with open(os.path.join("data/out", "train.pkl"), "rb") as f:
train = pickle.load(f)
# trainer = pycrfsuite.Trainer(verbose=True)
# trainer.set_params({'c1': 0.1, # coefficient for L1 penalty
# 'c2': 0.1, # coefficient for L2 penalty
# 'max_iterations': 250, # stop earlier
# 'feature.possible_transitions': False,
# 'feature.possible_states': False
# })
# trainer = pycrfsuite.Trainer(algorithm = 'ap',verbose=True)
trainer = pycrfsuite.Trainer(algorithm = 'pa',verbose=True)
#rainer = pycrfsuite.Trainer(algorithm = 'arow',verbose=True)
trainer.set_params({
'type':3,
'c': 0.1, # coefficient for L1 penalty
# 'c2': 0.01, # coefficient for L2 penalty
'max_iterations': 2000,
'feature.possible_transitions': False,
'feature.possible_states': False
})
for i, data in enumerate(train):
temp = prepare_data(data)
for features, ylabel in temp:
trainer.append(features, ylabel)
trainer.train("exp_{}".format(i))
print("Model {} Trained".format(i))
elif f['is_ball']:
p = "I-ball"
elif f['is_face']:
p = "I-face"
elif f['is_key']:
p = "I-key"
pred.append(p)
return pred
support_threshold = 0
stats = StatsManager(support_threshold)
for i, data in enumerate(validation):
y_pred = []
y_true = []
for features, ylabel in prepare_data(data):
y_pred.append(ftag(features))
y_true.append(ylabel)
# print(y_true[0][0:50])
# print("predict:", y_pred[0][0:50])
stats.append_report(y_true, y_pred)
print("Multi-class Classification Report Mean(Std)")
report, summary = stats.summarize()
# pretty_print_report(report)
# print("Run-length Report")
# rl_report = stats.runlength_report()
# pretty_rl_table(rl_report)