def main():
# Connect to experiment db
if args.host is not None:
client = MongoClient(host=args.host)
else:
client = MongoClient()
db = client.timb_experiments_db
fs = gridfs.GridFS(db)
# Pull out input/output sequence
o = db.experiments.find_one({'_id': ObjectId(args.id)})
from pprint import pprint
pprint(o['tracker_params'])
out_bmaps = np.empty((100, 100, len(o['log'])), dtype=np.uint32)
out_true_bmaps = np.empty((100, 100, len(o['log'])), dtype=np.uint32)
for i, log_entry in enumerate(o['log']):
entry = cPickle.loads(fs.get(log_entry['entry_file']).read())
out_true_bmaps[:,:,i] = entry['state']
trusted = timb.threshold_trusted_for_view2(entry['new_weight'])
machine_output = timb.sdf_to_zc(np.where(trusted, entry['new_phi'], np.nan))
out_bmaps[:,:,i] = machine_output
# out_bmaps = np.array(out_bmaps)
# out_true_bmaps = np.array(out_true_bmaps)
# Write i/o sequence to matlab
input_filename = '/tmp/%s.mat' % str(uuid.uuid4())
output_filename = '/tmp/%s.mat' % str(uuid.uuid4())
scipy.io.savemat(input_filename, {'bmaps': out_bmaps, 'true_bmaps': out_true_bmaps})
# Call matlab function to calculuate PR
subprocess.call(['matlab', '-nosplash', '-nodesktop', #'-nojvm', '-nodisplay',
'-r', "multiPR('%s', '%s'); exit" % (input_filename, output_filename)
], cwd=MATLAB_CWD)
output = scipy.io.loadmat(output_filename)
os.unlink(input_filename)
os.unlink(output_filename)
out_ps = np.squeeze(output['ps'])
out_rs = np.squeeze(output['rs'])
out_f1s = np.squeeze(output['f1s'])
print out_f1s
print 'average f1', np.mean(out_f1s[np.isfinite(out_f1s)])
import matplotlib.pyplot as plt
plt.plot(out_ps, label="precision")
plt.plot(out_rs, label="recall")
plt.plot(out_f1s, label="f1")
plt.legend(loc=4)
plt.ylim([0, 1])
plt.show()
fs = gridfs.GridFS(db)
o = db.experiments.find_one({'_id': ObjectId(args.id)})
entry = cPickle.loads(fs.get(o['log'][args.i]['entry_file']).read())
state = entry['state']
def sdf_to_zc(f):
p = np.pad(f, (1, 1), 'edge')
return (f * p[:-2, 1:-1] < 0) | (f * p[2:, 1:-1] < 0) | (
f * p[1:-1, :-2] < 0) | (f * p[1:-1, 2:] < 0)
trusted = timb.threshold_trusted_for_view2(entry['new_weight'])
output = sdf_to_zc(np.where(trusted, entry['new_phi'], np.nan))
import scipy.io
scipy.io.savemat(args.output, {'machine': output, 'true': state})
print 'state', state, state.shape
print 'output', output, output.shape
import sys
sys.exit(0)
import matplotlib
matplotlib.rcParams.update({'font.size': 8, 'image.origin': 'lower'})
import matplotlib.pyplot as plt
def main():
# Connect to experiment db
if args.host is not None:
client = MongoClient(host=args.host)
else:
client = MongoClient()
db = client.timb_experiments_db
fs = gridfs.GridFS(db)
# Pull out input/output sequence
o = db.experiments.find_one({'_id': ObjectId(args.id)})
from pprint import pprint
pprint(o['tracker_params'])
out_bmaps = np.empty((100, 100, len(o['log'])), dtype=np.uint32)
out_true_bmaps = np.empty((100, 100, len(o['log'])), dtype=np.uint32)
for i, log_entry in enumerate(o['log']):
entry = cPickle.loads(fs.get(log_entry['entry_file']).read())
out_true_bmaps[:, :, i] = entry['state']
trusted = timb.threshold_trusted_for_view2(entry['new_weight'])
machine_output = timb.sdf_to_zc(
np.where(trusted, entry['new_phi'], np.nan))
out_bmaps[:, :, i] = machine_output
# out_bmaps = np.array(out_bmaps)
# out_true_bmaps = np.array(out_true_bmaps)
# Write i/o sequence to matlab
input_filename = '/tmp/%s.mat' % str(uuid.uuid4())
output_filename = '/tmp/%s.mat' % str(uuid.uuid4())
scipy.io.savemat(input_filename, {
'bmaps': out_bmaps,
'true_bmaps': out_true_bmaps
})
# Call matlab function to calculuate PR
subprocess.call(
[
'matlab',
'-nosplash',
'-nodesktop', #'-nojvm', '-nodisplay',
'-r',
"multiPR('%s', '%s'); exit" % (input_filename, output_filename)
],
cwd=MATLAB_CWD)
output = scipy.io.loadmat(output_filename)
os.unlink(input_filename)
os.unlink(output_filename)
out_ps = np.squeeze(output['ps'])
out_rs = np.squeeze(output['rs'])
out_f1s = np.squeeze(output['f1s'])
print out_f1s
print 'average f1', np.mean(out_f1s[np.isfinite(out_f1s)])
import matplotlib.pyplot as plt
plt.plot(out_ps, label="precision")
plt.plot(out_rs, label="recall")
plt.plot(out_f1s, label="f1")
plt.legend(loc=4)
plt.ylim([0, 1])
plt.show()
def main():
# Connect to experiment db
if args.host is not None:
client = MongoClient(host=args.host)
else:
client = MongoClient()
db = client.timb_experiments_db
fs = gridfs.GridFS(db)
# Pull out parts of the sequence
ids = args.id.split(',')
baseline_ids = args.baseline_ids.split(',')
import matplotlib
matplotlib.rcParams.update({'font.size': 8, 'image.origin': 'lower'})
import matplotlib.pyplot as plt
ROWS = 3
plt.figure(figsize=(args.num_steps, ROWS * len(ids)), dpi=100)
plt.subplots_adjust(wspace=.001, hspace=.001)
for id_ind, id in enumerate(ids):
o = db.experiments.find_one({'_id': ObjectId(id)})
id_baseline = baseline_ids[id_ind]
o_baseline = db.experiments.find_one({'_id': ObjectId(id_baseline)})
step_inds = np.floor(np.linspace(1,
len(o['log']) - 2,
args.num_steps)).astype(int)
entries = [
cPickle.loads(fs.get(o['log'][i]['entry_file']).read())
for i in step_inds
]
entries_baseline = [
cPickle.loads(fs.get(o_baseline['log'][i]['entry_file']).read())
for i in step_inds
]
# def plot_binary(img):
# img = skimage.morphology.skeletonize(img)
# plt.imshow(img.T, aspect=1, origin='lower', cmap='binary_r', interpolation='spline36')
def plot_field(f, img=True, contour=False, colors='k'):
assert img or contour
if img:
plt.imshow(f.T,
aspect=1,
vmin=-1,
vmax=1,
cmap='bwr',
origin='lower')
else:
plt.imshow(np.zeros_like(f),
aspect=1,
vmin=-1,
vmax=1,
cmap='bwr',
origin='lower')
if contour:
x = np.linspace(0, f.shape[0] - 1, f.shape[0])
y = np.linspace(0, f.shape[1] - 1, f.shape[1])
X, Y = np.meshgrid(x, y, indexing='ij')
plt.contour(X, Y, f, levels=[0], colors=colors)
for i, e in enumerate(entries):
plt.subplot(ROWS * len(ids), len(entries),
i + 1 + len(entries) * (0 + ROWS * id_ind))
plt.axis('off')
img = np.empty(e['state'].shape + (3, ), dtype=np.uint8)
for k in range(3):
img[:, :, k] = np.where(e['state'], 0, 255)
plot_field(get_sdf(img), False, True)
e_baseline = entries_baseline[i]
plt.subplot(ROWS * len(ids), len(entries),
i + 1 + len(entries) * (1 + ROWS * id_ind))
plt.axis('off')
trusted = timb.threshold_trusted_for_view2(
e_baseline['new_weight'])
baseline_output = np.where(trusted, e_baseline['new_phi'], np.nan)
plot_field(baseline_output, False, True, colors='r')
plt.subplot(ROWS * len(ids), len(entries),
i + 1 + len(entries) * (2 + ROWS * id_ind))
plt.axis('off')
trusted = timb.threshold_trusted_for_view2(e['new_weight'])
machine_output = np.where(trusted, e['new_phi'], np.nan)
plot_field(machine_output, False, True, colors='b')
plt.savefig('out.pdf', bbox_inches='tight')
plt.show()
else:
client = MongoClient()
db = client.timb_experiments_db
fs = gridfs.GridFS(db)
o = db.experiments.find_one({'_id': ObjectId(args.id)})
entry = cPickle.loads(fs.get(o['log'][args.i]['entry_file']).read())
state = entry['state']
def sdf_to_zc(f):
p = np.pad(f, (1,1), 'edge')
return (f*p[:-2,1:-1] < 0) | (f*p[2:,1:-1] < 0) | (f*p[1:-1,:-2] < 0) | (f*p[1:-1,2:] < 0)
trusted = timb.threshold_trusted_for_view2(entry['new_weight'])
output = sdf_to_zc(np.where(trusted, entry['new_phi'], np.nan))
import scipy.io
scipy.io.savemat(args.output, {'machine':output, 'true':state})
print 'state', state, state.shape
print 'output', output, output.shape
import sys
sys.exit(0)
import matplotlib
matplotlib.rcParams.update({'font.size': 8, 'image.origin': 'lower'})
def main():
# Connect to experiment db
if args.host is not None:
client = MongoClient(host=args.host)
else:
client = MongoClient()
db = client.timb_experiments_db
fs = gridfs.GridFS(db)
# Pull out parts of the sequence
ids = args.id.split(',')
baseline_ids = args.baseline_ids.split(',')
import matplotlib
matplotlib.rcParams.update({'font.size': 8, 'image.origin': 'lower'})
import matplotlib.pyplot as plt
ROWS = 3
plt.figure(figsize=(args.num_steps, ROWS*len(ids)), dpi=100)
plt.subplots_adjust(wspace=.001, hspace=.001)
for id_ind, id in enumerate(ids):
o = db.experiments.find_one({'_id': ObjectId(id)})
id_baseline = baseline_ids[id_ind]
o_baseline = db.experiments.find_one({'_id': ObjectId(id_baseline)})
step_inds = np.floor(np.linspace(1, len(o['log'])-2, args.num_steps)).astype(int)
entries = [cPickle.loads(fs.get(o['log'][i]['entry_file']).read()) for i in step_inds]
entries_baseline = [cPickle.loads(fs.get(o_baseline['log'][i]['entry_file']).read()) for i in step_inds]
# def plot_binary(img):
# img = skimage.morphology.skeletonize(img)
# plt.imshow(img.T, aspect=1, origin='lower', cmap='binary_r', interpolation='spline36')
def plot_field(f, img=True, contour=False, colors='k'):
assert img or contour
if img:
plt.imshow(f.T, aspect=1, vmin=-1, vmax=1, cmap='bwr', origin='lower')
else:
plt.imshow(np.zeros_like(f), aspect=1, vmin=-1, vmax=1, cmap='bwr', origin='lower')
if contour:
x = np.linspace(0, f.shape[0]-1, f.shape[0])
y = np.linspace(0, f.shape[1]-1, f.shape[1])
X, Y = np.meshgrid(x, y, indexing='ij')
plt.contour(X, Y, f, levels=[0], colors=colors)
for i, e in enumerate(entries):
plt.subplot(ROWS*len(ids), len(entries), i+1 + len(entries)*(0 + ROWS*id_ind))
plt.axis('off')
img = np.empty(e['state'].shape+(3,), dtype=np.uint8)
for k in range(3):
img[:,:,k] = np.where(e['state'], 0, 255)
plot_field(get_sdf(img), False, True)
e_baseline = entries_baseline[i]
plt.subplot(ROWS*len(ids), len(entries), i+1 + len(entries)*(1 + ROWS*id_ind))
plt.axis('off')
trusted = timb.threshold_trusted_for_view2(e_baseline['new_weight'])
baseline_output = np.where(trusted, e_baseline['new_phi'], np.nan)
plot_field(baseline_output, False, True, colors='r')
plt.subplot(ROWS*len(ids), len(entries), i+1 + len(entries)*(2 + ROWS*id_ind))
plt.axis('off')
trusted = timb.threshold_trusted_for_view2(e['new_weight'])
machine_output = np.where(trusted, e['new_phi'], np.nan)
plot_field(machine_output, False, True, colors='b')
plt.savefig('out.pdf', bbox_inches='tight')
plt.show()
