, width * height)
noisy_emissions = real_emissions + noise
# Generate noisy image
noisy = image_from_array(noisy_emissions)
summary.paste(noisy, (30 + width, 10))
# Use K-means to initialize components
results = kmeans(noisy_emissions, num_comps)
init_gamma = results['best']
means = results['means']
# Analyze color space
if do_colormap:
col = { 'R': 0, 'G': 1, 'B': 2 }
plt.figure()
for i, (d, c1, c2) in enumerate([(real_emissions, 'R', 'G'),
(real_emissions, 'R', 'B'),
(real_emissions, 'G', 'B'),
(noisy_emissions, 'R', 'G'),
(noisy_emissions, 'R', 'B'),
(noisy_emissions, 'G', 'B')]):
plt.subplot(2, 3, i+1)
plt.hexbin(d[:,col[c1]], d[:,col[c2]], gridsize=30,
extent = (0, 255, 0, 255))
plt.plot(means[:,col[c1]], means[:,col[c2]], '.k')
plt.xlabel(c1)
plt.ylabel(c2)
plt.axis([-20, 275, -20, 275])
plt.savefig('image_test_color_colormap.png')
plt.show()
# Do EM
results = em(noisy_emissions,
[MultivariateNormal() for n in range(num_comps)],
count_restart = count_restart,
blocks = blocks,
max_reps = 100,
init_gamma = init_gamma,
trace = True,
pi_max = pi_max)
dists = results['dists']
dists_trace = results['dists_trace']
pi = results['pi']
print 'Iterations: %(reps)d' % results
gamma = np.transpose(results['gamma'])
means = np.array([d.mean() for d in dists])
covs = np.array([d.cov() for d in dists])
# Reconstruct with blocked gamma
rec_blocked_gamma = np.array([np.average(means, weights=g, axis=0)
for g in gamma])
im_blocked_gamma = image_from_array(rec_blocked_gamma)
summary.paste(im_blocked_gamma, (10, 40 + height))
# Reconstruct from distributions alone
pi_opt = pi_maximize(noisy_emissions, dists)
phi = np.empty((num_data, num_comps))
for c in range(num_comps):
phi[:,c] = dists[c].density(noisy_emissions)
phi = np.matrix(phi)
for i, pi in enumerate(pi_opt):
phi[:,i] *= pi
gamma_dists = phi / np.sum(phi, axis = 1)
rec_dists = np.array(np.dot(gamma_dists, means))
im_dists = image_from_array(rec_dists)
summary.paste(im_dists, (30 + width, 40 + height))
# Show summary image
if show_summary:
summary.show()
summary.save('image_test_color_reconstruction.png')
# Compare RMSE between reconstructions
def rmse(x):
return np.sqrt(np.mean((x - real_emissions) ** 2))
print 'Raw MSE: %.1f' % rmse(noisy_emissions)
print 'Blocked Gamma MSE: %.1f' % rmse(rec_blocked_gamma)
print 'Dists MSE: %.1f' % rmse(rec_dists)
# Visualize variance components
if do_variance_viz:
temp_files = []
col = { 'R': 0, 'G': 1, 'B': 2 }
fig = plt.figure()
for i, (d, c1, c2) in enumerate([(real_emissions, 'R', 'G'),
(real_emissions, 'R', 'B'),
(real_emissions, 'G', 'B'),
(noisy_emissions, 'R', 'G'),
(noisy_emissions, 'R', 'B'),
(noisy_emissions, 'G', 'B')]):
ax = fig.add_subplot(2, 3, i+1)
plt.hexbin(d[:,col[c1]], d[:,col[c2]], gridsize=30,
extent = (0, 255, 0, 255))
plt.xlabel(c1)
plt.ylabel(c2)
plt.axis([-20, 275, -20, 275])
for idx, dists in enumerate(dists_trace):
ells = []
for i, (d, c1, c2) in enumerate([(real_emissions, 'R', 'G'),
(real_emissions, 'R', 'B'),
(real_emissions, 'G', 'B'),
(noisy_emissions, 'R', 'G'),
(noisy_emissions, 'R', 'B'),
(noisy_emissions, 'G', 'B')]):
for dist in dists:
m, c = dist.mean(), dist.cov()
cm = (c[[col[c1], col[c2]]])[:,[col[c1], col[c2]]]
e, v = la.eigh(cm)
ell = Ellipse(xy = [m[col[c1]], m[col[c2]]],
width = np.sqrt(e[0]),
height = np.sqrt(e[1]),
angle = (180.0 / np.pi) * np.arccos(v[0,0]))
ells.append(ell)
ax = fig.add_subplot(2, 3, i+1)
ax.add_artist(ell)
ell.set_clip_box(ax.bbox)
ell.set_alpha(0.9)
ell.set_facecolor(np.fmax(np.fmin(m / 255, 1), 0))
file_name = 'tmp_%03d.png' % idx
temp_files.append(file_name)
plt.savefig(file_name, dpi = 100)
for ell in ells:
ell.remove()
command = ('mencoder',
'mf://tmp_*.png',
'-mf',
'type=png:w=800:h=600:fps=5',
'-ovc',
'lavc',
'-lavcopts',
'vcodec=mpeg4',
'-oac',
'copy',
'-o',
'image_test_color_components.avi')
os.spawnvp(os.P_WAIT, 'mencoder', command)
for temp_file in temp_files:
os.unlink(temp_file)
# Find common variance components
print 'True noise:'
print cov
chols = [la.cholesky(c) for c in covs]
chol_recon = np.zeros((3,3))
for i in range(3):
for j in range(3):
if j > i: continue
chol_recon[i,j] = np.Inf
for chol in chols:
if abs(chol[i,j]) < abs(chol_recon[i,j]):
chol_recon[i,j] = chol[i,j]
cov_recon = np.dot(chol_recon, np.transpose(chol_recon))
print 'Reconstructed noise:'
print cov_recon
import numpy as np
import em
observations = np.array([[1, 0, 0, 0, 1, 1, 0, 1, 0, 1],
[1, 1, 1, 1, 0, 1, 1, 1, 1, 1],
[1, 0, 1, 1, 1, 1, 1, 0, 1, 1],
[1, 0, 1, 0, 0, 0, 1, 1, 0, 0],
[0, 1, 1, 1, 0, 1, 1, 1, 0, 1]])
print em.em(observations, [0.6, 0.5])
print em.em(observations, [0.5, 0.6])
print em.em(observations, [0.3, 0.3])
print em.em(observations, [0.9999, 0.00000001])
def main():
tol = 0.00001
max_iter = 100
alpha = 2
beta = 1
"""SYN dataset"""
tag_syn = os.path.join("data", "SYN_data", "SYN_r")
num_trials = 10
plot_mv = []
plot_bp = []
plot_em = []
for value in range(
1, 20, 2
): # r=1,3,5,6,7,9,11,13,15,17,19, means number of tasks assigned to each worker
avg_error_mv = 0
avg_error_bp = 0
avg_error_em = 0
time_mv = 0
time_bp = 0
time_em = 0
for i in range(num_trials): # trial=1~10
data = io.loadmat(tag_syn +
"{}_{}.mat".format(value, i + 1))["data"][0][0]
true_labels = io.loadmat(
tag_syn +
"{}_{}_true.mat".format(value, i + 1))["true_labels"][0]
"""MV"""
start = time.time()
est_mv = mv(data)
error_mv = np.average(est_mv != true_labels)
avg_error_mv += error_mv
end = time.time()
time_mv += end - start
""""""
"""BP"""
start = time.time()
est_bp = bp(data, max_iter, tol, alpha, beta)
error_bp = np.average(est_bp != true_labels)
avg_error_bp += error_bp
end = time.time()
time_bp += end - start
""""""
"""EM"""
start = time.time()
est_em = em(data, max_iter, tol, alpha, beta)
error_em = np.average(est_em != true_labels)
avg_error_em += error_em
end = time.time()
time_em += end - start
""""""
avg_error_mv = avg_error_mv / num_trials
print(
"In SYN dataset using MV, when r={}, error rate is {}, time is {} secs"
.format(value, avg_error_mv, time_mv))
avg_error_bp = avg_error_bp / num_trials
print(
"In SYN dataset using BP, when r={}, error rate is {}, time is {} secs"
.format(value, avg_error_bp, time_bp))
avg_error_em = avg_error_em / num_trials
print(
"In SYN dataset using EM, when r={}, error rate is {}, time is {} secs"
.format(value, avg_error_em, time_em))
plot_mv.append(avg_error_mv)
plot_bp.append(avg_error_bp)
plot_em.append(avg_error_em)
plt.figure(figsize=(10, 8))
plt.title("SYN dataset")
plt.xlabel("r", )
plt.ylabel("error")
plt.xlim(0, 20)
plt.ylim(0, 1)
plt.plot([i for i in range(1, 20, 2)], plot_mv, label="MV")
plt.plot([i for i in range(1, 20, 2)], plot_bp, label="BP")
plt.plot([i for i in range(1, 20, 2)], plot_em, label="EM")
plt.legend(loc="upper left")
plt.show()
""""""
"""SIM dataset"""
tag_sim = os.path.join("data", "SIM_data", "SIM_l")
num_trials = 10
plot_mv = []
plot_bp = []
plot_em = []
for value in [
1, 5, 10, 15, 20, 25
]: # l=1,5,10,15,20,25, means number of workers assigned to each task
avg_error_mv = 0
avg_error_bp = 0
avg_error_em = 0
time_mv = 0
time_bp = 0
time_em = 0
for i in range(num_trials): # trial=1~10
data = io.loadmat(tag_sim +
"{}_{}.mat".format(value, i + 1))["data"][0][0]
true_labels = io.loadmat(
tag_sim +
"{}_{}_true.mat".format(value, i + 1))["true_labels"][0]
"""MV"""
start = time.time()
est_mv = mv(data)
error_mv = np.average(est_mv != true_labels)
avg_error_mv += error_mv
end = time.time()
time_mv += end - start
""""""
"""BP"""
start = time.time()
est_bp = bp(data, max_iter, tol, alpha, beta)
error_bp = np.average(est_bp != true_labels)
avg_error_bp += error_bp
end = time.time()
time_bp += end - start
""""""
"""EM"""
start = time.time()
est_em = em(data, max_iter, tol, alpha, beta)
error_em = np.average(est_em != true_labels)
avg_error_em += error_em
end = time.time()
time_em += end - start
""""""
avg_error_mv = avg_error_mv / num_trials
print(
"In SIM dataset using MV, when l={}, error rate is {}, time is {} secs"
.format(value, avg_error_mv, time_mv))
avg_error_bp = avg_error_bp / num_trials
print(
"In SIM dataset using BP, when l={}, error rate is {}, time is {} secs"
.format(value, avg_error_bp, time_bp))
avg_error_em = avg_error_em / num_trials
print(
"In SIM dataset using EM, when l={}, error rate is {}, time is {} secs"
.format(value, avg_error_em, time_em))
plot_mv.append(avg_error_mv)
plot_bp.append(avg_error_bp)
plot_em.append(avg_error_em)
plt.figure(figsize=(10, 8))
plt.title("SIM dataset")
plt.xlabel("l")
plt.ylabel("error")
plt.xlim(0, 30)
plt.ylim(0, 1)
plt.plot([1, 5, 10, 15, 20, 25], plot_mv, label="MV")
plt.plot([1, 5, 10, 15, 20, 25], plot_bp, label="BP")
plt.plot([1, 5, 10, 15, 20, 25], plot_em, label="EM")
plt.legend(loc="upper left")
plt.show()
""""""
"""TEMP dataset"""
data = io.loadmat("data/other_data/TEMP_data.mat")["data"][0][0]
true_labels = io.loadmat(
"data/other_data/TEMP_data_true.mat")["true_labels"][0]
"""MV"""
start = time.time()
est_mv = mv(data)
error_mv = np.average(est_mv != true_labels)
end = time.time()
time_mv = end - start
""""""
"""BP"""
start = time.time()
est_bp = bp(data, max_iter, tol, alpha, beta)
error_bp = np.average(est_bp != true_labels)
end = time.time()
time_bp = end - start
""""""
"""EM"""
start = time.time()
est_em = em(data, max_iter, tol, alpha, beta)
error_em = np.average(est_em != true_labels)
end = time.time()
time_em = end - start
""""""
print("In TEMP dataset using MV, error rate is {}, time is {} secs".format(
error_mv, time_mv))
print("In TEMP dataset using BP, error rate is {}, time is {} secs".format(
error_bp, time_bp))
print("In TEMP dataset using EM, error rate is {}, time is {} secs".format(
error_em, time_em))
""""""
from utils import ExpenseManagerUtils
import hashlib
import connection
import os
from nocache import nocache
import em
app = Flask(__name__)
app.secret_key = "highlysecret"
app.config['TEMPLATES_AUTO_RELOAD'] = True
emutils = ExpenseManagerUtils.UtilsLib()
conn = connection.connection()
emapi = em.em()
@app.route('/')
def reroute():
return redirect('/login', code=302)
@app.route('/login')
def login():
email = request.cookies.get('loggedin')
if email and request.cookies.get('security_verify:'+email) == emutils.hash_of_hashpass(email):
return redirect('/home', code=302)
else:
form = LoginForm.LoginForm()
return render_template('login.html', form=form)
@app.route('/register', methods = ['POST'])
def em_experiment(ax, clf, X_tr, y_tr, X_te, y_te, y_min=0, y_max=1.0):
mlb = MultiLabelBinarizer()
mlb.fit(np.expand_dims(np.hstack((y_tr, y_te)), 1))
y_tr_bin = mlb.transform(np.expand_dims(y_tr, 1))
y_te_bin = mlb.transform(np.expand_dims(y_te, 1))
train_priors = np.mean(y_tr_bin, 0)
test_priors = np.mean(y_te_bin, 0)
print("Fitting", clf)
clf.fit(X_tr, y_tr)
test_posteriors = clf.predict_proba(X_te)
posteriors_test_priors = np.mean(test_posteriors, axis=0)
print('train priors', train_priors, sep='\n')
print('test priors', test_priors, sep='\n')
print('posteriors mean', posteriors_test_priors, sep='\n')
print()
em_test_posteriors, em_test_priors, history = em(y_te, test_posteriors,
train_priors)
em_prior = [p[1] for _, p, _, _, _, _ in history]
accuracy = [a for _, _, _, a, _, _ in history]
f1 = [
2 * p * r / (p + r) if p + r > 0 else 0 for _, _, _, _, p, r in history
]
ax.set_ylim([y_min, y_max])
ax.plot(range(len(accuracy)),
accuracy,
linestyle='-.',
color='m',
label='accuracy')
ax.plot(range(len(f1)), f1, linestyle='--', color='#dd9f00', label='f1')
ax.plot(range(len(em_prior)), em_prior, color='b', label='em pr')
ax.hlines([train_priors[1]],
0,
len(em_prior) - 1,
colors=['r'],
linestyles=[':'],
label='train pr')
ax.hlines([posteriors_test_priors[1]],
0,
len(em_prior) - 1,
colors=['#b5651d'],
linestyles=['-.'],
label='clf pr')
ax.hlines([test_priors[1]],
0,
len(em_prior) - 1,
colors=['g'],
linestyles=['--'],
label='test pr')
ax.set()
ax.grid()
print('Results')
print('prior from: train test post em')
for i, (a, b, c, d) in enumerate(
zip(train_priors, test_priors, posteriors_test_priors,
em_test_priors)):
print(f'{i:11d} - {a:3.3f} {b:3.3f} {c:3.3f} {d:3.3f}')
return posteriors_test_priors[1], em_test_priors[1], accuracy[0], accuracy[
-1], f1[0], f1[-1]
plt.imshow(data)
plt.show()
# Initialize with K-means
if init == 'kmeans':
init_gamma = kmeans(data.reshape((dim * dim, 1)), 2)['best']
# Do (potentially adaptive) blocked EM, depending on strategy
for block_strategy in block_strategies:
# Only 'perfect' strategy uses the true states
blocks = block(data, block_strategy, true=x)
# Do EM
results = em(data.reshape((dim * dim, )),
model,
count_restart=count_restart,
blocks=blocks,
init_gamma=init_gamma,
pi_max=pi_max)
print 'Iterations: %d (%s)' % (results['reps'], block_strategy)
dists = results['dists']
pi = results['pi']
# Display results
if show_each:
for p, d in zip(np.transpose(pi), dists):
print '%s: %s' % (p, d.display())
print
if graphics:
display_densities(data.reshape((dim * dim, )), dists)
display_hist(data.reshape((dim * dim, )), dists)
if init == 'true':
init_gamma = data_comp
data_mu = mu[data_comp]
data = np.random.normal(data_mu, 1)
blocks = np.array_split(np.arange(n), num_blocks)
# Initialize with K-means
if init == 'kmeans':
init_gamma = kmeans(data.reshape((n, 1)), 2)['best']
# Do EM
results = em(data,
model,
count_restart=count_restart,
blocks=blocks,
init_gamma=init_gamma,
init_reps=em_steps,
max_reps=em_steps,
pi_max=pi_max,
trace=True)
if show_each:
print 'Iterations: %(reps)d' % results
dists, dists_trace = results['dists'], results['dists_trace']
pi, pi_trace = results['pi'], results['pi_trace']
# Display results
if show_each:
for p, d in zip(np.transpose(pi), dists):
print '%s: %s' % (p, d.display())
print
if graphics:
mat = spio.loadmat("HMMdata.mat")
X = mat['X']
print(X.shape)
# X = np.loadtxt('EMGaussian.data')
# Xtest = np.loadtxt('EMGaussian.test')
Xtest = X
K = 3
# Run simple EM (no HMM)
iterations = 40
assignments, centers, _ = kmeans.kmeans_best_of_n(X, K, n_trials=5)
new_centers = [distributions.Gaussian(c.mean, np.eye(2)) for c in centers]
tau, obs_distr, pi, gmm_ll_train, gmm_ll_test = em.em(X,
new_centers,
assignments,
n_iter=iterations,
Xtest=Xtest)
# example with fixed parameters
A = 1. / 6 * np.ones((K, K))
A[np.diag(np.ones(K)) == 1] = 0.5
lalpha, lbeta = alpha_beta(Xtest, pi, A, obs_distr)
log_p = smoothing(lalpha, lbeta)
p = np.exp(log_p)
# def plot_traj(p):
# plt.figure()
# ind = np.arange(100)
# for k in range(K):
import numpy as np
from plot_hist import plot_hist
from em import em
# Valeurs initiales pour 2 populations
pi0_2 = np.array([1. / 4, 3. / 4])
mu0_2 = np.array([.57, .67])
s20_2 = np.array([1. / 10000, 1. / 10000])
# Valeurs initiales pour 3 populations
pi0_3 = np.array([1. / 3, 1 / 3, 1. / 3])
mu0_3 = np.array([.57, .6, .67])
s20_3 = np.array([1. / 10000, 1. / 10000, 1. / 10000])
# Loi empirique
plot_hist('crabe.txt')
# 2 populations
pi, mu, s2 = em(pi0_2, mu0_2, s20_2)
plot_hist('crabe.txt', mu, s2, pi)
# 3 populations
pi, mu, s2 = em(pi0_3, mu0_3, s20_3)
plot_hist('crabe.txt', mu, s2, pi)
#programmed entirely by Paul An
import clusterparser
import kmeans
import em
#for testing and visualization
def writeKAssignments(assignments):
f = open('test.csv', 'w')
for assignment in assignments:
f.write(str(assignment) + '\n')
f.close()
def writeEMAssignments(assignments):
f = open('test.csv', 'w')
for assignment in assignments:
f.write(str(assignment.getAssignment()) + '\n')
f.close()
points = clusterparser.parsedatafile("clusters.txt")
centroids, kmassignments = kmeans.kmeans(points, 3)
print(centroids)
gausses, emassignments = em.em(points, 3)
for gauss in gausses:
print(gauss.mean)
print('done')
# Initialize summary image
summary = Image.new('L', (28 * num_components + 65, 28 * len(num_blocks)), 255)
# Do inference for varying numbers of blocks
idxs = np.argsort(map(np.sum, emissions))
reps = []
for block_i, num_block in enumerate(num_blocks):
# Block data
blocks = np.array_split(idxs, num_block)
# Run EM
results = em(emissions,
[Product([Bernoulli() for i in range(28 * 28)])
for n in range(num_components)],
count_restart = 3.0,
blocks = blocks,
gamma_seed = 137,
init_gamma = (init_to_labels and labels or None))
dists = results['dists']
print 'Reps: %d' % results['reps']
reps.append(results['reps'])
# Produce summary image
offset = 0
im = Image.new('L', (28 * len(dists), 28))
for d in results['dists']:
digit = Image.new('L', (28, 28))
digit.putdata(np.array(d.mean()) * 255)
im.paste(digit, (offset, 0))
offset += 28
iforest.fit(X_train)
s_X_iforest = iforest.decision_function(X_test)
print('LocalOutlierFactor processing...')
lof = LocalOutlierFactor(n_neighbors=20)
lof.fit(X_train)
s_X_lof = lof.decision_function(X_test)
print('OneClassSVM processing...')
ocsvm = OneClassSVM()
ocsvm.fit(X_train[:min(ocsvm_max_train, n_samples_train - 1)])
s_X_ocsvm = ocsvm.decision_function(X_test).reshape(1, -1)[0]
s_unif_iforest = iforest.decision_function(unif)
s_unif_lof = lof.decision_function(unif)
s_unif_ocsvm = ocsvm.decision_function(unif).reshape(1, -1)[0]
plt.subplot(121)
auc_iforest, em_iforest, amax_iforest = em(t, t_max,
volume_support,
s_unif_iforest,
s_X_iforest, n_generated)
auc_lof, em_lof, amax_lof = em(t, t_max, volume_support,
s_unif_lof, s_X_lof, n_generated)
auc_ocsvm, em_ocsvm, amax_ocsvm = em(t, t_max, volume_support,
s_unif_ocsvm, s_X_ocsvm,
n_generated)
if amax_iforest == -1 or amax_lof == -1 or amax_ocsvm == -1:
amax = -1
else:
amax = max(amax_iforest, amax_lof, amax_ocsvm)
plt.subplot(121)
plt.plot(t[:amax], em_iforest[:amax], lw=1,
label='%s (em_score = %0.3e)'
for wb in width_blocks:
block = [idx[h, w] for h in hb for w in wb]
blocks.append(np.array(block))
# Generate noise
noise = np.random.normal(0, noise_sd, width * height)
noisy_emissions = real_emissions + noise
# Generate noisy image
noisy = Image.new('L', (width, height))
noisy.putdata(noisy_emissions)
summary.paste(noisy, (30 + width, 10))
# Do EM
results = em(noisy_emissions,
[NormalFixedMean(m, max_sigma = max_sigma) for m in range(256)],
count_restart = count_restart,
blocks = blocks)
dists = results['dists']
pi = results['pi']
print 'Iterations: %(reps)d' % results
gamma = np.transpose(results['gamma'])
means = np.array([d.mean() for d in dists])
sds = np.array([d.sd() for d in dists])
# Display summary figures
display_densities(real_emissions, dists)
# Reconstruct with argmax
im_argmax = Image.new('L', (width, height))
reconstruct_argmax = means[np.argmax(gamma, axis=1)]
def main():
run_data = {}
run_id = 0
scale = 0.5
emissions_normal = { 1: Normal(0, 2.0 * scale),
2: Normal(3.5, 3.0 * scale),
3: Normal(6.5, 1.0 * scale) }
emissions_laplace = { 1: Laplace(0, 2.0 * scale),
2: Laplace(3.5, 3.0 * scale),
3: Laplace(6.5, 1.0 * scale) }
emission_spec = emissions_normal
dists = [Normal(max_sigma = 6.0) for n in range(3)]
num_state_reps = 50
num_emission_reps = 4
num_gamma_init_reps = 4
num_blocks = [1, 2, 5, 10, 20, 50]
verbose = False
graphics_on = False
total_work = (num_state_reps * num_emission_reps *
2 * num_gamma_init_reps * len(num_blocks))
work = 0
for state_rep in range(num_state_reps):
print 'State repetition %d' % state_rep
# Generate HMM states
while True:
model = HMM([('Start', (1,), (1.0,)),
(1, (1,2,3), (0.98, 0.02, 0.0)),
(2, (1,2,3), (0.02, 0.95, 0.03)),
(3, (1,2,3,'End'), (0.03, 0.03, 0.93, 0.01))],
emission_spec)
model.simulate()
num_data = len(model.state_vec)
if num_data < 5000 and num_data > 100: break
counts = {}
for state in model.state_vec:
if not state in counts:
counts[state] = 0
counts[state] += 1
if verbose: print 'Counts: %s' % str(counts)
# Generate shuffled indices for repeatable shuffling
shuffling = np.arange(num_data)
np.random.shuffle(shuffling)
for emission_rep in range(num_emission_reps):
if verbose: print 'Emission repetition %d' % emission_rep
model.emit()
for shuffled in [False, True]:
if verbose: print 'Shuffling HMM run: %s' % str(shuffled)
states = np.array(model.state_vec)
emissions = np.array(model.emission_vec)
if shuffled:
states = states[shuffling]
emissions = emissions[shuffling]
for num_block in num_blocks:
if verbose: print 'Blocks: %d' % num_block
blocks = np.array_split(np.arange(num_data), num_block)
for gamma_rep in range(num_gamma_init_reps):
if verbose: print 'Initial gamma seed: %d' % gamma_rep
init_gamma = np.array(states) - 1
run_id += 1
this_run = {}
this_run['num data'] = num_data
this_run['state rep'] = state_rep
this_run['emission rep'] = emission_rep
this_run['shuffled'] = shuffled
this_run['blocks'] = num_block
this_run['gamma init rep'] = gamma_rep
start_time = time.clock()
results = em(emissions,
dists,
blocks = blocks,
gamma_seed = gamma_rep,
init_gamma = init_gamma,
count_restart = 0.0)
pi = results['pi']
dists = results['dists']
reps = results['reps']
conv = results['converged']
run_time = time.clock() - start_time
this_run['run time'] = run_time
this_run['reps'] = reps
conv_status = conv and 'converged' or 'not converged'
this_run['convergence'] = conv_status
print 'Reps: %d (%s)' % (reps, conv_status)
print 'Time elapsed: %.2f' % run_time
if verbose: print_mixture(pi, dists)
if graphics_on:
display_densities(emissions, dists)
display_hist(emissions, dists)
act = emission_spec.values()
this_run['err mean max'] = max_error_mean(dists, act)
this_run['err mean mean'] = mean_error_mean(dists, act)
like = np.zeros(num_data)
pi_overall = np.mean(pi, 0)
for p, dist in zip(pi_overall, dists):
like += p * dist.density(states)
this_run['log likelihood'] = np.sum(np.log(like))
like = np.zeros(num_data)
for i, block in enumerate(blocks):
for p, dist in zip(pi[i], dists):
comp = p * dist.density(states[block])
like[block] += comp
this_run['log likelihood local'] = np.sum(np.log(like))
run_data[run_id] = this_run
work += 1
print 'Finished run %d/%d' % (work, total_work)
# Output data to CSV
cols = set()
for id in run_data:
for k in run_data[id]:
cols.add(k)
with open('outfile.csv', 'wb') as f:
writer = csv.writer(f)
writer.writerow(list(cols))
writer.writerows([[run_data[id][c] for c in cols] for id in run_data])
iforest.fit(X_train)
s_X_iforest = iforest.decision_function(X_test)
print('LocalOutlierFactor processing...')
lof = LocalOutlierFactor(n_neighbors=20)
lof.fit(X_train)
s_X_lof = lof.decision_function(X_test)
print('OneClassSVM processing...')
ocsvm = OneClassSVM()
ocsvm.fit(X_train[:min(ocsvm_max_train, n_samples_train - 1)])
s_X_ocsvm = ocsvm.decision_function(X_test).reshape(1, -1)[0]
s_unif_iforest = iforest.decision_function(unif)
s_unif_lof = lof.decision_function(unif)
s_unif_ocsvm = ocsvm.decision_function(unif).reshape(1, -1)[0]
plt.subplot(121)
auc_iforest, em_iforest, amax_iforest = em(t, t_max, volume_support,
s_unif_iforest, s_X_iforest,
n_generated)
auc_lof, em_lof, amax_lof = em(t, t_max, volume_support, s_unif_lof,
s_X_lof, n_generated)
auc_ocsvm, em_ocsvm, amax_ocsvm = em(t, t_max, volume_support,
s_unif_ocsvm, s_X_ocsvm, n_generated)
if amax_iforest == -1 or amax_lof == -1 or amax_ocsvm == -1:
amax = -1
else:
amax = max(amax_iforest, amax_lof, amax_ocsvm)
plt.subplot(121)
plt.plot(t[:amax],
em_iforest[:amax],
lw=1,
plt.show()
#GMM的构造
#Step 1.首先根据经验来分别对男女生的均值、方差和权值进行初始化
mu1 = 170
sigmal = 10
w1 = 0.7 #男生的
mu2 = 160
sigma2 = 10
w2 = 0.3 #以我们学校理工院校为例
d = 1
n = len(h) # 样本长度
# 开始EM算法的主循环
for iteration in range(100):
mu1, sigmal, w1, mu2, sigma2, w2 = em.em(h, mu1, sigmal, w1, mu2, sigma2,
w2)
#男生女生以及混合后身高的概率密度曲线
t = np.linspace(120, 220, 550) #500个
m = stats.norm.pdf(t, loc=mu1, scale=sigmal) # 男生分布的预测
f = stats.norm.pdf(t, loc=mu2, scale=sigma2) # 女生分别的预测
mix = w1 * m + w2 * f #混合后
plt.plot(t, m, color='b')
plt.plot(t, f, color='r')
plt.plot(t, mix, color='k')
#男生女生以及混合后身高的概率密度曲线
plt.title('Probability density curve for boys and girls and mixed height')
#plt.legend([p1,p2,p3],["male","female","mixing"],loc='upper right')
plt.legend(["male", "female", "mixing"], loc='upper right')
plt.xlabel('height/cm')
plt.ylabel('Probability') #坐标轴设置
import em
import matplotlib.pyplot as plt
print('Generating plot for first data set... \t', end='', flush=True)
# generate EM cluster plot for 1st data set
file_path = 'data/iris_flowers.csv'
x_axis = 'PetalLengthCm'
y_axis = 'SepalWidthCm'
n_clusters = 2
em.em(file_path, x_axis, y_axis, n_clusters)
print('first plot generated')
print('Generating plot for second data set... \t', end='', flush=True)
# generate EM cluster plot for 2nd data set
file_path = 'data/winequality-red.csv'
x_axis = 'citric acid'
y_axis = 'volatile acidity'
n_clusters = 2
em.em(file_path, x_axis, y_axis, n_clusters)
print('second plot generated')
# show both plots
plt.show()
