def main(train, dev, test, emb_path, hyperparams, run_id=None, res_path=None):
with open(train, 'rb') as fid:
X_train, Y_train, vocabulary, _ = cPickle.load(fid)
with open(dev, 'rb') as fid:
X_dev, Y_dev, _, _ = cPickle.load(fid)
with open(test, 'rb') as fid:
test_x, test_y, _, _ = cPickle.load(fid)
E, _ = embeddings.read_embeddings(emb_path, wrd2idx=vocabulary)
nn = nlse.NLSE(E, **hyperparams)
nn.fit(X_train, Y_train, X_dev, Y_dev)
y_hat = nn.predict(test_x)
avgF1 = f1_score(test_y, y_hat, average="macro")
acc = accuracy_score(test_y, y_hat)
run_id = run_id
dataset = os.path.basename(test)
hp = {p: hyperparams[p] for p in ["sub_size", "lrate"]}
if run_id is None: run_id = "NLSE"
results = {"acc":round(acc,3), \
"avgF1":round(avgF1,3), \
"model":"NLSE", \
"dataset":dataset, \
"run_id":run_id,
"sub_size":hyperparams["sub_size"],
"lrate":hyperparams["lrate"]}
cols = ["dataset", "model", "run_id", "acc", "avgF1", "sub_size"]
helpers.print_results(
results,
columns=["dataset", "run_id", "lrate", "subsize", "acc", "avgF1"])
if res_path is not None:
helpers.save_results(results, res_path, sep="\t", columns=cols)
return results
def hypertune(train, dev, emb_path, obj, hyperparams, res_path=None):
with open(train, 'rb') as fid:
X_train, Y_train, vocabulary, _ = cPickle.load(fid)
with open(dev, 'rb') as fid:
X_dev, Y_dev, _, _ = cPickle.load(fid)
E, _ = embeddings.read_embeddings(emb_path, wrd2idx=vocabulary)
best_hp = None
best_score = 0
for hp in hyperparams:
#initialize model with the hyperparameters
nn = nlse.NLSE(E, **hp)
nn.fit(X_train, Y_train, X_dev, Y_dev, silent=False)
Y_hat = nn.predict(X_dev)
score = obj(Y_dev, Y_hat)
print "[score: {} | hyperparameters: {}]".format(score, repr(hp))
if score > best_score:
best_score = score
best_hp = hp
results = {"score": round(score, 3), "hyper": repr(hp)}
if res_path is not None:
helpers.save_results(results, res_path, sep="\t")
helpers.print_results(results)
print ""
print "[best conf: {} | score: {}]".format(repr(best_hp), best_score)
return best_hp, best_score
def main(lex_path, test, label_map, run_id, conf={}, dev=None, res_path=None):
#read test data
dt = read_dataset(test, labels=label_map.keys())
X_test = [x[1] for x in dt]
Y_test = [label_map[x[0]] for x in dt]
#model
model = LexiconSentiment(path=lex_path,**conf)
#if dev data is passed, use this data to fit the threshold
if dev is not None:
dt_dev = read_dataset(dev, labels=label_map.keys())
X_dev = [x[1] for x in dt_dev]
Y_dev = [label_map[x[0]] for x in dt_dev]
print "[fitting]"
model.fit(X_dev,Y_dev,samples=SAMPLEZ,silent=True)
conf = model.get_params()
#test model
Y_hat = model.predict(X_test)
avgF1 = f1_score(Y_test, Y_hat,average="macro")
acc = accuracy_score(Y_test, Y_hat)
results = {"acc":round(acc,3),
"avgF1":round(avgF1,3),
"model":run_id,
"dataset":os.path.basename(test),
"run_id":run_id
}
results.update(conf)
cols = ["dataset", "model", "run_id", "acc", "avgF1"]
helpers.print_results(results, columns=cols)
if res_path is not None:
#cols+=["positive_threshold","keep_scores_below","keep_scores_above"]
helpers.save_results(results, res_path, sep="\t", columns=cols)
return results
def hypertuner(train,
dev,
test,
emb_path,
obj,
hyperparams,
run_id,
res_path=None,
model_path=None):
with open(train, 'rb') as fid:
X_train, Y_train, vocabulary = cPickle.load(fid)
with open(dev, 'rb') as fid:
X_dev, Y_dev, _ = cPickle.load(fid)
with open(test, 'rb') as fid:
X_test, Y_test, _ = cPickle.load(fid)
E, _ = embeddings.read_embeddings(emb_path, wrd2idx=vocabulary)
Ys = Y_train + Y_dev + Y_test
label_map = vectorizer.build_vocabulary(Ys)
Y_train = [label_map[y] for y in Y_train]
Y_test = [label_map[y] for y in Y_test]
Y_dev = [label_map[y] for y in Y_dev]
dataset = os.path.basename(test)
best_hp = None
best_score = 0
best_results = None
for hp in hyperparams:
#initialize model with the hyperparameters
nn = nlse.NLSE(E, label_map=label_map, vocab=vocabulary, **hp)
# nn = nlse.NLSE(E, **hp)
nn.fit(X_train, Y_train, X_dev, Y_dev, silent=False)
Y_hat = nn.predict(X_test)
score = obj(Y_test, Y_hat)
print "[score: {} | hyperparameters: {}]".format(score, repr(hp))
if score > best_score:
if model_path is not None:
nn.save(model_path)
best_score = score
best_hp = hp
acc = accuracy_score(Y_test, Y_hat)
avgF1 = f1_score(Y_test, Y_hat, average="macro")
rep_hp = {p: hp[p] for p in ["sub_size", "lrate"]}
best_results = {"acc":round(acc,3), \
"avgF1":round(avgF1,3), \
"model":"NLSE", \
"dataset":dataset, \
"run_id":run_id,
"hyper":repr(rep_hp)}
res = {"score": round(score, 3), "hyper": repr(hp)}
helpers.print_results(res)
if res_path is not None:
cols = ["dataset", "model", "run_id", "acc", "avgF1"]
helpers.save_results(best_results, res_path, cols, sep="\t")
print ""
print "[best conf: {} | score: {}]".format(repr(best_hp), best_score)
return best_hp, best_score
def main(train,
dev,
test,
emb_path,
hyperparams,
run_id=None,
res_path=None,
no_hidden=False):
with open(train, 'rb') as fid:
X_train, Y_train, vocabulary = cPickle.load(fid)
with open(dev, 'rb') as fid:
X_dev, Y_dev, _ = cPickle.load(fid)
with open(test, 'rb') as fid:
X_test, Y_test, _ = cPickle.load(fid)
E, _ = embeddings.read_embeddings(emb_path, wrd2idx=vocabulary)
Ys = Y_train + Y_dev + Y_test
label_map = vectorizer.build_vocabulary(Ys)
Y_train = [label_map[y] for y in Y_train]
Y_test = [label_map[y] for y in Y_test]
Y_dev = [label_map[y] for y in Y_dev]
print "[no hidden: {}]".format(no_hidden)
# set_trace()
if no_hidden:
del hyperparams["sub_size"]
nn = nlse.BOE_plus(E,
label_map=label_map,
vocab=vocabulary,
**hyperparams)
else:
nn = nlse.NLSE(E, label_map=label_map, vocab=vocabulary, **hyperparams)
nn.fit(X_train, Y_train, X_dev, Y_dev)
y_hat = nn.predict(X_test)
avgF1 = f1_score(Y_test, y_hat, average="macro")
acc = accuracy_score(Y_test, y_hat)
run_id = run_id
dataset = os.path.basename(test)
# hp = {p:hyperparams[p] for p in ["sub_size","lrate"]}
if run_id is None: run_id = "NLSE"
results = {"acc":round(acc,3), \
"avgF1":round(avgF1,3), \
"model":"NLSE", \
"dataset":dataset, \
"run_id":run_id}
helpers.print_results(results,
columns=["dataset", "run_id", "acc", "avgF1"])
if res_path is not None:
cols = ["dataset", "model", "run_id", "acc", "avgF1"]
helpers.save_results(results, res_path, sep="\t", columns=cols)
return results, nn
def main(train, test, run_id, features, hyperparameters={}, res_path=None):
#train and evalute model
if features[0].lower() == "naive_bayes":
X_train, Y_train = get_features(train, ["bow-bin"])
X_test, Y_test = get_features(test, ["bow-bin"])
model = BernoulliNB()
model_name = "NaiveBayes"
elif features[0].lower() == "mlp":
X_train, Y_train = get_features(train, ["bow-bin"])
X_test, Y_test = get_features(test, ["bow-bin"])
model = MLPClassifier(solver='lbfgs',
activation="logistic",
hidden_layer_sizes=[400])
model_name = "MLP"
elif features[0].lower() == "mlp-2":
X_train, Y_train = get_features(train, ["bow-bin"])
X_test, Y_test = get_features(test, ["bow-bin"])
model = MLPClassifier(solver='lbfgs',
activation="logistic",
hidden_layer_sizes=[400, 100])
model_name = "MLP-2"
else:
X_train, Y_train = get_features(train, features)
X_test, Y_test = get_features(test, features)
#initialize model with the hyperparameters
model = SGDClassifier(random_state=1234, **hyperparameters)
model_name = "+".join(features)
model.fit(X_train, Y_train)
Y_hat = model.predict(X_test)
avgF1 = f1_score(Y_test, Y_hat, average="macro")
acc = accuracy_score(Y_test, Y_hat)
results = {"acc":round(acc,3), \
"avgF1":round(avgF1,3), \
"model":model_name, \
"dataset":os.path.basename(test), \
"run_id":run_id, \
"train_size":len(X_train), \
"test_size":len(X_test), \
"hyper":repr(hyperparameters)}
cols = ["dataset", "run_id", "acc", "avgF1", "hyper"]
helpers.print_results(results, columns=cols)
if res_path is not None:
cols = ["dataset", "model", "run_id", "acc", "avgF1"]
helpers.save_results(results, res_path, sep="\t", columns=cols)
return results
def hypertune(train, dev, features, obj, hyperparams, res_path=None):
X_train, Y_train = get_features(train, features)
X_dev, Y_dev = get_features(dev, features)
best_hp = None
best_score = 0
for hp in hyperparams:
#initialize model with the hyperparameters
model = SGDClassifier(random_state=1234, **hp)
model.fit(X_train, Y_train)
Y_hat = model.predict(X_dev)
score = obj(Y_dev, Y_hat)
# print "[score: {} | hyperparameters: {}]".format(score, repr(hp))
if score > best_score:
best_score = score
best_hp = hp
results = {"score": round(score, 3), "hyper": repr(hp)}
if res_path is not None:
helpers.save_results(results, res_path)
helpers.print_results(results)
print ""
print "[best conf: {} | score: {}]".format(repr(best_hp),
round(best_score, 3))
return best_hp, best_score
def main(train,
test,
dev,
embs_path,
total_epochs=10,
weights_file=None,
results_path=None):
print "[reading data]"
train_data = data.read_dataset(train)
train_docs = [x[1] for x in train_data]
train_Y = [x[0] for x in train_data]
test_data = data.read_dataset(test)
test_docs = [x[1] for x in test_data]
test_Y = [x[0] for x in test_data]
dev_data = data.read_dataset(dev)
dev_docs = [x[1] for x in dev_data]
dev_Y = [x[0] for x in dev_data]
#convert labels to one-hot
label_map = vectorizer.build_vocabulary(test_Y + train_Y + dev_Y)
train_Y = vectorizer.one_hot(label_map, train_Y)
dev_Y = vectorizer.one_hot(label_map, dev_Y)
test_Y = vectorizer.one_hot(label_map, test_Y)
#convert to argmax
test_Y = np.argmax(test_Y, axis=1)
n_labels = len(train_Y[0])
print "[loading embeddings]"
wvs = embeddings.embeddings_to_dict(embs_path)
# preprocessor for texts
print "[preprocessing...]"
all_docs = train_docs + test_docs + dev_docs
max_len = max([len(x.split()) for x in all_docs])
print "[max len: {}]".format(max_len)
p = CNN_text.Preprocessor(max_features=len(wvs), maxlen=max_len, wvs=wvs)
p.preprocess(all_docs)
train_X = p.build_sequences(train_docs)
test_X = p.build_sequences(test_docs)
dev_X = p.build_sequences(dev_docs)
# then the CNN
cnn = CNN_text.TextCNN(p,
n_labels=n_labels,
filters=[2, 3],
n_filters=50,
dropout=0.0)
if weights_file:
cnn.model.load_weights('weights.hdf5')
epochs_per_iter = 1
epochs_so_far = 0
print "training"
while epochs_so_far < total_epochs:
cnn.train(train_X,
train_Y,
nb_epoch=epochs_per_iter,
X_val=dev_X,
y_val=dev_Y)
epochs_so_far += epochs_per_iter
Y_hat = cnn.predict(dev_X)
acc = accuracy_score(np.argmax(dev_Y, axis=1), Y_hat)
avgF1 = f1_score(np.argmax(dev_Y, axis=1), Y_hat, average="macro")
res={"acc":round(acc,3), \
"avgF1":round(avgF1,3)}
helpers.print_results(res)
#print("acc @ epoch %s: %s" % (epochs_so_far, acc))
Y_hat = cnn.predict(test_X)
acc = accuracy_score(test_Y, Y_hat)
avgF1 = f1_score(test_Y, Y_hat, average="macro")
results = {"acc":round(acc,3), \
"avgF1":round(avgF1,3), \
"model":"CNN", \
"dataset":os.path.basename(test), \
"run_id":"NEURAL"}
helpers.print_results(results)
if results_path is not None:
cols = ["dataset", "model", "run_id", "acc", "avgF1"]
helpers.save_results(results, results_path, cols, sep="\t")
conf, _ = hypertune(tr_fname, ts_fname, args.emb_path,\
scorer, hyperparams_grid, res_path=hyper_results_path)
#run model with the best hyperparams
res = main(tr_fname,
dev_fname,
ts_fname,
args.emb_path,
conf,
run_id=args.run_id,
res_path=cv_results_path)
results.append(res)
accs = [res["acc"] for res in results]
f1s = [res["avgF1"] for res in results]
cv_res = {"acc_mean":round(np.mean(accs),3), \
"acc_std":round(np.std(accs),3), \
"avgF1_mean":round(np.mean(f1s),3), \
"avgF1_std":round(np.std(f1s),3), \
"model":"NLSE", \
"dataset":os.path.basename(args.test), \
"run_id":args.run_id}
helpers.print_results(cv_res)
#save the results of each run
if args.res_path is not None:
cols = [
"dataset", "run_id", "model", "acc_mean", "acc_std",
"avgF1_mean", "avgF1_std"
]
helpers.save_results(cv_res, args.res_path, sep="\t", columns=cols)
predictor.fit(X_train, Y_train)
Y_hat = predictor.predict(X_test)
#evaluate
avgF1 = f1_score(Y_test, Y_hat, average="macro")
acc = accuracy_score(Y_test, Y_hat)
fname = os.path.basename(args.test)
run_id = args.run_id
if run_id is None: run_id = "+".join(args.features)
results = {"dataset": fname,
"acc": round(acc, 3),
"avgF1": round(avgF1, 3),
"features": "+".join(args.features)+"@"+args.model,
"run_id": run_id}
cols = ["dataset", "run_id", "features", "acc", "avgF1"]
#report
helpers.print_results(results, columns=cols)
if args.res_path is not None:
helpers.save_results(results, args.res_path, columns=cols)
elif args.type == "continuous":
#choose model
if args.model == "linear":
predictor = SVR(kernel='linear')
elif args.model == "l1":
predictor = LinearSVR(loss='epsilon_insensitive')
elif args.model == "rbf":
predictor = SVR(kernel='rbf')
#train and predict
predictor.fit(X_train, Y_train)
Y_hat = predictor.predict(X_test)
#evaluate
pred_rank = sp.stats.stats.rankdata(Y_hat)
# logging.info('Initialized model from embeddings %s' % args.emb)
# logging.info("Training data: %s" % args.tr)
# logging.info("Dev data: %s" % args.dev)
with open(args.tr, 'rb') as fid:
train_x, train_y, vocabulary, label_map, E, st, ed = cPickle.load(fid)
E = E.astype(theano.config.floatX)
with open(args.dev, 'rb') as fid:
dev_x, dev_y, _, _ = cPickle.load(fid)
with open(args.ts, 'rb') as fid:
ts_x, ts_y, _, _ = cPickle.load(fid)
nn = train_nlse_reg(train_x, train_y, vocabulary, E, st, ed, args)
tau = evaluate(nn, ts_x, ts_y)
fname = os.path.basename(args.ts)
run_id = args.run_id
if run_id is None: run_id = "NLSE"
results = {
"tau": round(tau, 3),
"features": "NLSE",
"subsize": args.sub_size,
"lrate": args.lrate,
"dataset": fname,
"run_id": run_id
}
cols = ["dataset", "run_id", "features", "tau"]
helpers.print_results(
results, columns=["dataset", "run_id", "lrate", "subsize", "tau"])
if args.res_path is not None:
helpers.save_results(results, args.res_path, columns=cols)
