#!/usr/bin/env python
# coding: utf-8

# In[1]:


get_ipython().run_line_magic('reload_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')
get_ipython().run_line_magic('matplotlib', 'inline')
import os
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID";
os.environ["CUDA_VISIBLE_DEVICES"]="0"; 


# In[2]:


import ktrain
from ktrain import text


# # Building a Chinese-Language Sentiment Analyzer
# 
# In this notebook, we will build a Chinese-language text classification model in 4 simple steps. More specifically, we will build a model that classifies Chinese hotel reviews as either positive or negative.
# 
# The dataset can be downloaded from Chengwei Zhang's GitHub repository [here](https://github.com/Tony607/Chinese_sentiment_analysis/tree/master/data/ChnSentiCorp_htl_ba_6000).
# 
# (**Disclaimer:** I don't speak Chinese. Please forgive mistakes.)  

# ## STEP 1:  Load and Preprocess the Data
# 
# First, we use the `texts_from_folder` function to load and preprocess the data.  We assume that the data is in the following form:
# ```
#     ├── datadir
#     │   ├── train
#     │   │   ├── class0       # folder containing documents of class 0
#     │   │   ├── class1       # folder containing documents of class 1
#     │   │   ├── class2       # folder containing documents of class 2
#     │   │   └── classN       # folder containing documents of class N
# ```
# We set `val_pct` as 0.1, which will automatically sample 10% of the data for validation.  We specifiy `preprocess_mode='standard'` to employ normal text preprocessing.  If you are using the BERT model (i.e., 'bert'), you should use `preprocess_mode='bert'`.
# 
# **Notice that there is nothing speical or extra we need to do here for non-English text.**  *ktrain* automatically detects the language and character encoding and prepares the data and configures the model appropriately.
# 
# 

# In[3]:


(x_train, y_train), (x_test, y_test), preproc = text.texts_from_folder('data/ChnSentiCorp_htl_ba_6000', 
                                                                       maxlen=75, 
                                                                       max_features=30000,
                                                                       preprocess_mode='standard',
                                                                       train_test_names=['train'],
                                                                       val_pct=0.1,
                                                                       classes=['pos', 'neg'])


# ## STEP 2:  Create a Model and Wrap in Learner Object

# In[4]:


model = text.text_classifier('fasttext', (x_train, y_train) , preproc=preproc)
learner = ktrain.get_learner(model, 
                             train_data=(x_train, y_train), 
                             val_data=(x_test, y_test), 
                             batch_size=32)


# ## STEP 3: Estimate the LR
# We'll use the *ktrain* learning rate finder to find a good learning rate to use with *fasttext*. We select a high learning rate that is associated with a still falling loss from the plot.
# 

# In[5]:


learner.lr_find(show_plot=True)


# ## STEP 4: Train the Model
# 
# We will use the `fit_onecycle` method that employs a [1cycle learning rate policy](https://arxiv.org/pdf/1803.09820.pdf) for 10 epochs (i.e., roughly 20 seconds).

# In[10]:


learner.fit_onecycle(5e-3, 10)


# In[11]:


learner.validate(class_names=preproc.get_classes())


# ### Inspecting Misclassifications

# In[12]:


learner.view_top_losses(n=1, preproc=preproc)


# Using Google Translate, the above roughly translates to:
# ```
# 
# The so-called mountain view room is just a non-sea view room, there is no mountain view at all, the sea view room is indeed, there are conditions to choose as much as possible. It’s just that the curtains in this room are not pulled up. The morning light comes in like a lit lamp, which may be triggered. In addition, the sound insulation outside the window is not good. If the whirring is obvious, it must be no longer necessary.
# ```
# 
# Mistranslations aside, this is clearly a negative review.  It appears to have been incorrectly assigned a ground-truth label of positive.

# ### Making Predictions on New Data

# In[13]:


p = ktrain.get_predictor(learner.model, preproc)


# Predicting label for the text
# > "*The view and service of this hotel were terrible and our room was dirty.*"

# In[16]:


p.predict("这家酒店的看法和服务都很糟糕，我们的房间很脏。")


# Predicting label for:
# > "*I like the service of this hotel.*"

# In[17]:


p.predict('我喜欢这家酒店的服务')


# ### Saving Predictor for Later Deployment

# In[18]:


p.save('/tmp/mypred')


# In[19]:


p = ktrain.load_predictor('/tmp/mypred')


# In[20]:


# still works
p.predict("这家酒店的风景和服务都非常糟糕")


# In[ ]: