w = process_data(w,
decomposable=('orthography',),
decomposable_names=('letters',),
feature_layers=('letters',),
feature_sets=('fourteen',),
negative_features=True,
length_adaptation=False)
rla = {k: 'global' for k in {'letters-features', 'letters'}}
rla['orthography'] = 'frequency'
s = Builder(IA_WEIGHTS,
rla,
-.05,
outputs=('orthography',),
monitors=('orthography',),
step_size=.5,
weight_adaptation=True)
m = s.build_model(w)
result = m.activate(w,
max_cycles=n_cyc,
threshold=.7,
strict=False)
cycles = np.array([len(x['orthography']) for x in result])
right = cycles == n_cyc
cycles[right] = -1
for word, c in zip(w, cycles):
results.append([word['orthography'][0],
weights = deepcopy(IA_WEIGHTS)
# Manually adapt weights to length 4
if not length_adaptation:
weights[("letters", "orthography")][0] /= 4
weights[("letters", "orthography")][1] *= 4
weights[("orthography", "letters")][0] /= 4
weights[("orthography", "letters")][1] *= 4
names = set(chain.from_iterable(weights))
rla = {k: 'global' for k in names}
rla['orthography'] = 'frequency'
s = Builder(weights,
rla,
-.05,
outputs=('orthography', ),
monitors=('orthography', ),
step_size=.5,
weight_adaptation=length_adaptation)
m = s.build_model(w)
result = m.activate(w,
max_cycles=n_cyc,
threshold=.7,
strict=False)
cycles = np.array([len(x['orthography']) for x in result])
right = cycles == n_cyc
cycles[right] = -1
for x, word, c in zip(result, w, cycles):
results.append([
word['orthography'], (idx * 100) + idx_2, word['rt'],