Deploy Machine Learning Model for simple use. Part II

July 13, 2020

Deploy Machine Learning Model for simple use. Part II

4th week of blogging
Hai,Everyone 👋 ,
Without further adieu let us get started.
we have already set up the flask app and HTML to get the input, now lets put the model to work. To put our Model we will create a file called decision.py and import it in from decision import Emotion in our flask_app.py file. To load model

import json
	import keras
	import numpy as np
	import sklearn
	from sklearn.preprocessing import LabelBinarizer
	from sklearn.preprocessing import LabelEncoder
	from tensorflow.keras.layers import Input, Embedding, SpatialDropout1D, LSTM
	from tensorflow.keras.layers import GlobalAveragePooling1D, GlobalMaxPooling1D
	from tensorflow.keras.layers import Bidirectional, Conv1D, Dense, concatenate
	from tensorflow.keras.models import Model
	import pickle
	from tensorflow.keras.preprocessing.sequence import pad_sequences
	import os
	THIS_FOLDER = os.path.dirname(os.path.abspath(__file__))
	 
	def token():
	 file = os.path.join(THIS_FOLDER, 'static/tokenizer.pickle')
	 infile = open(file,'rb')
	 tokenizer = pickle.load(infile)
	 infile.close()
	 return tokenizer
	 
	def encoder():
	 labels = ['Happy','Sad','Anger','Disgust','Surprise','Fear','Bad']
	 encoder = LabelBinarizer()
	 encoder.fit(labels)
	 return encoder
	 
	def model():
	 input_length = 428
	 input_dim = 203169
	 num_classes = 7
	 embedding_dim = 500
	 lstm_units = 128
	 lstm_dropout = 0.1
	 recurrent_dropout = 0.1
	 filters=64
	 kernel_size=3
	 input_layer = Input(shape=(input_length,))
	 output_layer = Embedding(
	 input_dim=input_dim,
	 output_dim=embedding_dim,
	 input_shape=(input_length,)
	 )(input_layer)
	 output_layer = Bidirectional(
	 LSTM(lstm_units, return_sequences=True,
	 dropout=lstm_dropout, recurrent_dropout=recurrent_dropout)
	 )(output_layer)
	 output_layer = Conv1D(filters, kernel_size=kernel_size, padding='valid',
	 kernel_initializer='glorot_uniform')(output_layer)
	 
	 avg_pool = GlobalAveragePooling1D()(output_layer)
	 max_pool = GlobalMaxPooling1D()(output_layer)
	 output_layer = concatenate([avg_pool, max_pool])
	 
	 output_layer = Dense(num_classes, activation='softmax')(output_layer)
	 model = Model(input_layer, output_layer)
	 file = os.path.join(THIS_FOLDER, 'static/model.h5')
	 model.load_weights(file)
	 return model
	 
	 
	class Emotion:
	 
	 
	 
	 def __init__(self):
	 self.model = model()
	 self.tokenizer = token()
	 self.encoder = encoder()
	 
	 def test(self,text):
	 labels = ['Happy','Sad','Anger','Disgust','Surprise','Fear','Bad']
	 tokenized = self.tokenizer.texts_to_sequences([text])
	 pad_data = pad_sequences(tokenized,428)
	 pred = self.model.predict(pad_data)
	 print(pred,flush=True)
	 emotion = labels[np.argmax(pred)]
	 confidence = pred[0][np.argmax(pred)] * 100
	 return (emotion,confidence)

Here we design the Model using Keras, then load the weights from the saved .h5 file for easy access, we create a class for model Emotion, which contains a test method that tests the model with input after initializing model, tokenizer, encoder as the result we return emotion and confidence from the model, which is rendered with HTML for the user to view.

Deploying to Pythonanywhere

Host the script pythonanywhere by following a few simple steps:

Step 1. Sign up for a new account.

For now, let’s stick with the free account. Sign up and log in to your account. Choose the flask and whichever python version you want. I will be using Python 3.7. After creating the web app, you will get a URL that points to your flask endpoint. By default, Your endpoint looks something like this: [username].pythonanywhere.com

Step 2. Upload the files

Inside the default folder — /mysite/ you need to upload your complete folder. You can do it either using the files page on the website or using the bash console by using wget command to download your files. and enable Force HTTPS File Structure looks like this

/mysite/
 |
 +-- flask_app.py
 | 
 +-- decision.py
 | 
 +-- static/
 | | 
 | +-- encoder.pickle
 | | 
 | +-- lstm_model.h5
 | | 
 | +-- tokenizer.pickle
 +-- templates/
 | | 
 | +-- index.html

Step 3. Reload the web app

Web Application is ready Now and can be used for predicting emotions of text.