import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import train_test_split
import tensorflow as tf
from sklearn.metrics import classification_report, confusion_matrix, accuracy_score

print("---------------------- Downloading Dataset -------------------------\n")

dataset = pd.read_csv('https://raw.githubusercontent.com/adityaiiitmk/Datasets/master/SMSSpamCollection',sep='\t',names=['label','message'])

print("----------------------  -------------------------\n")
print(dataset.head())
print("----------------------  -------------------------")
print(dataset.groupby('label').describe())
print("----------------------  -------------------------")
dataset['label'] = dataset['label'].map( {'spam': 1, 'ham': 0} )
X = dataset['message'].values
y = dataset['label'].values

print("---------------------- Train Test Split -------------------------\n")
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=42)
tokeniser = tf.keras.preprocessing.text.Tokenizer()
tokeniser.fit_on_texts(X_train)        #learn vocabulary assigning unique int to each word
encoded_train = tokeniser.texts_to_sequences(X_train)
encoded_test = tokeniser.texts_to_sequences(X_test)
print(encoded_train[0:2])
print("----------------------  Padding  -------------------------\n")
max_length = 10   #legth of sequence 
padded_train = tf.keras.preprocessing.sequence.pad_sequences(encoded_train, maxlen=max_length, padding='post')
padded_test = tf.keras.preprocessing.sequence.pad_sequences(encoded_test, maxlen=max_length, padding='post')
print(padded_train[0:2])

print("----------------------  -------------------------\n")

vocab_size = len(tokeniser.word_index)+1

# define the model

print("---------------------- Modelling -------------------------\n")


model=tf.keras.models.Sequential([
   tf.keras.layers.Embedding(input_dim=vocab_size,output_dim= 24, input_length=max_length),
   tf.keras.layers.SimpleRNN(24, return_sequences=False),  #return last time output
   tf.keras.layers.Dense(64, activation='relu'),
   tf.keras.layers.Dense(32, activation='relu'),
   tf.keras.layers.Dense(1, activation='sigmoid')
])

# compile the model
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

print("----------------------  -------------------------\n")

# summarize the model
print(model.summary())

print("----------------------  -------------------------\n")

early_stop = tf.keras.callbacks.EarlyStopping(monitor='accuracy', mode='min', patience=10)  #callback set of functions can applied different stages of training

print("----------------------  Training -------------------------\n")

# fit the model
model.fit(x=padded_train,
         y=y_train,
         epochs=50,
         validation_data=(padded_test, y_test),
         callbacks=[early_stop]
         )
print("----------------------  -------------------------\n")


def c_report(y_true, y_pred):
   print("Classification Report")
   print(classification_report(y_true, y_pred))
   acc_sc = accuracy_score(y_true, y_pred)
   print(f"Accuracy : {str(round(acc_sc,2)*100)}")
   return acc_sc

def plot_confusion_matrix(y_true, y_pred):
   mtx = confusion_matrix(y_true, y_pred)
   sns.heatmap(mtx, annot=True, fmt='d', linewidths=.5, cmap="Blues", cbar=False)
   plt.ylabel('True label')
   plt.xlabel('Predicted label')
   plt.savefig(r"C:\Users\Dell\Documents\Sem !!!\Deep Learning\App\Results\test.jpg")


preds = (model.predict(padded_test) > 0.5).astype("int32")
c_report(y_test, preds)

plot_confusion_matrix(y_test, preds)
model.save("RNN/results/model/spam_model")