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

# # Tutorial to invoke LIME explainers via aix360
# 
# There are two ways to use [LIME](https://github.com/marcotcr/lime) explainers after installing aix360:
# - [Approach 1  (aix360 style)](#approach1): LIME explainers can be invoked in a manner similar to other explainer algorithms in aix360 via the implemented wrapper classes.
# - [Approach 2 (original style)](#approach2): Since LIME comes pre-installed in aix360, the explainers can simply be invoked directly.
# 
# This notebook showcases both these approaches to invoke LIME. The notebook is based on the following example from the original LIME tutorial: https://marcotcr.github.io/lime/tutorials/Lime%20-%20multiclass.html

# ## Approach 1  (aix360 style)
# <a id='approach1'></a>
# - Note the import statement related to LimeTextExplainer

# In[1]:


from __future__ import print_function
import sklearn
import numpy as np
import sklearn
import sklearn.ensemble
import sklearn.metrics

# Importing LimeTextExplainer (aix360 sytle)
from aix360.algorithms.lime import LimeTextExplainer


# In[2]:


# Supress jupyter warnings if required for cleaner output
import warnings
warnings.simplefilter('ignore')


# ### Fetching data, training a classifier

# In[3]:


from sklearn.datasets import fetch_20newsgroups
newsgroups_train = fetch_20newsgroups(subset='train')
newsgroups_test = fetch_20newsgroups(subset='test')
# making class names shorter
class_names = [x.split('.')[-1] if 'misc' not in x else '.'.join(x.split('.')[-2:]) for x in newsgroups_train.target_names]
class_names[3] = 'pc.hardware'
class_names[4] = 'mac.hardware'


# In[4]:


print(','.join(class_names))


# In[5]:


vectorizer = sklearn.feature_extraction.text.TfidfVectorizer(lowercase=False)
train_vectors = vectorizer.fit_transform(newsgroups_train.data)
test_vectors = vectorizer.transform(newsgroups_test.data)


# In[6]:


from sklearn.naive_bayes import MultinomialNB
nb = MultinomialNB(alpha=.01)
nb.fit(train_vectors, newsgroups_train.target)


# In[7]:


pred = nb.predict(test_vectors)
sklearn.metrics.f1_score(newsgroups_test.target, pred, average='weighted')


# ### Explaining predictions using lime

# In[8]:


from sklearn.pipeline import make_pipeline
c = make_pipeline(vectorizer, nb)


# In[9]:


print(c.predict_proba([newsgroups_test.data[0]]).round(3))


# In[10]:


limeexplainer = LimeTextExplainer(class_names=class_names)
print(type(limeexplainer))


# In[11]:


idx = 1340
# aix360 style for explaining input instances
exp = limeexplainer.explain_instance(newsgroups_test.data[idx], c.predict_proba, num_features=6, labels=[0, 17])
print('Document id: %d' % idx)
print('Predicted class =', class_names[nb.predict(test_vectors[idx]).reshape(1,-1)[0,0]])
print('True class: %s' % class_names[newsgroups_test.target[idx]])


# In[12]:


print ('Explanation for class %s' % class_names[0])
print ('\n'.join(map(str, exp.as_list(label=0))))
print ()
print ('Explanation for class %s' % class_names[17])
print ('\n'.join(map(str, exp.as_list(label=17))))


# In[13]:


# aix360 style for explaining input instances
exp = limeexplainer.explain_instance(newsgroups_test.data[idx], c.predict_proba, num_features=6, top_labels=2)
print(exp.available_labels())


# In[14]:


exp.show_in_notebook(text=False)


# In[15]:


exp.show_in_notebook(text=newsgroups_test.data[idx], labels=(0,))


# ## Approach 2 (original style)
# <a id='approach2'></a>
# - Note the import statement related to LimeTextExplainer

# In[16]:


from __future__ import print_function
import sklearn
import numpy as np
import sklearn
import sklearn.ensemble
import sklearn.metrics

# Importing LimeTextExplainer (original style)
from lime.lime_text import LimeTextExplainer


# In[17]:


# Supress jupyter warnings if required for cleaner output
import warnings
warnings.simplefilter('ignore')


# ### Fetching data, training a classifier

# In[18]:


from sklearn.datasets import fetch_20newsgroups
newsgroups_train = fetch_20newsgroups(subset='train')
newsgroups_test = fetch_20newsgroups(subset='test')
# making class names shorter
class_names = [x.split('.')[-1] if 'misc' not in x else '.'.join(x.split('.')[-2:]) for x in newsgroups_train.target_names]
class_names[3] = 'pc.hardware'
class_names[4] = 'mac.hardware'


# In[19]:


print(','.join(class_names))


# In[20]:


vectorizer = sklearn.feature_extraction.text.TfidfVectorizer(lowercase=False)
train_vectors = vectorizer.fit_transform(newsgroups_train.data)
test_vectors = vectorizer.transform(newsgroups_test.data)


# In[21]:


from sklearn.naive_bayes import MultinomialNB
nb = MultinomialNB(alpha=.01)
nb.fit(train_vectors, newsgroups_train.target)


# In[22]:


pred = nb.predict(test_vectors)
sklearn.metrics.f1_score(newsgroups_test.target, pred, average='weighted')


# ### Explaining predictions using lime

# In[23]:


from sklearn.pipeline import make_pipeline
c = make_pipeline(vectorizer, nb)


# In[24]:


print(c.predict_proba([newsgroups_test.data[0]]).round(3))


# In[25]:


explainer = LimeTextExplainer(class_names=class_names)
print(type(explainer))


# In[26]:


idx = 1340
# LIME original style for explaining input instances
exp = explainer.explain_instance(newsgroups_test.data[idx], c.predict_proba, num_features=6, labels=[0, 17])
print('Document id: %d' % idx)
print('Predicted class =', class_names[nb.predict(test_vectors[idx]).reshape(1,-1)[0,0]])
print('True class: %s' % class_names[newsgroups_test.target[idx]])


# In[27]:


print ('Explanation for class %s' % class_names[0])
print ('\n'.join(map(str, exp.as_list(label=0))))
print ()
print ('Explanation for class %s' % class_names[17])
print ('\n'.join(map(str, exp.as_list(label=17))))


# In[28]:


# LIME original style for explaining input instances
exp = explainer.explain_instance(newsgroups_test.data[idx], c.predict_proba, num_features=6, top_labels=2)
print(exp.available_labels())


# In[29]:


exp.show_in_notebook(text=False)


# In[30]:


exp.show_in_notebook(text=newsgroups_test.data[idx], labels=(0,))