Intent Classification

Credit: AITS Cainvas Community

Photo by Jakub Jezovic 🦔 on Dribbble

# get the data file
!wget -N https://cainvas-static.s3.amazonaws.com/media/user_data/cainvas-admin/train.csv

--2021-07-14 15:08:10--  https://cainvas-static.s3.amazonaws.com/media/user_data/cainvas-admin/train.csv
Resolving cainvas-static.s3.amazonaws.com (cainvas-static.s3.amazonaws.com)... 52.219.160.59
Connecting to cainvas-static.s3.amazonaws.com (cainvas-static.s3.amazonaws.com)|52.219.160.59|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 441947 (432K) [text/csv]
Saving to: ‘train.csv’

train.csv           100%[===================>] 431.59K  --.-KB/s    in 0.005s  

2021-07-14 15:08:10 (88.6 MB/s) - ‘train.csv’ saved [441947/441947]

# import the required libraries
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.utils import shuffle
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.layers import LSTM, Dense, Bidirectional, Dropout, Dense, Activation, Flatten, Embedding
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing import sequence
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Sequential,Model,load_model
from tensorflow.keras.optimizers import SGD

# read into csv file
df = pd.read_csv('train.csv')
df.head()

	sentence	BookRestaurant	GetWeather	PlayMusic	RateBook
0	book The Middle East restaurant in IN for noon	1	0	0	0
1	Book a table at T-Rex distant from Halsey St.	1	0	0	0
2	I'd like to eat at a taverna that serves chili...	1	0	0	0
3	I have a party of four in Japan and need a res...	1	0	0	0
4	Please make a restaurant reservation for somew...	1	0	0	0

# shuffle data
df = shuffle(df)
df.head()

	sentence	BookRestaurant	GetWeather	PlayMusic	RateBook
6013	Rate the current series a one	0	0	0	1
4709	Play the playlist Grime Shutdown	0	0	1	0
5976	rate The Ape-Man Within 4	0	0	0	1
6814	rate the current essay three of 6\n points	0	0	0	1
6146	give Portions from a Wine-stained Notebook: Sh...	0	0	0	1

X_train = df["sentence"].fillna("fillna").values
y_train = df[["BookRestaurant", "GetWeather", "PlayMusic", "RateBook"]].values

X_train.shape, y_train.shape

((7929,), (7929, 4))

X_train[1], y_train[1] 

('Play the playlist Grime Shutdown', array([0, 0, 1, 0]))

Preprocessing

text = X_train

Tokenizer = Tokenizer()

# text preprocessing
Tokenizer.fit_on_texts(text) 
Tokenizer_vocab_size = len(Tokenizer.word_index) + 1
Tokenizer_vocab_size

7522

X_train.shape, y_train.shape

((7929,), (7929, 4))

samples = 2500

X_train = X_train[samples:] 
y_train = y_train[samples:] 

X_val = X_train[:samples] 
y_val = y_train[:samples] 

X_train.shape, y_train.shape

((5429,), (5429, 4))

X_train_encoded_words = Tokenizer.texts_to_sequences(X_train)
X_val_encoded_words = Tokenizer.texts_to_sequences(X_val)

X_train_encoded_padded_words = sequence.pad_sequences(X_train_encoded_words, maxlen = 100)
X_val_encoded_padded_words = sequence.pad_sequences(X_val_encoded_words, maxlen = 100)

X_train_encoded_padded_words.shape, X_val_encoded_padded_words.shape

((5429, 100), (2500, 100))

X_val_encoded_padded_words, X_train_encoded_padded_words

(array([[   0,    0,    0, ...,   94,   36,   32],
        [   0,    0,    0, ...,    3,    1,  304],
        [   0,    0,    0, ...,   23,    5,   11],
        ...,
        [   0,    0,    0, ...,    8,  239, 5753],
        [   0,    0,    0, ...,    3, 5754,  419],
        [   0,    0,    0, ...,   29,    4,   70]], dtype=int32),
 array([[  0,   0,   0, ...,  94,  36,  32],
        [  0,   0,   0, ...,   3,   1, 304],
        [  0,   0,   0, ...,  23,   5,  11],
        ...,
        [  0,   0,   0, ...,  23,   5,  11],
        [  0,   0,   0, ...,   5,  11,  32],
        [  0,   0,   0, ...,  78,   8, 466]], dtype=int32))

y_train.shape, y_val.shape

((5429, 4), (2500, 4))

Build and Train the Model

model = Sequential()

model.add(Embedding(Tokenizer_vocab_size, 32, input_length = 100)) 

model.add(LSTM(10))
model.add(Dropout(0.5))
model.add(Dense(400, activation='relu'))
model.add(Dropout(0.5))
#model.add(Dense(200, activation='relu'))
#model.add(Dropout(0.5))
model.add(Dense(4, activation='softmax'))

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
embedding (Embedding)        (None, 100, 32)           240704    
_________________________________________________________________
lstm (LSTM)                  (None, 10)                1720      
_________________________________________________________________
dropout (Dropout)            (None, 10)                0         
_________________________________________________________________
dense (Dense)                (None, 400)               4400      
_________________________________________________________________
dropout_1 (Dropout)          (None, 400)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 4)                 1604      
=================================================================
Total params: 248,428
Trainable params: 248,428
Non-trainable params: 0
_________________________________________________________________

Nadam = tf.keras.optimizers.Nadam(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08, schedule_decay=0.004)
model.compile(loss='categorical_crossentropy', optimizer=Nadam, metrics=['accuracy'])

history  = model.fit(X_train_encoded_padded_words,y_train, epochs = 3, batch_size=32, verbose=1, validation_data=(X_val_encoded_padded_words, y_val))

Epoch 1/3
170/170 [==============================] - 2s 10ms/step - loss: 0.4124 - accuracy: 0.8598 - val_loss: 0.0185 - val_accuracy: 0.9964
Epoch 2/3
170/170 [==============================] - 1s 7ms/step - loss: 0.0807 - accuracy: 0.9711 - val_loss: 0.0101 - val_accuracy: 0.9980
Epoch 3/3
170/170 [==============================] - 1s 7ms/step - loss: 0.0580 - accuracy: 0.9801 - val_loss: 0.0042 - val_accuracy: 0.9992

# save the model
model.save("Intent_Classification.h5")

Plots

#loss
from matplotlib import pyplot as plt
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('model loss')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='upper right')
plt.show()

#accuracy
from matplotlib import pyplot as plt
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='lower right')
plt.show()

Loss and Accuracy

model.evaluate(X_val_encoded_padded_words, y_val)

79/79 [==============================] - 0s 2ms/step - loss: 0.0042 - accuracy: 0.9992

[0.004162235651165247, 0.9991999864578247]

Predictions

def predict(text):
    sentence = text
    tokens = Tokenizer.texts_to_sequences([text])
    tokens = pad_sequences(tokens, maxlen = 100)
    prediction = model.predict(np.array(tokens))
    pred = np.argmax(prediction)
    classes = ['BookRestaurant','GetWeather','PlayMusic','RateBook']
    result = classes[pred]
    return result

predict("is it raining ?")

'GetWeather'

predict("i would like to book a table at hotel Orion for 29th june")

'BookRestaurant'

predict("An Artist released a new music album ")

'PlayMusic'

predict("this novel deserves a rating of 10")

'RateBook'