def msrc_over_weak(n_train_full=40, n_train=276,
C=100, alpha=0.1,
test_method='gco', test_n_iter=5, n_iter=5,
max_iter=1000, verbose=1,
check_every=50, complete_every=100, update_w_every=50,
relaxed_test=False,
use_latent_first_iter=100):
# save parameters as meta
meta_data = locals()
logger = logging.getLogger(__name__)
crf_test = HCRF(n_states=24, n_features=2028, n_edge_features=4, alpha=alpha,
inference_method=test_method, n_iter=test_n_iter)
crf_latent = HCRF(n_states=24, n_features=2028, n_edge_features=4, alpha=alpha,
inference_method='gco', n_iter=n_iter)
trainer = OverWeak(crf_latent, n_states=24, n_features=2028, n_edge_features=4,
C=C, alpha=alpha,
max_iter=max_iter, verbose=verbose,
check_every=check_every, complete_every=complete_every, update_w_every=update_w_every)
x_train, y_train, y_train_full, x_test, y_test = load_msrc(n_train_full, n_train, dense=True)
logger.info('start training')
start = time()
trainer.fit(x_train, y_train,
train_scorer=lambda w: compute_score(crf_test, w, x_train, y_train_full, relaxed=relaxed_test),
test_scorer=lambda w: compute_score(crf_test, w, x_test, y_test, relaxed=relaxed_test),
use_latent_first_iter=use_latent_first_iter)
stop = time()
time_elapsed = stop - start
logger.info('testing')
test_score = compute_score(crf_test, trainer.w, x_test, y_test)
train_score = compute_score(crf_test, trainer.w, x_train, y_train_full)
logger.info('========================================')
logger.info('train score: %f', train_score)
logger.info('test score: %f', test_score)
exp_data = {}
exp_data['timestamps'] = trainer.timestamps
exp_data['objective'] = trainer.objective_curve
exp_data['w'] = trainer.w
exp_data['train_scores'] = trainer.train_score
exp_data['test_scores'] = trainer.test_score
exp_data['w_history'] = trainer.w_history
meta_data['dataset_name'] = 'msrc'
meta_data['annotation_type'] = 'full+weak'
meta_data['label_type'] = 'image-level labelling'
meta_data['trainer'] = 'komodakis+latent+kappa'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
return ExperimentResult(exp_data, meta_data)
def msrc_over(n_train=276, C=100,
max_iter=1000, check_every=50, verbose=1,
test_method='gco', test_n_iter=5, relaxed_test=False):
# save parameters as meta
meta_data = locals()
logger = logging.getLogger(__name__)
crf = HCRF(n_states=24, n_features=2028, n_edge_features=4, alpha=1,
inference_method=test_method, n_iter=test_n_iter)
trainer = Over(n_states=24, n_features=2028, n_edge_features=4,
C=C, max_iter=max_iter, verbose=verbose, check_every=check_every)
x_train, y_train, y_train_full, x_test, y_test = \
load_msrc(n_train, n_train)
logger.info('start training')
start = time()
trainer.fit(x_train, y_train_full,
train_scorer=lambda w: compute_score(crf, w, x_train, y_train, invert=True, relaxed=relaxed_test),
test_scorer=lambda w: compute_score(crf, w, x_test, y_test, invert=True, relaxed=relaxed_test))
stop = time()
time_elapsed = stop - start
logger.info('testing')
test_score = compute_score(crf, trainer.w, x_test, y_test, invert=True, relaxed=relaxed_test)
train_score = compute_score(crf, trainer.w, x_train, y_train, invert=True, relaxed=relaxed_test)
logger.info('========================================')
logger.info('train score: %f', train_score)
logger.info('test score: %f', test_score)
exp_data = {}
exp_data['timestamps'] = trainer.timestamps
exp_data['objective'] = trainer.objective_curve
exp_data['w'] = trainer.w
exp_data['train_scores'] = trainer.train_score
exp_data['test_scores'] = trainer.test_score
exp_data['w_history'] = trainer.w_history
meta_data['dataset_name'] = 'msrc'
meta_data['annotation_type'] = 'full'
meta_data['label_type'] = 'full'
meta_data['trainer'] = 'komodakis'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
return ExperimentResult(exp_data, meta_data)
def msrc_full_fw(n_train=276, C=100, max_iter=500, check_dual_every=50,
inference_method='gco', n_inference_iter=5):
# save parameters as meta
meta_data = locals()
logger = logging.getLogger(__name__)
crf = HCRF(n_states=24, n_features=2028, n_edge_features=4, alpha=1,
inference_method=inference_method, n_iter=n_inference_iter)
clf = FrankWolfeSSVM(crf, verbose=2, n_jobs=1,
check_dual_every=check_dual_every, max_iter=max_iter, C=C)
x_train, y_train, y_train_full, x_test, y_test = \
load_msrc(n_train, n_train)
logger.info('start training')
start = time()
clf.fit(x_train, y_train, Xtest=x_test, Ytest=y_test)
stop = time()
train_score = clf.score(x_train, y_train_full)
test_score = clf.score(x_test, y_test)
time_elapsed = stop - start
logger.info('============================================================')
logger.info('Score on train set: %f', train_score)
logger.info('Score on test set: %f', test_score)
logger.info('Elapsed time: %f s', time_elapsed)
exp_data = {}
exp_data['timestamps'] = clf.timestamps_
exp_data['primal_objective'] = clf.primal_objective_curve_
exp_data['objective'] = clf.objective_curve_
exp_data['w_history'] = clf.w_history
exp_data['test_scores'] = clf.test_scores
exp_data['train_scores'] = clf.train_scores
exp_data['w'] = clf.w
meta_data['dataset_name'] = 'syntetic'
meta_data['annotation_type'] = 'full'
meta_data['label_type'] = 'full'
meta_data['trainer'] = 'frank-wolfe'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
return ExperimentResult(exp_data, meta_data)
def msrc_over_weak(n_train_full=40,
n_train=276,
C=100,
alpha=0.1,
test_method='gco',
test_n_iter=5,
n_iter=5,
max_iter=1000,
verbose=1,
check_every=50,
complete_every=100,
update_w_every=50,
relaxed_test=False,
use_latent_first_iter=100):
# save parameters as meta
meta_data = locals()
logger = logging.getLogger(__name__)
crf_test = HCRF(n_states=24,
n_features=2028,
n_edge_features=4,
alpha=alpha,
inference_method=test_method,
n_iter=test_n_iter)
crf_latent = HCRF(n_states=24,
n_features=2028,
n_edge_features=4,
alpha=alpha,
inference_method='gco',
n_iter=n_iter)
trainer = OverWeak(crf_latent,
n_states=24,
n_features=2028,
n_edge_features=4,
C=C,
alpha=alpha,
max_iter=max_iter,
verbose=verbose,
check_every=check_every,
complete_every=complete_every,
update_w_every=update_w_every)
x_train, y_train, y_train_full, x_test, y_test = load_msrc(n_train_full,
n_train,
dense=True)
logger.info('start training')
start = time()
trainer.fit(x_train,
y_train,
train_scorer=lambda w: compute_score(
crf_test, w, x_train, y_train_full, relaxed=relaxed_test),
test_scorer=lambda w: compute_score(
crf_test, w, x_test, y_test, relaxed=relaxed_test),
use_latent_first_iter=use_latent_first_iter)
stop = time()
time_elapsed = stop - start
logger.info('testing')
test_score = compute_score(crf_test, trainer.w, x_test, y_test)
train_score = compute_score(crf_test, trainer.w, x_train, y_train_full)
logger.info('========================================')
logger.info('train score: %f', train_score)
logger.info('test score: %f', test_score)
exp_data = {}
exp_data['timestamps'] = trainer.timestamps
exp_data['objective'] = trainer.objective_curve
exp_data['w'] = trainer.w
exp_data['train_scores'] = trainer.train_score
exp_data['test_scores'] = trainer.test_score
exp_data['w_history'] = trainer.w_history
meta_data['dataset_name'] = 'msrc'
meta_data['annotation_type'] = 'full+weak'
meta_data['label_type'] = 'image-level labelling'
meta_data['trainer'] = 'komodakis+latent+kappa'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
return ExperimentResult(exp_data, meta_data)
def msrc_over(n_train=276,
C=100,
max_iter=1000,
check_every=50,
verbose=1,
test_method='gco',
test_n_iter=5,
relaxed_test=False):
# save parameters as meta
meta_data = locals()
logger = logging.getLogger(__name__)
crf = HCRF(n_states=24,
n_features=2028,
n_edge_features=4,
alpha=1,
inference_method=test_method,
n_iter=test_n_iter)
trainer = Over(n_states=24,
n_features=2028,
n_edge_features=4,
C=C,
max_iter=max_iter,
verbose=verbose,
check_every=check_every)
x_train, y_train, y_train_full, x_test, y_test = \
load_msrc(n_train, n_train)
logger.info('start training')
start = time()
trainer.fit(
x_train,
y_train_full,
train_scorer=lambda w: compute_score(
crf, w, x_train, y_train, invert=True, relaxed=relaxed_test),
test_scorer=lambda w: compute_score(
crf, w, x_test, y_test, invert=True, relaxed=relaxed_test))
stop = time()
time_elapsed = stop - start
logger.info('testing')
test_score = compute_score(crf,
trainer.w,
x_test,
y_test,
invert=True,
relaxed=relaxed_test)
train_score = compute_score(crf,
trainer.w,
x_train,
y_train,
invert=True,
relaxed=relaxed_test)
logger.info('========================================')
logger.info('train score: %f', train_score)
logger.info('test score: %f', test_score)
exp_data = {}
exp_data['timestamps'] = trainer.timestamps
exp_data['objective'] = trainer.objective_curve
exp_data['w'] = trainer.w
exp_data['train_scores'] = trainer.train_score
exp_data['test_scores'] = trainer.test_score
exp_data['w_history'] = trainer.w_history
meta_data['dataset_name'] = 'msrc'
meta_data['annotation_type'] = 'full'
meta_data['label_type'] = 'full'
meta_data['trainer'] = 'komodakis'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
return ExperimentResult(exp_data, meta_data)
def msrc_full_fw(n_train=276,
C=100,
max_iter=500,
check_dual_every=50,
inference_method='gco',
n_inference_iter=5):
# save parameters as meta
meta_data = locals()
logger = logging.getLogger(__name__)
crf = HCRF(n_states=24,
n_features=2028,
n_edge_features=4,
alpha=1,
inference_method=inference_method,
n_iter=n_inference_iter)
clf = FrankWolfeSSVM(crf,
verbose=2,
n_jobs=1,
check_dual_every=check_dual_every,
max_iter=max_iter,
C=C)
x_train, y_train, y_train_full, x_test, y_test = \
load_msrc(n_train, n_train)
logger.info('start training')
start = time()
clf.fit(x_train, y_train, Xtest=x_test, Ytest=y_test)
stop = time()
train_score = clf.score(x_train, y_train_full)
test_score = clf.score(x_test, y_test)
time_elapsed = stop - start
logger.info('============================================================')
logger.info('Score on train set: %f', train_score)
logger.info('Score on test set: %f', test_score)
logger.info('Elapsed time: %f s', time_elapsed)
exp_data = {}
exp_data['timestamps'] = clf.timestamps_
exp_data['primal_objective'] = clf.primal_objective_curve_
exp_data['objective'] = clf.objective_curve_
exp_data['w_history'] = clf.w_history
exp_data['test_scores'] = clf.test_scores
exp_data['train_scores'] = clf.train_scores
exp_data['w'] = clf.w
meta_data['dataset_name'] = 'syntetic'
meta_data['annotation_type'] = 'full'
meta_data['label_type'] = 'full'
meta_data['trainer'] = 'frank-wolfe'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
return ExperimentResult(exp_data, meta_data)
def msrc_weak(n_full=20,
n_train=276,
C=100,
latent_iter=25,
max_iter=500,
inner_tol=0.001,
outer_tol=0.01,
min_changes=0,
initialize=True,
alpha=0.1,
n_inference_iter=5,
inactive_window=50,
inactive_threshold=1e-5,
warm_start=False,
inference_cache=0,
save_inner_w=False,
inference_method='gco'):
meta_data = locals()
logger = logging.getLogger(__name__)
crf = HCRF(n_states=24,
n_features=2028,
n_edge_features=4,
alpha=alpha,
inference_method=inference_method,
n_iter=n_inference_iter)
base_clf = OneSlackSSVM(crf,
verbose=2,
n_jobs=4,
tol=inner_tol,
max_iter=max_iter,
C=C,
inference_cache=inference_cache,
inactive_window=inactive_window,
inactive_threshold=inactive_threshold)
clf = LatentSSVM(base_clf,
latent_iter=latent_iter,
verbose=2,
tol=outer_tol,
min_changes=min_changes,
n_jobs=4)
x_train, y_train, y_train_full, x_test, y_test = \
load_msrc(n_full, n_train)
start = time()
clf.fit(x_train,
y_train,
initialize=initialize,
warm_start=warm_start,
save_inner_w=save_inner_w)
stop = time()
train_score = clf.score(x_train, y_train_full)
test_score = clf.score(x_test, y_test)
time_elapsed = stop - start
logger.info('============================================================')
logger.info('Score on train set: %f', train_score)
logger.info('Score on test set: %f', test_score)
logger.info('Norm of weight vector: |w|=%f', np.linalg.norm(clf.w))
logger.info('Elapsed time: %f s', time_elapsed)
test_scores = []
for score in clf.staged_score(x_test, y_test):
test_scores.append(score)
train_scores = []
for score in clf.staged_score(x_train, y_train_full):
train_scores.append(score)
raw_scores = []
for score in clf.staged_score2(x_train, y_train):
raw_scores.append(score)
exp_data = clf._get_data()
exp_data['test_scores'] = np.array(test_scores)
exp_data['train_scores'] = np.array(train_scores)
exp_data['raw_scores'] = np.array(raw_scores)
meta_data['dataset_name'] = 'msrc'
meta_data['annotation_type'] = 'image-level labelling'
meta_data['label_type'] = 'full+weak'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
meta_data['iter_done'] = clf.iter_done
return ExperimentResult(exp_data, meta_data)
def msrc_weak(n_full=20, n_train=276, C=100, latent_iter=25,
max_iter=500, inner_tol=0.001, outer_tol=0.01, min_changes=0,
initialize=True, alpha=0.1, n_inference_iter=5,
inactive_window=50, inactive_threshold=1e-5,
warm_start=False, inference_cache=0,
save_inner_w=False, inference_method='gco'):
meta_data = locals()
logger = logging.getLogger(__name__)
crf = HCRF(n_states=24, n_features=2028, n_edge_features=4, alpha=alpha,
inference_method=inference_method, n_iter=n_inference_iter)
base_clf = OneSlackSSVM(crf, verbose=2, n_jobs=4,
tol=inner_tol, max_iter=max_iter, C=C,
inference_cache=inference_cache,
inactive_window=inactive_window,
inactive_threshold=inactive_threshold)
clf = LatentSSVM(base_clf, latent_iter=latent_iter, verbose=2,
tol=outer_tol, min_changes=min_changes, n_jobs=4)
x_train, y_train, y_train_full, x_test, y_test = \
load_msrc(n_full, n_train)
start = time()
clf.fit(x_train, y_train,
initialize=initialize,
warm_start=warm_start,
save_inner_w=save_inner_w)
stop = time()
train_score = clf.score(x_train, y_train_full)
test_score = clf.score(x_test, y_test)
time_elapsed = stop - start
logger.info('============================================================')
logger.info('Score on train set: %f', train_score)
logger.info('Score on test set: %f', test_score)
logger.info('Norm of weight vector: |w|=%f', np.linalg.norm(clf.w))
logger.info('Elapsed time: %f s', time_elapsed)
test_scores = []
for score in clf.staged_score(x_test, y_test):
test_scores.append(score)
train_scores = []
for score in clf.staged_score(x_train, y_train_full):
train_scores.append(score)
raw_scores = []
for score in clf.staged_score2(x_train, y_train):
raw_scores.append(score)
exp_data = clf._get_data()
exp_data['test_scores'] = np.array(test_scores)
exp_data['train_scores'] = np.array(train_scores)
exp_data['raw_scores'] = np.array(raw_scores)
meta_data['dataset_name'] = 'msrc'
meta_data['annotation_type'] = 'image-level labelling'
meta_data['label_type'] = 'full+weak'
meta_data['train_score'] = train_score
meta_data['test_score'] = test_score
meta_data['time_elapsed'] = time_elapsed
meta_data['iter_done'] = clf.iter_done
return ExperimentResult(exp_data, meta_data)