def __init__(self,args):
self.args = args
logging.basicConfig(stream=sys.stdout,
format='%(asctime)s %(levelname)s:%(message)s',
level=logging.INFO, datefmt='%I:%M:%S')
#with open(os.path.join(self.args.model_dir, 'result.json'), 'r') as f:
# 获取 绝对路径
with open(r'G:\Postgraduate\myClassesOfGraduateStudentOne\张华平大数据人工智能课程\大作业\字符级LSTM文本自动生成模型\TangPoemGenerator\poet\output_poem\result.json', 'r') as f:
result = json.load(f)
params = result['params']
best_model = result['best_model']
best_valid_ppl = result['best_valid_ppl']
if 'encoding' in result:
self.args.encoding = result['encoding']
else:
self.args.encoding = 'utf-8'
base_path = args.data_dir
"""
源代码中都是使用os.path.join 形成的路径在windows下面，类似于
“FileNotFoundError: [Errno 2] No such file or directory: './data/poem\\rhyme_words.txt'”
这样的错误，
所以本次试验使用 绝对路径进行替换
"""
#w2v_file = os.path.join(base_path, "vectors_poem.bin")
w2v_file=r'G:\Postgraduate\myClassesOfGraduateStudentOne\张华平大数据人工智能课程\大作业\字符级LSTM文本自动生成模型\TangPoemGenerator\poet\data\poem\vectors_poem.bin'
#将词嵌入加载进来
self.w2v = Word2Vec(w2v_file)
#读取押韵词
RhymeWords.read_rhyme_words(r'G:\Postgraduate\myClassesOfGraduateStudentOne\张华平大数据人工智能课程\大作业\字符级LSTM文本自动生成模型\TangPoemGenerator\poet\data\poem\rhyme_words.txt')
if args.seed >= 0:
np.random.seed(args.seed)
logging.info('best_model: %s\n', best_model)
best_model=r'G:\Postgraduate\myClassesOfGraduateStudentOne\张华平大数据人工智能课程\大作业\字符级LSTM文本自动生成模型\TangPoemGenerator\poet\output_poem\best_model\model-16312'
self.sess = tf.Session()
w2v_vocab_size = len(self.w2v.model.vocab)
with tf.name_scope('evaluation'):
self.model = CharRNNLM(is_training=False,w2v_model = self.w2v.model,vocab_size=w2v_vocab_size, infer=True, **params)
saver = tf.train.Saver(name='model_saver')
saver.restore(self.sess, best_model)
def __init__(self, args):
self.args = args
logging.basicConfig(stream=sys.stdout,
format='%(asctime)s %(levelname)s:%(message)s',
level=logging.INFO,
datefmt='%I:%M:%S')
with open(os.path.join(self.args.model_dir, 'result.json'), 'r') as f:
result = json.load(f)
params = result['params']
best_model = result['best_model']
best_valid_ppl = result['best_valid_ppl']
if 'encoding' in result:
self.args.encoding = result['encoding']
else:
self.args.encoding = 'utf-8'
base_path = args.data_dir
w2v_file = os.path.join(base_path, "vectors_poem.bin")
self.w2v = Word2Vec(w2v_file)
RhymeWords.read_rhyme_words(os.path.join(base_path, 'rhyme_words.txt'))
if args.seed >= 0:
np.random.seed(args.seed)
logging.info('best_model: %s\n', best_model)
# self.sess = tf.Session()
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.2
self.sess = tf.Session(config=config)
# self.sess = tf.Session(config=tf.ConfigProto(device_count={'gpu':0}))
w2v_vocab_size = len(self.w2v.model.vocab)
with tf.name_scope('evaluation'):
self.model = CharRNNLM(is_training=False,
w2v_model=self.w2v.model,
vocab_size=w2v_vocab_size,
infer=True,
**params)
saver = tf.train.Saver(name='model_saver')
saver.restore(self.sess, best_model)
def select_rhyme(self, rhyme_ref, probs):
if rhyme_ref:
rhyme_set = RhymeWords.get_rhyme_words(rhyme_ref)
if rhyme_set: # 如果生成的第二句的韵脚在韵律字典里面，则取出全部的韵律字典
seq_probs = zip(probs, range(0, self.vocab_size))
topn = heapq.nlargest(50, seq_probs, key=lambda sp: sp[0])
# 一句概率找到前50个，如果前50个里有，就选取前50里面最大的，否则选取最大概率的那个，就不管韵律了
for _, seq in topn:
if self.w2v_model.vocab[seq] in rhyme_set:
return seq
return np.argmax(probs)
def select_rhyme(self, rhyme_ref, probs):
if rhyme_ref:
rhyme_set = RhymeWords.get_rhyme_words(rhyme_ref)
if rhyme_set:
seq_probs = zip(probs, range(0, self.vocab_size))
topn = heapq.nlargest(50, seq_probs, key=lambda sp: sp[0])
for _, seq in topn:
if self.w2v_model.vocab[seq] in rhyme_set:
return seq
return np.argmax(probs)
def select_rhyme(self,rhyme_ref,probs):
if rhyme_ref:
rhyme_set = RhymeWords.get_rhyme_words(rhyme_ref)
if rhyme_set:
seq_probs = zip(probs,range(0,self.vocab_size))
topn = heapq.nlargest(50,seq_probs,key=lambda sp :sp[0])
for _,seq in topn:
if self.w2v_model.vocab[seq] in rhyme_set:
return seq
return np.argmax(probs)