Merge branch 'master' of https://github.com/elvisun/Chinese-poem-gene…

…rator

README.md

@@ -5,6 +5,20 @@ Chinese poem generator with LSTM network
 
 > trained on one GTX1080 for 12 hours. Went through 35% of the data and both loss and validation loss settled. [See full training output here ](./result.md)
 
+## Findings
+The first few epochs started with random(very rare) charaters, with no or random punctuations. 
+It figured out how to add punctuations into sentences after 5 epochs- one comma and one period. 
+
+After around 30 epochs the charaters used start to have a theme.
+
+Adding dropout layer helped improve the issue of overfitting - because punctuations were so frequent comparing to other charaters, it was just generating lines of punctuations with no other contents. 
+
+## Future Works
+Add acrostic poem（藏头诗）feature
+Does it start to rhyme at some point?
+Try GAN network. 
+
+## Output 
 ==================Epoch 0=====================
 
 
@@ -43,4 +57,4 @@ Chinese poem generator with LSTM network
 。须镜何平心，行阴世白身。若国为路想，湖尔大雾因。山在雪断老，然闭绿尘下。送西土无君，罗云叶何园。心音何自未，会关醉花乐。欲上枝碧露，玉江唯上入。蜀林衣如寒，子时未深犹。心来天今半，莫静山山金。流高何
 
 ------------Diversity 1.5--------------
-，为纵得尽风。门临觉归轻，言行共露众。太我低分与，不即浪汉食。百看宵须日，须期谁公与。霭林忧忽室，携溪景新城。满霁人情晨，雨兴千看军。赋怀还当君，永阁夕霞前。乐龙张青好，迹浦更飞穷。空尚得为流，未生诗
+，为纵得尽风。门临觉归轻，言行共露众。太我低分与，不即浪汉食。百看宵须日，须期谁公与。霭林忧忽室，携溪景新城。满霁人情晨，雨兴千看军。赋怀还当君，永阁夕霞前。乐龙张青好，迹浦更飞穷。空尚得为流，未生诗