Series 1 LSTM Gender Classification Tensorflow

Hello… this post will be part of a serial posts about how we using deep learning approach for simple text classification model, starting from training the model until serving the model into “production ready” application with TensorflowServing or Flask. This series posts will be divided into 3 posts, the first one is preparing and training our simple model to recognize a gender based on name.

Lets say we want a model for our application needs to predict user’s gender based on their name, so at first we have a datasets a pair of name and gender in .csv file. Then, we will stacking our model with LSTM architecture with input user’s characters name and output gender (male/female). on this series, we will using tensorflow and tensorflow keras with minimum additional components and architecture so you still can train the model on your own laptop without excessive resources.

In this series, I use jupyter notebook from tensorflow docker image so you no need to build the environment and install requirements package one by one. to use tensorflow docker with jupyter please read here: https://www.tensorflow.org/install/docker

• Import Necessary Packages

  import re
  
  import numpy as np 
  import pandas as pd 
  
  import tensorflow as tf
  
  from sklearn.feature_extraction.text import CountVectorizer
  from tensorflow.keras.preprocessing.text import Tokenizer
  from tensorflow.keras.preprocessing.sequence import pad_sequences
  from tensorflow.keras.models import Sequential
  from tensorflow.keras.layers import Input, Dense, Embedding, LSTM, Dropout
  from sklearn.model_selection import train_test_split
  

• Load Datasets

  data = pd.read_csv('name_gender_pair.csv')
  

• Create Character Dictionary for Name Features

  data['name'] = data['name'].apply(lambda x: x.lower())
  
  human_vocab = set()
  for name in data['name']:
        human_vocab.update(tuple(name))
  
  vocab_index = {v: k + 1 for k, v in enumerate(human_vocab)}
  
  print(vocab_index)
  print(len(vocab_index))
  

• Preprocess datasets and split for train and test set

  name_datasets = data['name'].apply(lambda x: [vocab_index[key] for key in list(x)])
  X = pad_sequences(name_datasets)
  Y = pd.get_dummies(data['gender']).values
  
  X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.2, random_state=42)
  
  print('data shape')
  print(X_train.shape, y_train.shape)
  print(X_test.shape, y_test.shape)
  

• Build the classification model using LSTM

  #build model
  model = Sequential()
  
  # use embedding at first layer for handle character dictionary input
  model.add(Embedding(len(vocab_index) + 1, 16, input_length=X.shape[1]))
  model.add(LSTM(64, dropout=0.2, recurrent_dropout=0.2))
  model.add(Dense(2, activation='sigmoid'))
  
  #compile model 
  model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
  
  print('model summary')
  print(model.summary())
  

• Train the model

  batch_size = 64
  n_epochs = 12
  # train the model
  model.fit(X_train, y_train, batch_size=batch_size, epochs=n_epochs, validation_data=(X_test, y_test), verbose=1)
  

• Evaluate the model

  #test evaluate
  score, acc = model.evaluate(X_test, y_test, batch_size=64)
  print('score', score)
  print('accuracy', acc)
  

• Test the model

  #test 
  name = 'Aminarti'
  name = list(name.lower())
  test_dt = [vocab_index[x] for x in name]
  test_dt = pad_sequences([test_dt], maxlen=X.shape[1])
  print(test_dt) #print encoded test input sequence
  
  pad = np.array(test_dt[0])
  # predict with the model
  res = model.predict(pad.reshape(1, pad.shape[0]), batch_size=1, verbose=2)[0]
  print(res)
  if np.argmax(res) == 0:
      print('Female')
  elif np.argmax(res) == 1:
      print('Male')
  

for the complete codes and notebook you can download the jupyter notbook from this repo: https://github.com/yudanta/lstm-gender-classification/blob/master/LSTM-Character-Level-Gender-Classification.ipynb

The second part of the series will be exporting the trained model for tensorflow serving and run with tensorflow serving.

ps. the second post already publised in this link: https://yudanta.github.io/posts/series-2-exporting-lstm-gender-classification-and-serving-with-tensorflowserving/