def main():
print("\n", "/" * 120, "\n")
print(d.now())
print("\nOUTPUTS DIRECTORY:\n{}\n".format(activity_dir))
print("\nTST H5:\n{}\n".format(tst_h5))
tst_ip = ip.for_callids(
tst_h5,
callids=tst_callids,
data_context=dctx,
add_channel_at_end=add_channel_dim,
label_subcontext=lctx,
label_from_subcontext_fn=lctx_fn,
steps_per_chunk=steps_per_chunk,
classkeyfn=np.argmax, # for categorical labels
class_subsample_to_ratios=tst_class_subsampling,
shuffle_seed=None, # never shuffled
npasses=1,
mean_it=mean_it,
std_it=std_it,
)
print(
"{}: max-totlen: {:,}; nchunks: {:,}; steps_per_pass: {:,}; npasses: {:,}"
.format("TST", tst_ip.totlen, tst_ip.nchunks, tst_ip.steps_per_pass,
tst_ip.npasses))
print("data shape: {}; label shape: {}".format(tst_ip.inputdatashape,
tst_ip.inputlabelshape))
with hFile(priors_fp, 'r') as pf:
init = np.array(pf['init'])
tran = np.array(pf['tran'])
priors = np.array(pf['priors'])
init = init / init.sum()
tran = nu.normalize_confusion_matrix(tran)[1]
priors = priors / priors.sum()
print("\n", "/" * 120, "\n")
print("DONE SETTING PRIORS - NOW, MODEL SUMMARY")
kmodel.summary()
print("\n", "/" * 120, "\n")
print("PREDICTING ON TEST")
predict_on_inputs_provider(
model=kmodel,
inputs_provider=tst_ip,
subsampling=sub,
export_to_dir=os.path.join(activity_dir, activity_name),
init=init,
tran=tran,
priors=priors,
)
print("\nDONE")
print(d.now())
def test_predP_batB_seqlQ_normconfmat(predP_batB_seqlQ_cls3_preds_confmat):
provider = predP_batB_seqlQ_cls3_preds_confmat
confmat, confrecall, confprecision = [
provider[k] for k in ['confmat', 'confrecall', 'confprecision']
]
print(confmat.shape)
confprecp, confrecp = nu.normalize_confusion_matrix(confmat)
assert_almost_equal(confprecp, confprecision)
assert_almost_equal(confrecp, confrecall)
confmatg, confrecallg, confprecisiong = [
provider[k] for k in ['confmatg', 'confrecallg', 'confprecisiong']
]
print("G:", confmatg.shape)
confprecp, confrecp = nu.normalize_confusion_matrix(confmatg)
assert_almost_equal(confprecp, confprecisiong)
assert_almost_equal(confrecp, confrecallg)
def predict_on_inputs_provider( # pylint: disable=too-many-locals,too-many-statements
model, inputs_provider, export_to, init, tran):
def _save(paths, datas):
with hFile(export_to, 'a') as f:
for path, data in zip(paths, datas):
if path not in f.keys():
f.create_dataset(path,
data=data,
compression='lzf',
fletcher32=True)
f.flush()
currn = None
ctrue = []
cpred = []
tot_conf = None
tot_conf_vp = None
for xy, (_, chunking) in inputs_provider.flow(
indefinitely=False,
only_labels=False,
with_chunking=True,
):
ctrue.append(xy[1])
cpred.append(model.predict_on_batch(xy[0]))
if currn is None:
currn = chunking.labelpath
continue
if chunking.labelpath != currn:
t = np.concatenate(ctrue[:-1])
p = np.concatenate(cpred[:-1])
if sub != 'keepzero': # from activity_name above
z = t[:, 0].astype(bool)
p[z, 0] = 1.
p[z, 1:] = 0.
# raw confusion
conf = nu.confusion_matrix_forcategorical(
t, nu.to_categorical(p.argmax(axis=-1), nclasses=t.shape[-1]))
# viterbi decoded - no scaling
vp = lu.viterbi_smoothing(p, init, tran)
conf_vp = nu.confusion_matrix_forcategorical(
t, nu.to_categorical(vp, nclasses=t.shape[-1]))
_save(
paths=["{}/{}".format(_p, currn) for _p in ('raw', 'viterbi')],
datas=[conf, conf_vp],
)
print(currn, end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(conf),
onlydiag=True)
if tot_conf is None:
tot_conf = conf
tot_conf_vp = conf_vp
else:
tot_conf += conf
tot_conf_vp += conf_vp
currn = chunking.labelpath
ctrue = ctrue[-1:]
cpred = cpred[-1:]
# last chunking
t = np.concatenate(ctrue)
p = np.concatenate(cpred)
if sub != 'keepzero': # from activity_name above
z = t[:, 0].astype(bool)
p[z, 0] = 1.
p[z, 1:] = 0.
conf = nu.confusion_matrix_forcategorical(
t, nu.to_categorical(p.argmax(axis=-1), nclasses=t.shape[-1]))
vp = lu.viterbi_smoothing(p, init, tran)
conf_vp = nu.confusion_matrix_forcategorical(
t, nu.to_categorical(vp, nclasses=t.shape[-1]))
_save(
paths=["{}/{}".format(_p, currn) for _p in ('raw', 'viterbi')],
datas=[conf, conf_vp],
)
print(currn, end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(conf), onlydiag=True)
tot_conf += conf
tot_conf_vp += conf_vp
# print out total-statistics
_save(
paths=["{}/{}".format(_p, 'final') for _p in ('raw', 'viterbi')],
datas=[tot_conf, tot_conf_vp],
)
print("\nFINAL - RAW", end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(tot_conf), onlydiag=False)
print("\nFINAL - VITERBI", end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(tot_conf_vp),
onlydiag=False)
def predict_on_inputs_provider(model, inputs_provider, subsampling,
export_to_dir, init, tran, priors):
export_to = os.path.join(export_to_dir, "confs.h5")
def _save(paths, datas):
with hFile(export_to, 'a') as f:
for path, data in zip(paths, datas):
if path not in f.keys():
f.create_dataset(path,
data=data,
compression='lzf',
fletcher32=True)
f.flush()
currn = None
ctrue = []
cpred = []
tot_conf = None
tot_conf_vp = None
tot_conf_svp = None
for xy, (_, chunking) in inputs_provider.flow(
indefinitely=False,
only_labels=False,
with_chunking=True,
):
ctrue.append(xy[1])
cpred.append(model.predict_on_batch(xy[0]))
if currn is None:
currn = chunking.labelpath
continue
if chunking.labelpath != currn:
t = np.concatenate(ctrue[:-1])
p = np.concatenate(cpred[:-1])
if subsampling != 'nosub':
z = t[:, 0].astype(bool)
p[z, 0] = 1.
p[z, 1:] = 0.
# raw confusion
conf = nu.confusion_matrix_forcategorical(
t, nu.to_categorical(p.argmax(axis=-1), nclasses=t.shape[-1]))
# viterbi decoded - no scaling
vp = viterbi(p, init, tran, priors=None)
conf_vp = nu.confusion_matrix_forcategorical(
t, nu.to_categorical(vp, nclasses=t.shape[-1]))
# viterbi decoded - scaling
vp = viterbi(p, init, tran, priors=priors)
conf_svp = nu.confusion_matrix_forcategorical(
t, nu.to_categorical(vp, nclasses=t.shape[-1]))
_save(
paths=[
"{}/{}".format(_p, currn)
for _p in ('raw', 'viterbi', 'sviterbi')
],
datas=[conf, conf_vp, conf_svp],
)
print(currn, end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(conf),
onlydiag=True)
if tot_conf is None:
tot_conf = conf
tot_conf_vp = conf_vp
tot_conf_svp = conf_svp
else:
tot_conf += conf
tot_conf_vp += conf_vp
tot_conf_svp += conf_svp
currn = chunking.labelpath
ctrue = ctrue[-1:]
cpred = cpred[-1:]
_save(
paths=[
"{}/{}".format(_p, 'final')
for _p in ('raw', 'viterbi', 'sviterbi')
],
datas=[tot_conf, tot_conf_vp, tot_conf_svp],
)
print("\nFINAL - RAW", end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(tot_conf), onlydiag=False)
print("\nFINAL - VITERBI", end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(tot_conf_vp),
onlydiag=False)
print("\nFINAL - VITERBI - SCALED", end=' ')
nu.print_prec_rec(*nu.normalize_confusion_matrix(tot_conf_svp),
onlydiag=False)