app.py

# -*- coding: utf-8 -*-
"""Gradio_Demo.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1eXu8deuh8Jl5Mmm0DxX0XlR4-2E6IN0l
"""


import pandas as pd
import seaborn as sns 
import matplotlib.pyplot as plt
import numpy as np

# import the necessary libraries
import nltk
import string
import re

from time import time

from gensim.parsing.preprocessing import STOPWORDS
from gensim.utils import simple_preprocess
from sklearn.model_selection import GridSearchCV

from sklearn.model_selection import train_test_split
from sklearn.utils.class_weight import compute_sample_weight
from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer
from sklearn.metrics import accuracy_score, confusion_matrix
from sklearn.naive_bayes import MultinomialNB
from sklearn.pipeline import Pipeline
from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.linear_model import SGDClassifier,LogisticRegression
from sklearn.ensemble import ExtraTreesRegressor
from sklearn import model_selection, svm
import gradio as gr
from datasets import load_dataset

import warnings
warnings.filterwarnings('ignore')

"""### First Let's read our data"""

url = 'https://raw.githubusercontent.com/omnabill/NLP_Task/main/flask-app/Job%20titles%20and%20industries.csv'
df = pd.read_csv(url)

df.head()

df.tail()

"""### Let's Explore our data to understand it .."""

df.info()

df.describe()

df_count = df['industry'].value_counts().rename_axis('Category').reset_index(name='Count')

df_count.head()

plt.figure(figsize=(10,7))
sns.barplot(x=df_count.Category,y=df_count.Count/len(df))
plt.show()

df_count.Count/len(df)

"""Here there is a Class Less than 5% of existance which is "Accountancy" , So this can be a Way of imbalanced data that can disturb our model .

This is a Problem That should be dealed with before passing our data to a model .

### Let's complete First our EDA for our data .

#### Let's check first for duplicated rows
"""

df[df.duplicated()==True]

"""There are many duplicate rows let's start deal with them through keep one copy of duplicated rows"""

df.drop_duplicates(subset="job title",inplace=True)

df.info()

plt.figure(figsize=(10,6))
df.groupby('industry').count().plot.bar()

df_count = df['industry'].value_counts().rename_axis('Category').reset_index(name='Count')

plt.figure(figsize=(10,7))
sns.barplot(x=df_count.Category,y=df_count.Count/len(df))
plt.show()

df_count.Count/len(df)

"""### Let's Deal With text (Text Preprocessing)

This Through : 

1- Remove Stop words Like 'The','and','of',etc...

2- change numerical numbers to text notation

3- Remove Punctiuations:  We remove punctuations so that we don’t have different forms of the same word. If we don’t remove the punctuation, then been. been, been! will be treated separately.

4- Text Lowercase:  to reduce the size of the vocabulary of our text data.

5- Remove White spaces
"""

import inflect
p = inflect.engine()

def Text_Cleaner(text):
    """
        text: a string 
        return: modified clean string
    """
    result = ""
    
    for token in text.split(' '):
        
        if token not in STOPWORDS and len(token) >= 1:
            
            if token.isdigit():
                
                temp = p.number_to_words(token)  #change numerical number to text
                result+=temp+" "
            else:
                
                if not re.match(r'£[0-9]+',  token):  
                    
                    token = token.lower()  #lower case string
                    result+=token+" "     
    translator = str.maketrans('', '', string.punctuation)  #remove punctiuation signs
    return " ".join(result.translate(translator).split())


df['job title'] = df['job title'].map(Text_Cleaner)
df['job title'].head(20)

"""### Deal With imbalanced data and pass it to model .. 

Train More Than 1 Model For Text Classification

1- SGD Classifier with Hyper Parameter Tuning 

2- Multi nominal Naive Bayes 

3- Logistic Regression

4- SVM (Support vector machine calssifier)

### hyper parameter tuning for SGDClassifier .
"""

X = df['job title']
y = df['industry']
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state = 42)
sample_weight = compute_sample_weight("balanced",y_train)



pipeline = Pipeline([
    ('vect', CountVectorizer()),
    ('tfidf', TfidfTransformer()),
    ('clf', SGDClassifier()),
])

# uncommenting more parameters will give better exploring power but will
# increase processing time in a combinatorial way
parameters = {
    'vect__max_df': (0.5, 0.75, 1.0),
    'vect__max_features': (None, 5000, 10000, 50000),
    'vect__ngram_range': ((1, 1), (1, 2)),  # unigrams or bigrams
    # 'tfidf__use_idf': (True, False),
    # 'tfidf__norm': ('l1', 'l2'),
    'clf__max_iter': (20,),
    'clf__alpha': (0.00001, 0.000001),
    'clf__penalty': ('l2', 'elasticnet'),
    'clf__loss': ['log','hinge'],
    'clf__max_iter': (10, 50, 80),
}


# multiprocessing requires the fork to happen in a __main__ protected
# block

# find the best parameters for both the feature extraction and the
# classifier
grid_search = GridSearchCV(pipeline, parameters, n_jobs=-1, verbose=1)

print("Performing grid search...")
print("pipeline:", [name for name, _ in pipeline.steps])
print("parameters:")
print(parameters)
t0 = time()
grid_search.fit(X_train, y_train)
print("done in %0.3fs" % (time() - t0))
print()

print("Best score: %0.3f" % grid_search.best_score_)
print("Best parameters set:")
best_parameters = grid_search.best_estimator_.get_params()
for param_name in sorted(parameters.keys()):
        print("\t%s: %r" % (param_name, best_parameters[param_name]))

make_pipline = Pipeline([('vect', CountVectorizer(max_df=0.5,max_features= 50000,ngram_range= (1, 1))),
               ('tfidf', TfidfTransformer()),
               ('SGD' , SGDClassifier(loss='log', penalty='l2',alpha=1e-05, random_state=42, max_iter=80)),])

make_pipline.fit(X_train, y_train, **{'SGD__sample_weight': sample_weight})

y_pred = make_pipline.predict(X_test)
print('accuracy = {}'.format(accuracy_score(y_pred, y_test)) )



"""### Naive Bayes"""

make_pipline = Pipeline([('vect', CountVectorizer()),
               ('tfidf', TfidfTransformer()),
               ('clf', MultinomialNB()),])

make_pipline.fit(X_train, y_train, **{'clf__sample_weight': sample_weight})

y_pred = make_pipline.predict(X_test)
print('accuracy %s' % accuracy_score(y_pred, y_test))

y_pred

"""### Logistic Regression"""

make_pipline = Pipeline([('vect', CountVectorizer()),
               ('tfidf', TfidfTransformer()),
               ('log' , LogisticRegression(n_jobs=1, C=1e5)),])

make_pipline.fit(X_train, y_train, **{'log__sample_weight': sample_weight})

y_pred = make_pipline.predict(X_test)
print('accuracy %s' % accuracy_score(y_pred, y_test))



"""### Support Vector Machine"""

make_pipline = Pipeline([('vect', CountVectorizer()),
               ('tfidf', TfidfTransformer()),
               ('clf', svm.SVC(C=1.0, kernel='linear', degree=3, gamma='auto',probability=True)),])

make_pipline.fit(X_train, y_train, **{'clf__sample_weight': sample_weight})

y_pred = make_pipline.predict(X_test)
print('accuracy %s' % accuracy_score(y_pred, y_test))

X_test.to_csv('out.csv', index=False)



"""### Confusion Matrix for the final model Used SVM """

a = np.flip(df['industry'].unique())

from sklearn.metrics import confusion_matrix
mat = confusion_matrix(y_test, y_pred)
plt.figure(figsize=(10,8))
ax = sns.heatmap(mat.T, square=True, annot=True, fmt='d', cbar=False,
            xticklabels=a, yticklabels=a)
bottom, top = ax.get_ylim()
ax.set_ylim(bottom + 0.5, top - 0.5)

plt.title('Confussion Matrix')
plt.xlabel('true label')
plt.ylabel('predicted label');

"""IT is the highest class which missed it's labels , while it appeares that Marketing and education missed the leaset"""

from sklearn import metrics 
print(metrics.classification_report(y_test, y_pred, target_names=a))

"""The performance measurment prove our thoughts from confusion matrix as the marketing and education are the highest classes with true positive rate precision that for classes that are marketing or education and they are classified as marketing or education
while Accountancy have the leaset true positive rate 
"""

metrics.roc_auc_score(y_test, make_pipline.predict_proba(X_test), multi_class='ovr')

"""The Area Under the Receiver Operating Characteristic Curve (AUC) is between 0.5 to 1 which 1 denates the out standing performance of the model, 0.5 the opposite , here 0.97 is a very good score that indicates that our model performance is very good .

### Deploy model as Gradio Api Service
"""

def Job_Class(text):
    # img = img.reshape(1, 100, 100, 1)
    prediction = make_pipline.predict([text])
    # class_names = ["Accountancy", "Education","IT","Marketing"]
    return prediction[0]

#set the user uploaded image as the input array
#match same shape as the input shape in the model
# im = gr.inputs.Image(shape=(100, 100), image_mode='L', invert_colors=False, source="upload")

#setup the interface
iface = gr.Interface(
  fn=Job_Class, 
  inputs=gr.inputs.Textbox(lines=2, placeholder="Write here the job description ..."), 
  outputs="text")
iface.launch(share=True)

# prediction = make_pipline.predict(['senior technical support engineer'])

# type(prediction[0])