training_companies.py

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

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1xkxzi8wZ5rOC731yBeITMF-BXjJGh7ui
"""

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Embedding, LSTM, Dense
import pandas as pd

from sklearn.feature_extraction.text import CountVectorizer
import tensorflow as tf
import re
import argparse

from keras.preprocessing.sequence import pad_sequences
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import LabelBinarizer
import numpy as np
from sklearn.model_selection import train_test_split

parser = argparse.ArgumentParser()
parser.add_argument('-p', '--path', help='CSV path')
parser.add_argument('-s', '--save', help='Save path')

args = parser.parse_args()
path = args.path
saved_path = args.save

# '/content/drive/MyDrive/bookings_Erwin.csv'

df = pd.read_csv(path, low_memory= False)

df.dropna(subset = ["booking_account"], inplace=True)

## cleaning texts

df['description'] = df.apply(lambda row : row['description'].lower(), axis = 1)
df['description'] = df.apply(lambda row : re.sub(r'[0-9]', '', row['description']), axis = 1)
df['description'] = df.apply(lambda row : re.sub(r'[\W_]+', ' ', row['description']), axis = 1)


vectorizer = CountVectorizer(min_df=0, lowercase=False)
vectorizer.fit(df['description'])
corpus  = vectorizer.vocabulary_

reverse_corpus = dict([(value, key) for (key, value) in corpus.items()])

def text_to_num(s):
  temp = []
  s = s.split(' ')
  for word in s:
    if word in corpus: temp.append(corpus[word])
    else: temp.append(0)
  return temp

df['X'] = df.apply(lambda row: text_to_num(row['description']), axis = 1)


num_words= 5
data = pad_sequences(df['X'], maxlen=num_words)

bin = LabelBinarizer()
y = bin.fit_transform(df.booking_account)

X_train, X_test, y_train, y_test = train_test_split(data, y, test_size=0.2, random_state=42)

model = Sequential()
model.add(Embedding(875, 64, input_length= 5))
model.add(LSTM(64, dropout=0.2, recurrent_dropout=0.2, return_sequences=True))
model.add(LSTM(64))
model.add(Dense(21, activation='relu'))

model.compile(loss='MSE',
              optimizer='adam',
              metrics=['accuracy'])

labels = bin.classes_.tolist()

model.fit(X_train, y_train,
          batch_size=1000,
          epochs=30,
          validation_data=(X_test, y_test))
print('Evaluating >>>>>')
score, acc = model.evaluate(X_test, y_test,
                            batch_size=1000)


print(X_test[1].shape)
xx = np.expand_dims(X_test[100], axis = 0)
yy = np.expand_dims(y_test[100], axis = 0)

print('Original result:', bin.inverse_transform(yy))

hist = model.predict(xx)
print(hist)
print('Prediction:', labels[np.argmax(hist)])

# '/content/saved_erwin'
model.save(saved_path)