def test(self): trace('loading model ...') src_vocab = Vocabulary.load(self.model + '.srcvocab') trg_vocab = Vocabulary.load(self.model + '.trgvocab') encdec = EncoderDecoder.load_spec(self.model + '.spec') serializers.load_hdf5(self.model + '.weights', encdec) trace('generating translation ...') generated = 0 with open(self.target, 'w') as fp: for src_batch in gens.batch(gens.word_list(self.source), self.minibatch): src_batch = fill_batch(src_batch) K = len(src_batch) trace('sample %8d - %8d ...' % (generated + 1, generated + K)) hyp_batch = self.forward(src_batch, None, src_vocab, trg_vocab, encdec, False, self.generation_limit) source_cuont = 0 for hyp in hyp_batch: hyp.append('</s>') hyp = hyp[:hyp.index('</s>')] print("src : " + "".join(src_batch[source_cuont]).replace("</s>", "")) print('hyp : ' +''.join(hyp)) print(' '.join(hyp), file=fp) source_cuont = source_cuont + 1 generated += K trace('finished.')
def test(self): trace('loading model ...') src_vocab = Vocabulary.load(self.model + '.srcvocab') trg_vocab = Vocabulary.load(self.model + '.trgvocab') encdec = EncoderDecoder.load_spec(self.model + '.spec') serializers.load_hdf5(self.model + '.weights', encdec) trace('generating translation ...') generated = 0 with open(self.target, 'w') as fp: for src_batch in gens.batch(gens.word_list(self.source), self.minibatch): src_batch = fill_batch(src_batch) K = len(src_batch) trace('sample %8d - %8d ...' % (generated + 1, generated + K)) hyp_batch = self.forward(src_batch, None, src_vocab, trg_vocab, encdec, False, self.generation_limit) source_cuont = 0 for hyp in hyp_batch: hyp.append('</s>') hyp = hyp[:hyp.index('</s>')] print("src : " + "".join(src_batch[source_cuont]).replace("</s>", "")) print('hyp : ' + ''.join(hyp)) print(' '.join(hyp), file=fp) source_cuont = source_cuont + 1 generated += K trace('finished.')
def test(args): trace('loading model ...') word_vocab = Vocabulary.load(args.model + '.words') phrase_vocab = Vocabulary.load(args.model + '.phrases') semiterminal_vocab = Vocabulary.load(args.model + '.semiterminals') parser = Parser.load_spec(args.model + '.spec') if args.use_gpu: parser.to_gpu() serializers.load_hdf5(args.model + '.weights', parser) embed_cache = {} parser.reset() trace('generating parse trees ...') with open(args.source) as fp: for l in fp: word_list = to_vram_words(convert_word_list(l.split(), word_vocab)) tree = combine_xbar( restore_labels( parser.forward(word_list, None, args.unary_limit, embed_cache), phrase_vocab, semiterminal_vocab)) print('( ' + tree_to_string(tree) + ' )') trace('finished.')
def load(filename): self = AttentionalTranslationModel() with ModelFile(filename) as fp: self.__src_vocab = Vocabulary.load(fp.get_file_pointer()) self.__trg_vocab = Vocabulary.load(fp.get_file_pointer()) self.__n_embed = int(fp.read()) self.__n_hidden = int(fp.read()) self.__make_model() wrapper.begin_model_access(self.__model) fp.read_embed(self.__model.w_xi) fp.read_linear(self.__model.w_ia) fp.read_linear(self.__model.w_aa) fp.read_linear(self.__model.w_ib) fp.read_linear(self.__model.w_bb) fp.read_linear(self.__model.w_aw) fp.read_linear(self.__model.w_bw) fp.read_linear(self.__model.w_pw) fp.read_linear(self.__model.w_we) fp.read_linear(self.__model.w_ap) fp.read_linear(self.__model.w_bp) fp.read_embed(self.__model.w_yp) fp.read_linear(self.__model.w_pp) fp.read_linear(self.__model.w_cp) fp.read_linear(self.__model.w_dp) fp.read_linear(self.__model.w_py) wrapper.end_model_access(self.__model) return self
def test(args): trace('loading model ...') src_vocab = Vocabulary.load(args.model + '.srcvocab') trg_vocab = Vocabulary.load(args.model + '.trgvocab') attmt = AttentionMT.load_spec(args.model + '.spec') if args.use_gpu: attmt.to_gpu() serializers.load_hdf5(args.model + '.weights', attmt) trace('generating translation ...') generated = 0 with open(args.target, 'w') as fp: for src_batch in gens.batch(gens.word_list(args.source), args.minibatch): src_batch = fill_batch(src_batch) K = len(src_batch) trace('sample %8d - %8d ...' % (generated + 1, generated + K)) hyp_batch = forward(src_batch, None, src_vocab, trg_vocab, attmt, False, args.generation_limit) for hyp in hyp_batch: hyp.append('</s>') hyp = hyp[:hyp.index('</s>')] print(' '.join(hyp), file=fp) generated += K trace('finished.')
def __predict_sentence(self, src_batch): dialogue = EncoderDecoderModelForwardSlack(self.parameter) src_vocab = Vocabulary.load(self.model_name + '.srcvocab') trg_vocab = Vocabulary.load(self.model_name + '.trgvocab') model = EncoderDecoder.load_spec(self.model_name + '.spec') serializers.load_hdf5(dialogue.model + '.weights', model) hyp_batch = dialogue.forward(src_batch, None, src_vocab, trg_vocab, model, False, self.generation_limit) return hyp_batch
def train(args): if os.path.exists("./model/vocab.bin"): src_vocab = Vocabulary.load("./model/vocab.bin") else: src_vocab = Vocabulary.new(gens.word_list(args.source), args.n_vocab) src_vocab.save('./model/vocab.bin') if os.path.exists("./model/tag.bin"): trg_tag = Vocabulary.load("./model/tag.bin") else: trg_tag = Vocabulary.new(gens.word_list(args.target), args.n_tag) trg_tag.save('./model/tag.bin') print("vocab_len:{}".format(src_vocab.__len__)) print("tag_len:{}".format(trg_tag.__len__)) encdec = BiEncDecLSTM(args.n_vocab, args.layer, args.embed, args.hidden, args.n_tag) optimizer = optimizers.Adam() optimizer.setup(encdec) for e_i in range(args.epoch): tt_list = [[src_vocab.stoi(char) for char in char_arr] for char_arr in gens.word_list(args.source_tr)] tag_list = [ trg_tag.stoi(tag[0]) for tag in gens.word_list(args.target_tr) ] print("{}:{}".format(len(tt_list), len(tag_list))) assert len(tt_list) == len(tag_list) ind_arr = [ri for ri in range(len(tt_list))] random.shuffle(ind_arr) tt_now = (tt_list[ri] for ri in ind_arr) tag_now = (tag_list[ri] for ri in ind_arr) tt_gen = gens.batch(tt_now, args.batchsize) tag_gen = gens.batch(tag_now, args.batchsize) for tt, tag in zip(tt_gen, tag_gen): y_ws = encdec(tt) teac_arr = [src_vocab.itos(t) for t in tt[0]] pred_arr = [trg_tag.itos(y_each.data.argmax(0)) for y_each in y_ws] print("teach:{}:{}:{}".format(teac_arr, trg_tag.itos(tag[0]), pred_arr[0])) tag = xp.array(tag, dtype=xp.int32) loss = F.softmax_cross_entropy(y_ws, tag) encdec.cleargrads() loss.backward() optimizer.update() # loss.backward() # optimizer.target.cleargrads() # loss.backward() # loss.unchain_backward() # optimizer.update() serializers.save_npz('./model/attn_tag_model_{}.npz'.format(e_i), encdec)
def __init__(self, args): trace('loading model ...') self.args = args self.src_vocab = Vocabulary.load(args.model + '.srcvocab') self.trg_vocab = Vocabulary.load(args.model + '.trgvocab') self.encdec = EncoderDecoder.load_spec(args.model + '.spec') if args.use_gpu: self.encdec.to_gpu() serializers.load_hdf5(args.model + '.weights', self.encdec) trace('generating translation ...')
def __predict_sentence(self, src_batch): """ predict sentence :param src_batch: get the source sentence :return: """ dialogue = EncoderDecoderModelAttention(self.parameter) src_vocab = Vocabulary.load(self.model_name + '.srcvocab') trg_vocab = Vocabulary.load(self.model_name + '.trgvocab') model = AttentionDialogue.load_spec(self.model_name + '.spec', self.XP) serializers.load_hdf5(self.model_name + '.weights', model) hyp_batch = dialogue.forward_implement(src_batch, None, src_vocab, trg_vocab, model, False, self.generation_limit) return hyp_batch
def load(self, filename): with ModelFile(filename) as fp: self.src_vocab = Vocabulary.load(fp.get_file_pointer()) self.trg_vocab = Vocabulary.load(fp.get_file_pointer()) self.n_embed = int(fp.read()) self.n_hidden = int(fp.read()) self.make_model() wrapper.begin_model_access(self.model) fp.read_embed(self.model.weight_xi) fp.read_linear(self.model.weight_ip) fp.read_linear(self.model.weight_pp) fp.read_linear(self.model.weight_pq) fp.read_linear(self.model.weight_qj) fp.read_linear(self.model.weight_jy) fp.read_embed(self.model.weight_yq) fp.read_linear(self.model.weight_qq) wrapper.end_model_access(self.model) return self
def test(args, epoch): model_name = "./model/attn_tag_model_{}.npz".format(epoch) encdec = BiEncDecLSTM(args.n_vocab, args.layer, args.embed, args.hidden, args.n_tag) serializers.load_npz(model_name, encdec) src_vocab = Vocabulary.load("./model/vocab.bin") trg_tag = Vocabulary.load("./model/tag.bin") tt_now = ([src_vocab.stoi(char) for char in char_arr] for char_arr in gens.word_list(args.source_te)) tag_now = (trg_tag.stoi(tag[0]) for tag in gens.word_list(args.target_te)) tt_gen = gens.batch(tt_now, args.batchsize) tag_gen = gens.batch(tag_now, args.batchsize) correct_num = 0 wrong_num = 0 fw = codecs.open("./output/result_attn_tw{}.csv".format(epoch), "w", encoding="utf-8") fw.write("台詞,教師キャラ,予測キャラ,予測値,›単語\n") for tt, tag in zip(tt_gen, tag_gen): y, att_w = encdec.callAndAtt(tt) max_y = [ max( F.softmax(F.reshape(y_each.data, (1, len(y_each.data)))).data[0]) for y_each in y ] y = [y_each.data.argmax(0) for y_each in y] for tt_e, y_e, tag_e, max_y_e, att_w_e in zip(tt, y, tag, max_y, att_w): txt = ",".join([src_vocab.itos(id) for id in tt_e]) tag_e = trg_tag.itos(tag_e) y_e = trg_tag.itos(y_e) att_ind = att_w_e.data.argmax() most_word = src_vocab.itos(tt_e[att_ind]) fw.write("{}:{}:{}:{}:{}\n".format(txt, tag_e, y_e, max_y_e, most_word)) correct_num += len([1 for y_e, tag_e in zip(y, tag) if y_e == tag_e]) wrong_num += len([1 for y_e, tag_e in zip(y, tag) if y_e != tag_e]) print("epoch:{}".format(epoch)) print(" correct:{}".format(correct_num)) print(" wrong:{}".format(wrong_num)) fw.write("correct{}\n".format(correct_num)) fw.write("wrong:{}\n".format(wrong_num)) fw.close()
def load(filename): self = EncoderDecoderModel() with ModelFile(filename) as fp: self.__src_vocab = Vocabulary.load(fp.get_file_pointer()) self.__trg_vocab = Vocabulary.load(fp.get_file_pointer()) self.__n_embed = int(fp.read()) self.__n_hidden = int(fp.read()) self.__make_model() wrapper.begin_model_access(self.__model) fp.read_embed(self.__model.w_xi) fp.read_linear(self.__model.w_ip) fp.read_linear(self.__model.w_pp) fp.read_linear(self.__model.w_pq) fp.read_linear(self.__model.w_qj) fp.read_linear(self.__model.w_jy) fp.read_embed(self.__model.w_yq) fp.read_linear(self.__model.w_qq) wrapper.end_model_access(self.__model) return self
def test(self): """ Test method You have to parepare the train model """ trace('loading model ...') prefix = self.model model_path = APP_ROOT + "/model/" + prefix src_vocab = Vocabulary.load(model_path + '.srcvocab') trg_vocab = Vocabulary.load(model_path + '.trgvocab') self.attention_dialogue = AttentionDialogue.load_spec( model_path + '.spec', self.XP) serializers.load_hdf5(model_path + '.weights', self.attention_dialogue) trace('generating translation ...') generated = 0 with open(self.test_target, 'w') as fp: for src_batch in gens.batch(gens.word_list(self.source), self.minibatch): src_batch = fill_batch(src_batch) K = len(src_batch) trace('sample %8d - %8d ...' % (generated + 1, generated + K)) hyp_batch = self.forward_implement(src_batch, None, src_vocab, trg_vocab, self.attention_dialogue, False, self.generation_limit) source_cuont = 0 for hyp in hyp_batch: hyp.append('</s>') hyp = hyp[:hyp.index('</s>')] print("src : " + "".join(src_batch[source_cuont]).replace("</s>", "")) print('hyp : ' + ''.join(hyp)) print(' '.join(hyp), file=fp) source_cuont = source_cuont + 1 generated += K trace('finished.')
def test(args): trace('loading model ...') word_vocab = Vocabulary.load(args.model + '.words') phrase_vocab = Vocabulary.load(args.model + '.phrases') semi_vocab = Vocabulary.load(args.model + '.semiterminals') parser = Parser.load_spec(args.model + '.spec') if USE_GPU: parser.to_gpu() serializers.load_hdf5(args.model + '.weights', parser) trace('generating parse trees ...') with open(args.source) as fp: for l in fp: word_list = convert_word_list(l.split(), word_vocab) tree = restore_labels( parser.forward(word_list, None, args.unary_limit), phrase_vocab, semi_vocab) print('( ' + tree_to_string(tree) + ' )') trace('finished.')
def test(args): trace('loading model ...') src_vocab = Vocabulary.load(args.model + '.srcvocab') trg_vocab = Vocabulary.load(args.model + '.trgvocab') encdec = EncoderDecoder.load_spec(args.model + '.spec') if args.use_gpu: encdec.to_gpu() serializers.load_hdf5(args.model + '.weights', encdec) trace('generating translation ...') generated = 0 temp = gens.to_words(args.target) # temp.append("</s>") src_batch = [] src_batch.append(temp) # src_batch = [['私は', '太郎', 'です', '(´', 'ー', '`*)', 'ウンウン', '</s>']] src_batch = fill_batch(src_batch) print("src_batch:", src_batch) K = len(src_batch) trace('sample %8d - %8d ...' % (generated + 1, generated + K)) print("question:") for srp in src_batch: srp.append('</s>') srp = srp[:srp.index('</s>')] print(''.join(srp)) hyp_batch = forward(src_batch, None, src_vocab, trg_vocab, encdec, False, args.generation_limit) print("answser:") for hyp in hyp_batch: hyp.append('</s>') hyp = hyp[:hyp.index('</s>')] print(''.join(hyp)) print("----------------") generated += K trace('finished.')
def load(filename): self = TransSegmentationModel() with ModelFile(filename) as fp: self.__vocab = Vocabulary.load(fp.get_file_pointer()) self.__n_context = int(fp.read()) self.__n_hidden = int(fp.read()) self.__make_model() wrapper.begin_model_access(self.__model) fp.read_embed(self.__model.w_xh) fp.read_linear(self.__model.w_hy) wrapper.end_model_access(self.__model) return self
def load(filename): self = SegmentationModel() with ModelFile(filename) as fp: self.__vocab = Vocabulary.load(fp.get_file_pointer()) self.__n_context = int(fp.read()) self.__n_hidden = int(fp.read()) self.__make_model() wrapper.begin_model_access(self.__model) fp.read_embed(self.__model.w_xh) fp.read_linear(self.__model.w_hy) wrapper.end_model_access(self.__model) return self
def setCateg(self, args): categ_name = "./{}/categ_{}.bin".format(args.dataname, args.dataname) if os.path.exists(categ_name): categ_vocab = Vocabulary.load(categ_name) else: set_cat = set() [[set_cat.add(word) for word in word_arr] for word_arr in gens.word_list(args.category)] n_categ = len(set_cat) + 3 print("n_categ:{}".format(n_categ)) categ_vocab = Vocabulary.new(gens.word_list(args.category), n_categ) categ_vocab.save(categ_name) self.categ_vocab = categ_vocab return categ_vocab
def test(args): trace('loading model ...') word_vocab = Vocabulary.load(args.model + '.words') phrase_vocab = Vocabulary.load(args.model + '.phrases') semi_vocab = Vocabulary.load(args.model + '.semiterminals') parser = Parser.load_spec(args.model + '.spec') if USE_GPU: parser.to_gpu() serializers.load_hdf5(args.model + '.weights', parser) trace('generating parse trees ...') with open(args.source) as fp: for l in fp: word_list = convert_word_list(l.split(), word_vocab) tree = restore_labels( parser.forward(word_list, None, args.unary_limit), phrase_vocab, semi_vocab ) print('( ' + tree_to_string(tree) + ' )') trace('finished.')
def test(self): """ Test method You have to parepare the train model """ trace("loading model ...") prefix = self.model model_path = APP_ROOT + "/model/" + prefix src_vocab = Vocabulary.load(model_path + ".srcvocab") trg_vocab = Vocabulary.load(model_path + ".trgvocab") self.attention_dialogue = AttentionDialogue.load_spec(model_path + ".spec", self.XP) serializers.load_hdf5(model_path + ".weights", self.attention_dialogue) trace("generating translation ...") generated = 0 with open(self.test_target, "w") as fp: for src_batch in gens.batch(gens.word_list(self.source), self.minibatch): src_batch = fill_batch(src_batch) K = len(src_batch) trace("sample %8d - %8d ..." % (generated + 1, generated + K)) hyp_batch = self.forward_implement( src_batch, None, src_vocab, trg_vocab, self.attention_dialogue, False, self.generation_limit ) source_cuont = 0 for hyp in hyp_batch: hyp.append("</s>") hyp = hyp[: hyp.index("</s>")] print("src : " + "".join(src_batch[source_cuont]).replace("</s>", "")) print("hyp : " + "".join(hyp)) print(" ".join(hyp), file=fp) source_cuont = source_cuont + 1 generated += K trace("finished.")
def setVocab(self, args): vocab_name = "./{}/vocab_{}.bin".format(args.dataname, args.dataname) if os.path.exists(vocab_name): src_vocab = Vocabulary.load(vocab_name) else: set_vocab = set() [[set_vocab.add(word) for word in word_arr] for word_arr in gens.word_list(args.source)] n_vocab = len(set_vocab) + 3 print("n_vocab:{}".format(n_vocab)) print("arg_vocab:{}".format(args.n_vocab)) src_vocab = Vocabulary.new(gens.word_list(args.source), args.n_vocab) src_vocab.save(vocab_name) self.vocab = src_vocab return src_vocab
def load(filename): self = RNNSegmentationModel() with ModelFile(filename) as fp: self.__vocab = Vocabulary.load(fp.get_file_pointer()) self.__n_embed = int(fp.read()) self.__n_hidden = int(fp.read()) self.__make_model() wrapper.begin_model_access(self.__model) fp.read_embed(self.__model.w_xe) fp.read_linear(self.__model.w_ea) fp.read_linear(self.__model.w_aa) fp.read_linear(self.__model.w_eb) fp.read_linear(self.__model.w_bb) fp.read_linear(self.__model.w_ay1) fp.read_linear(self.__model.w_by1) fp.read_linear(self.__model.w_ay2) fp.read_linear(self.__model.w_by2) wrapper.end_model_access(self.__model) return self
def train(args): trace('making vocabularies ...') src_vocab = Vocabulary.new(gens.input_word_list(), args.vocab) trg_vocab = Vocabulary.new(gens.output_word_list(), args.vocab) trace('making model ...') encdec = EncoderDecoder(args.vocab, args.embed, args.hidden) if args.load_model != "": print("model load %s ... " % (args.load_model)) src_vocab = Vocabulary.load(args.load_model + '.srcvocab') trg_vocab = Vocabulary.load(args.load_model + '.trgvocab') encdec = EncoderDecoder.load_spec(args.load_model + '.spec') serializers.load_hdf5(args.load_model + '.weights', encdec) if args.use_gpu: encdec.to_gpu() for epoch in range(args.epoch): trace('epoch %d/%d: ' % (epoch + 1, args.epoch)) trained = 0 gen1 = gens.input_word_list() gen2 = gens.output_word_list() gen3 = gens.batch( gens.sorted_parallel(gen1, gen2, 100 * args.minibatch), args.minibatch) opt = optimizers.AdaGrad(lr=0.01) opt.setup(encdec) opt.add_hook(optimizer.GradientClipping(5)) for src_batch, trg_batch in gen3: src_batch = fill_batch(src_batch) trg_batch = fill_batch(trg_batch) K = len(src_batch) hyp_batch, loss = forward(src_batch, trg_batch, src_vocab, trg_vocab, encdec, True, 0) loss.backward() opt.update() for k in range(K): trace('epoch %3d/%3d, sample %8d' % (epoch + 1, args.epoch, trained + k + 1)) trace( ' src = ' + ' '.join([x if x != '</s>' else '*' for x in src_batch[k]])) trace( ' trg = ' + ' '.join([x if x != '</s>' else '*' for x in trg_batch[k]])) trace( ' hyp = ' + ' '.join([x if x != '</s>' else '*' for x in hyp_batch[k]])) trained += K if epoch % args.model_save_timing == 0: trace('saving model ...') prefix = args.model + '.%03.d' % (epoch + 1) src_vocab.save(prefix + '.srcvocab') trg_vocab.save(prefix + '.trgvocab') encdec.save_spec(prefix + '.spec') serializers.save_hdf5(prefix + '.weights', encdec) trace('finished.')
from util.vocabulary import Vocabulary from util import generators as gens from util.controller import Controller from util.wrapper import wrapper from util.const import * if __name__ == '__main__': args = parse_args() trace('initializing ...') wrapper = wrapper(args.gpu_id) wrapper.init() trace('loading vocab ...') # src_vocab = Vocabulary.load(args.src_vocab) # trg_vocab = Vocabulary.load(args.trg_vocab) src_vocab = Vocabulary.load(VOCAB_SRC) trg_vocab = Vocabulary.load(VOCAB_TRG) controller = Controller(args.folder_name) if args.mode == 'train': controller.train_model(BasicEncoderDecoderModel, src_vocab, trg_vocab, args) elif args.mode == 'dev': controller.dev_model(BasicEncoderDecoderModel, src_vocab, trg_vocab, args) elif args.mode == 'test': controller.test_model(BasicEncoderDecoderModel, src_vocab, trg_vocab, args)