更新编码问题和生成对话

Author: zhaoyingjun

lessonTen/seqGan chatbotv2.0/execute.py

@@ -306,36 +306,44 @@ def decoder_online(sess,gen_config, model, vocab,rev_vocab, inputs):
     # Get output logits for the sentence.
     _, _, output_logits = model.step(sess, encoder_inputs, decoder_inputs, target_weights, bucket_id, True)
 
-    # This is a greedy decoder - outputs are just argmaxes of output_logits.
-    outputs = [int(np.argmax(logit, axis=1)) for logit in output_logits[0]]
-    
     # If there is an EOS symbol in outputs, cut them at that point.
-    if data_utils.EOS_ID in outputs:
-        outputs = outputs[:outputs.index(prepareData.EOS_ID)]
-
-    return " ".join([tf.compat.as_str(rev_vocab[output]) for output in outputs])
-
-     
-            
+    tokens = []
+    resps = []
+    for seq in output_logits:
+        token = []
+        for t in seq:
+            token.append(int(np.argmax(t, axis=0)))
+        tokens.append(token)
+        tokens_t = []
+        for col in range(len(tokens[0])):
+            tokens_t.append([tokens[row][col] for row in range(len(tokens))])
+
+        for seq in tokens_t:
+            if data_utils.EOS_ID in seq:
+                resps.append(seq[:seq.index(data_utils.EOS_ID)][:gen_config.buckets[bucket_id][1]])
+            else:
+                resps.append(seq[:gen_config.buckets[bucket_id][1]])
+    for resp in resps:
+        resq_str= " ".join([tf.compat.as_str(rev_vocab[output]) for output in resp])
+    return resq_str
 
-        
 
 def main(_):
     # step_1 training gen model
-    gen_pre_train()
+    #gen_pre_train()
 
-    print("*****请注释掉本行代码，以及上行代码gen_pre_train()，下行代码sys.exit(0)然后继续执行execute.py********")
-    sys.exit(0)
+    #print("*****请注释掉本行代码，以及上行代码gen_pre_train()，下行代码sys.exit(0)然后继续执行execute.py********")
+    #sys.exit(0)
     # step_2 gen training data for disc
-    gen_disc()
+    #gen_disc()
 
-    print("*****请注释掉本行代码，以及上行代码gen_disc()，下行代码sys.exit(0)然后继续执行execute.py********")
-    sys.exit(0)
+    #print("*****请注释掉本行代码，以及上行代码gen_disc()，下行代码sys.exit(0)然后继续执行execute.py********")
+    #sys.exit(0)
 
     # step_3 training disc model
-    disc_pre_train()
-    print("*****请注释掉本行代码，以及上行代码disc_pre_train()，下行代码sys.exit(0)然后继续执行execute.py********")
-    sys.exit(0)
+    #disc_pre_train()
+    #print("*****请注释掉本行代码，以及上行代码disc_pre_train()，下行代码sys.exit(0)然后继续执行execute.py********")
+    #sys.exit(0)
     # step_4 training al model
     al_train()
 

lessonTen/seqGan chatbotv2.0/gen/generator.py

@@ -110,7 +110,7 @@ def train(gen_config):
         gen_loss_summary = tf.Summary()
         gen_writer = tf.summary.FileWriter(gen_config.tensorboard_dir, sess.graph)
 
-        while current_step<100:
+        while current_step<1000:
             # Choose a bucket according to disc_data distribution. We pick a random number
             # in [0, 1] and use the corresponding interval in train_buckets_scale.
             random_number_01 = np.random.random_sample()
@@ -239,6 +239,7 @@ def decoder(gen_config):
 
             encoder_inputs, decoder_inputs, target_weights, batch_source_encoder, batch_source_decoder = \
                 model.get_batch(train_set, bucket_id, gen_config.batch_size)
+          
 
             _, _, out_logits = model.step(sess, encoder_inputs, decoder_inputs, target_weights, bucket_id,
                                           forward_only=True)
@@ -255,14 +256,7 @@ def decoder(gen_config):
 
             for seq in tokens_t:
                 if data_utils.EOS_ID in seq:
-                    '''
-                    seq[:seq.index(data_utils.EOS_ID)][:gen_config.buckets[bucket_id][1]]
-                    seq是一维的，乍一看以为上面表达式把他当作二维处理，但实际上s=[1,2,3,4]
-                    s[:3][:2]输出的是[1,2]，也就是先截取[0:3]的数据，在截取[0:2]的数据
-                    而没有冒号，即s[3][2]是当做二维处理，在这边这么写是错的
-                    '''
-
-                    #resps的shape为[[[vocab_size]],.......]  倒数第二层：decoder_size  最外一层：batch_size
+                    
                     resps.append(seq[:seq.index(data_utils.EOS_ID)][:gen_config.buckets[bucket_id][1]])
                 else:
                     resps.append(seq[:gen_config.buckets[bucket_id][1]])
@@ -272,7 +266,6 @@ def decoder(gen_config):
                 answer_str = " ".join([str(rev_vocab[an]) for an in answer[:-1]])
                 disc_train_answer.write(answer_str)
                 disc_train_answer.write("\n")
-
                 query_str = " ".join([str(rev_vocab[qu]) for qu in query])
                 disc_train_query.write(query_str)
                 disc_train_query.write("\n")

lessonTen/seqGan chatbotv2.0/utils/conf.py

@@ -5,7 +5,7 @@ class disc_config(object):
     batch_size = 64
     lr = 0.2
     lr_decay = 0.9
-    vocab_size = 25000
+    vocab_size = 2500
     embed_dim = 64
     steps_per_checkpoint = 20
     #hidden_neural_size = 128
@@ -36,7 +36,7 @@ class gen_config(object):
     batch_size = 64
     emb_dim = 64
     num_layers = 2
-    vocab_size = 25000
+    vocab_size = 2500
     train_dir = "./gen_data/"
     name_model = "st_model"
     tensorboard_dir = "./tensorboard/gen_log/"
@@ -59,7 +59,7 @@ class GSTConfig(object):
     batch_size = 256
     emb_dim = 1024
     num_layers = 2
-    vocab_size = 25000
+    vocab_size = 2500
     train_dir = "./gst_data/"
     name_model = "st_model"
     tensorboard_dir = "./tensorboard/gst_log/"

lessonTen/seqGan chatbotv2.0/utils/data_utils.py

@@ -119,7 +119,7 @@ def initialize_vocabulary(vocabulary_path):
   """
   if gfile.Exists(vocabulary_path):
     rev_vocab = []
-    with gfile.GFile(vocabulary_path, mode="rb") as f:
+    with open(vocabulary_path, mode="r") as f:
       rev_vocab.extend(f.readlines())
     rev_vocab = [line.strip() for line in rev_vocab]
     vocab = dict([(x, y) for (y, x) in enumerate(rev_vocab)])