update

README.md

@@ -5,25 +5,49 @@ Fine grain sentiment analysis from AI challenger, with data and runnable code an
 
 Important Notice: this project is based on pengshuang/AI-Comp's project(baseline), most of code is reused here. 
 
-Here we reused this baseline, the main purpose is to save time as this code is runnable and have a reasonable performance 
+the main reason why we reused this baseline is to save time as this baseline is runnable and has a reasonable performance 
 
-already. On line f1 score is 0.702 as reported in this baseline. we just want to try some new ideas, especially some new
-
-models, new technique introduced recently such as pre-train of language model on large corpus and gain fine. this project 
+already. Online f1 score is 0.702 as reported in this baseline. since we just want to try some new ideas, to do some improvement,
 
-is on its early stage.
+for example, we want to add some new models, and use some new techniques introduced recently such as pre-train of language model 
+
+on large corpus and gain fine. this project is on its early stage.
 
 ## Experiment on New Models
 
 add something here.
 
 ## Usage
 
+    1. generate train/validation/test set:
+       
+       Preprocess_char.ipynb
+
+    2. train the model:
+       
+       python model_*_char.py to 
+
+    3. make prediction using validation data, and write prediction to file
+
+       python validation_*_char.py
+
+    4. compute f1 score on validation set
+     
+       python evaluate_char.py
+      
+    5. submmit prediction on test set, and generate submition file 
+
+       python predict_*_char.py
+
+
     class_*.py file for model
     model_*_char.py file for training
     validation_*_char.py genreate validation result  
     evaluate_char.py compute f1 score on validation set 
     predict_*_char.py generate online submit file 
 
 ## Reference
-1. <a href='https://github.com/pengshuang/AI-Comp'>pengshuang/AI-Comp</a>
+
+1. <a href='https://github.com/pengshuang/AI-Comp'>pengshuang/AI-Comp</a>
+
+2. <a href='https://github.com/AIChallenger/AI_Challenger_2018'>AI Challenger 2018</a>