from google.colab import drive drive.mount('/content/drive')

!wget --header="Host: doc-10-a0-docs.googleusercontent.com" --header="User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.77 Safari/537.36" --header="Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9" --header="Accept-Language: en-US,en;q=0.9" --header="Referer: https://drive.google.com/" --header="Cookie: AUTH_g1nhg9bn7rppf3mem1b5js72hmqescc8_nonce=6m23eklhub6l6; _ga=GA1.2.276044829.1600093820" --header="Connection: keep-alive" "https://doc-10-a0-docs.googleusercontent.com/docs/securesc/okunrllqorgp640n1aj11ir5butkb2rr/6nk26rum3s1kuglq82dcunlarf7lu0b4/1623493800000/00484516897554883881/17936686973419313426/1rxD15nyeIPIAZ-J2VYPrDRZI66-TBWvM?e=download&authuser=0&nonce=6m23eklhub6l6&user=17936686973419313426&hash=ava5qubbbd8282557e86l03mt9av8cj3" -c -O 'documents.rar'

!mkdir textfiles

pip install pyunpack

pip install patool

# Extracting al text files from documents.rar to textfiles from pyunpack import Archive Archive('documents.rar').extractall('/content/textfiles')

import os # extracting the name of the documents and saved in the txt_class list txt_class = [] for file in os.listdir('/content/textfiles/documents'): if file.endswith('.txt'): filename = file.split('.txt')[0].split('_')[0] txt_class.append(filename)

len(txt_class)

### count plot of all the class labels. import matplotlib.pyplot as plt import seaborn as sns plt.figure(figsize=(12,7)) sns.set_theme(style="darkgrid") sns.countplot(y = txt_class) plt.show()

'65882@mimsy.umd.edu'.split('@')[1].split('.')

import re import string from nltk import ne_chunk, pos_tag, word_tokenize from nltk.tree import Tree import nltk nltk.download('popular')

with open('/content/textfiles/documents/'+'alt.atheism_49960.txt',"rb") as fp: k = fp.read().decode('ISO-8859-1') #print(k) l = re.findall(r"[a-z0-9\.\-+_]+@[a-z0-9\.\-+_]+\.[a-z]+", k) #print(l) temp1 = re.findall(r'^Subject.*$',k, re.MULTILINE) sub = temp1[0] sub = sub[7:] #print(sub) #j = re.sub(r'^Subject.*$', "", k, re.MULTILINE) k = re.sub(r"[a-z0-9\.\-+_]+@[a-z0-9\.\-+_]+\.[a-z]+", ' ',k) k = re.sub(r"Subject:.*", " ", k, re.MULTILINE) k = re.sub(r"From.*", " ", k, re.MULTILINE) k = re.sub(r"[\n\t\-\\\/]", "", k, re.MULTILINE) k = re.sub(r".*:", "", k, re.MULTILINE) chunks=(list(ne_chunk(pos_tag(word_tokenize(k))))) for i in chunks: if type(i)==Tree: if i.label() == "GPE": j = i.leaves() #if new_york or bigger name if len(j)>1: gpe = "_".join([term for term,pos in j]) k = re.sub(rf'{j[1][0]}', gpe, k, flags=re.MULTILINE) #replacing york with new_york k = re.sub(rf'\b{j[0][0]}\b',"", k, flags=re.MULTILINE) #deleting new, \b is important # deleting Srikanth Varma if i.label()=="PERSON": for term,pog in i.leaves(): k = re.sub(re.escape(term),"", k, flags=re.MULTILINE) k = re.sub(r'\d',"", k, flags=re.MULTILINE) print(k)

c = [] procc = [] for i in l: p = i.split('@')[1].split('.') for j in p: if (len(j)>2) and (j != 'com') : procc.append(j) b = ' '.join(procc) c.append(b) print(c) print(procc)

preprocessed_email = [] # storing all the peprocessed emails for file in os.listdir('/content/textfiles/documents/'): # for each file with open('/content/textfiles/documents/'+file,"rb") as fp: text = fp.read().decode('ISO-8859-1') # from stackoverflow to read the text emails = re.findall(r"[a-z0-9\.\-+_]+@[a-z0-9\.\-+_]+\.[a-z]+", text) # regular expression to extract emails from documents splitted_emails = [] for email in emails: splitted_emails += email.split('@')[1].split('.') preprocessed_email_list = [] for word in splitted_emails: if (len(word) > 2) and (word != 'com'): preprocessed_email_list.append(word) process = ' '.join(preprocessed_email_list) preprocessed_email.append(process) # removing email from text re.sub(r"[a-z0-9\.\-+_]+@[a-z0-9\.\-+_]+\.[a-z]+", ' ',text)

# we have collected all emails and preprocessed them, this is sample output preprocessed_email[:5]

len(preprocessed_email)

preprocessed_subject = [] # storing the subject for file in os.listdir('/content/textfiles/documents/'): # for each file with open('/content/textfiles/documents/'+file, "rb" ) as fp: # from stackoverflow to read the text text = fp.read().decode('ISO-8859-1') # extracting subject sub = re.findall(r'^Subject:.*$',text , re.MULTILINE) subject = sub[0] subject = subject[9:] #remove all the punctuation for i in string.punctuation: subject = subject.replace(i, " ") # removing Re subject = re.sub(r"Re:", "", subject) preprocessed_subject.append(subject) text = re.sub(r'Subject:.*', "", text, re.MULTILINE) # removing subject text = re.sub(r'Write to:.*',"", text, re.MULTILINE) # removing Write to: text = re.sub(r'From:.*', "", text, re.MULTILINE) # removing From: text = re.sub(r'<.*>', "", text, re.MULTILINE) # removing tags like "<anyword>" text = re.sub(r'[\(\[].*?[\)\]]',"", text, re.MULTILINE) # removing data present in brackets '\(.*\)' text = re.sub(r'[\n\t\-\\\/]', "", text, re.MULTILINE) # removing all the newlines('\n'), tabs('\t'), "-", "\" text = re.sub(r".*:", "", text, re.MULTILINE) # removing words which ends with ":" #can't -> can not, 's -> is, i've -> i have, i'm -> i am, you're -> you are, i'll --> i will text = re.sub(r"can\'t", "can not", text) text = re.sub(r"won\'t", "will not", text) text = re.sub(r"i\'ve", "i have", text) text = re.sub(r"i\'m", "i am", text) text = re.sub(r"you\'re", "you are", text) text = re.sub(r"i\'ll", "i will", text) text = re.sub(r"n'\t", "not", text) text = re.sub(r"\'s", "is", text) text = re.sub(r"\'d", "would", text) text = re.sub(r"\'ll", "will", text) text = re.sub(r"\'t", "not", text) text = re.sub(r"\'ve", "have", text) text = re.sub(r"\'m", "am", text) # chuncking chunks=(list(ne_chunk(pos_tag(word_tokenize(text))))) for i in chunks: if type(i)==Tree: if i.label() == "GPE": j = i.leaves() #if new_york or bigger name if len(j)>1: gpe = "_".join([term for term,pos in j]) text = re.sub(rf'{j[1][0]}', gpe, text, flags=re.MULTILINE) #replacing york with new_york text = re.sub(rf'\b{j[0][0]}\b',"", text, flags=re.MULTILINE) #deleting new, \b is important # deleting Srikanth Varma if i.label()=="PERSON": for term,pog in i.leaves(): text = re.sub(re.escape(term),"", text, flags=re.MULTILINE) # Removing digits text = re.sub(r'\d'," ", text, re.MULTILINE) # Replacing “_word_” , “_word” , “word_” to word #replace _word_ to word text = re.sub(r"\b_([a-zA-z]+)_\b",r"\1", text) # replace _word to word text = re.sub(r"\b_([a-zA-z]+)\b",r"\1", text) # replace word_ to word text = re.sub(r"\b([a-zA-z]+)_\b",r"\1", text) # d_berlin to berlin text = re.sub(r"\b[a-zA-Z]{1}_([a-zA-Z]+)",r"\1", text) # de_berlin to berlin text = re.sub(r"\b[a-zA-Z]{2}_([a-zA-Z]+)",r"\1", text) # lower case text = text.lower() # remove words >15 text = re.sub(r"\b\w{15,}\b"," ", text) # remove words <2 text = re.sub(r'\b\w{1,2}\b'," ", text) # keep only alphabets and _ text = re.sub(r"[^a-zA-Z_]"," ", text)

print(len(preprocessed_subject))

#i am living in the New York print("i am living in the New York -->", list(chunks)) print(" ") print("-"*50) print(" ") #My name is Srikanth Varma print("My name is Srikanth Varma -->", list(chunks1))

data.columns

data.iloc[400]

def preprocess(Input_Text): """Do all the Preprocessing as shown above and return a tuple contain preprocess_email,preprocess_subject,preprocess_text for that Text_data""" #preprocessed_email = [] # storing all the peprocessed emails #preprocessed_subject = [] # storing the subject #preprocessed_text = [] # storing the text #text = [] #txt_class = [] #for file in os.listdir(Input_Text): # for each file #with open('/content/textfiles/documents/'+file, "rb" ) as fp: # from stackoverflow to read the text #texts = fp.read().decode('ISO-8859-1') #text.append(texts) # extracting the name of the documents and saved in the txt_class list #if Input_Text.endswith('.txt'): #txt_class = Input_Text.split('.txt')[0].split('_')[0] #txt_class.append(filename) # regular expression to extract emails from documents emails = re.findall(r"[a-z0-9\.\-+_]+@[a-z0-9\.\-+_]+\.[a-z]+", Input_Text) splitted_emails = [] for email in emails: splitted_emails += email.split('@')[1].split('.') preprocessed_email_list = [] for word in splitted_emails: if (len(word) > 2) and (word != 'com'): preprocessed_email_list.append(word) preprocessed_email = ' '.join(preprocessed_email_list) #preprocessed_email.append(process) # removing email from text re.sub(r"[a-z0-9\.\-+_]+@[a-z0-9\.\-+_]+\.[a-z]+", ' ',Input_Text) # extracting subject sub = re.findall(r'^Subject:.*$', Input_Text , re.MULTILINE) subject = sub[0] subject = subject[9:] #remove all the punctuation for i in string.punctuation: subject = subject.replace(i, " ") # removing Re subject = re.sub(r"Re:", "", subject) preprocessed_subject = subject Input_Text = re.sub(r'Subject:.*', "", Input_Text, re.MULTILINE) # removing subject Input_Text = re.sub(r'Write to:.*',"", Input_Text, re.MULTILINE) # removing Write to: Input_Text = re.sub(r'From:.*', "", Input_Text, re.MULTILINE) # removing From: Input_Text = re.sub(r'<.*>', "", Input_Text, re.MULTILINE) # removing tags like "<anyword>" Input_Text = re.sub(r'\([^)]*\)|[\(\)]',"", Input_Text, re.MULTILINE) # removing data present in brackets Input_Text = re.sub(r'[\n\t\-\\\/]', "", Input_Text, re.MULTILINE) # removing all the newlines('\n'), tabs('\t'), "-", "\" Input_Text = re.sub(r".*:", "", Input_Text, re.MULTILINE) # removing words which ends with ":" #can't -> can not, 's -> is, i've -> i have, i'm -> i am, you're -> you are, i'll --> i will Input_Text = re.sub(r"can\'t", "can not", Input_Text) Input_Text = re.sub(r"won\'t", "will not", Input_Text) Input_Text = re.sub(r"i\'ve", "i have", Input_Text) Input_Text = re.sub(r"i\'m", "i am", Input_Text) Input_Text = re.sub(r"you\'re", "you are", Input_Text) Input_Text = re.sub(r"i\'ll", "i will", Input_Text) Input_Text = re.sub(r"n'\t", "not", Input_Text) Input_Text = re.sub(r"\'s", "is", Input_Text) Input_Text = re.sub(r"\'d", "would", Input_Text) Input_Text = re.sub(r"\'ll", "will", Input_Text) Input_Text = re.sub(r"\'t", "not", Input_Text) Input_Text = re.sub(r"\'ve", "have", Input_Text) Input_Text = re.sub(r"\'m", "am", Input_Text) # chuncking chunks=(list(ne_chunk(pos_tag(word_tokenize(Input_Text))))) for i in chunks: if type(i)==Tree: if i.label() == "GPE": j = i.leaves() # if new_york or bigger name if len(j)>1: gpe = "_".join([term for term,pos in j]) Input_Text = re.sub(rf'{j[1][0]}', gpe, Input_Text, flags=re.MULTILINE) #replacing york with new_york Input_Text = re.sub(rf'\b{j[0][0]}\b',"", Input_Text, flags=re.MULTILINE) #deleting new, \b is important # deleting Srikanth Varma if i.label()=="PERSON": for term, position in i.leaves(): Input_Text = re.sub(re.escape(term),"", Input_Text, flags=re.MULTILINE) # Removing digits Input_Text = re.sub(r'\d'," ", Input_Text, re.MULTILINE) # Replacing “_word_” , “_word” , “word_” to word #replace _word_ to word Input_Text = re.sub(r"\b_([a-zA-z]+)_\b",r"\1", Input_Text) # replace _word to word Input_Text = re.sub(r"\b_([a-zA-z]+)\b",r"\1", Input_Text) # replace word_ to word Input_Text = re.sub(r"\b([a-zA-z]+)_\b",r"\1", Input_Text) # d_berlin to berlin Input_Text = re.sub(r"\b[a-zA-Z]{1}_([a-zA-Z]+)",r"\1", Input_Text) # de_berlin to berlin Input_Text = re.sub(r"\b[a-zA-Z]{2}_([a-zA-Z]+)",r"\1", Input_Text) # lower case Input_Text = Input_Text.lower() # remove words >15 Input_Text = re.sub(r"\b\w{15,}\b"," ", Input_Text) # remove words <2 Input_Text = re.sub(r'\b\w{1,2}\b'," ", Input_Text) # keep only alphabets and _ Input_Text = re.sub(r"[^a-zA-Z_]"," ", Input_Text) # Removing extra white space from the string Input_Text = " ".join(Input_Text.split()) #preprocessed_text.append(texts) preprocessed_text = Input_Text return (preprocessed_email, preprocessed_subject, preprocessed_text)

preprocessed_email = [] # storing all the peprocessed emails preprocessed_subject = [] # storing the subject preprocessed_text = [] # storing the processed text text = [] # storing text txt_class = [] # storing text class for file in os.listdir('/content/textfiles/documents/'): with open('/content/textfiles/documents/'+file, "rb" ) as fp: if file.endswith('.txt'): file_name = file.split('.txt')[0].split('_')[0] # from stackoverflow to read the text input_text = fp.read().decode('ISO-8859-1') text.append(input_text) # appending raw text # calling preprocess function for text preprocessing pre_processed = preprocess(input_text) txt_class.append(file_name) preprocessed_email.append(pre_processed[0]) preprocessed_subject.append(pre_processed[1]) preprocessed_text.append(pre_processed[2])

with open('/content/textfiles/documents/alt.atheism_49960.txt' , "rb" ) as f: input_text = f.read().decode('ISO-8859-1') pre_process = preprocess(input_text) process_text = pre_process[2] print(process_text)

import pandas as pd import numpy as np from sklearn.preprocessing import LabelEncoder from sklearn.model_selection import train_test_split from keras.utils import np_utils from keras.preprocessing.text import Tokenizer from keras.preprocessing.sequence import pad_sequences

df = pd.DataFrame(list(zip(text, txt_class, preprocessed_email, preprocessed_subject, preprocessed_text)), columns = ['Text', 'Txt_Class', 'Preprocessed_email', 'Preprocessed_subject', 'Preprocessed_text'])

df.head(5)

df.shape

#Data = pd.DataFrame((df['Preprocessed_subject']+df['Preprocessed_email']+df['Preprocessed_text']), columns = ['processed_text']) Data = pd.DataFrame((df[['Preprocessed_subject', 'Preprocessed_email', 'Preprocessed_text']].agg(' '.join, axis = 1)), columns = ['processed_text']) Data['Txt_Class'] = df['Txt_Class'] Data

y = Data['Txt_Class'] x = Data.drop(['Txt_Class'], axis=1)

print(x.shape) print(y.shape)

# encode class values as integers label_encoder = LabelEncoder() label_encoder.fit(y) label_encoded_y = label_encoder.transform(y) # convert integers to dummy variables (i.e. one hot encoded) encoded_y = np_utils.to_categorical(label_encoded_y) encoded_y

text_train, text_test, y_train, y_test = train_test_split(x, encoded_y, test_size = 0.25, stratify = y)

print('Train Data :', text_train.shape, y_train.shape) print('Test Data :', text_test.shape, y_test.shape)

sentence_train = [] for item in text_train.values.tolist(): for i in item: sentence_train.append(i)

sentence_test = [] for item in text_test.values.tolist(): for i in item: sentence_test.append(i)

tokenizer = Tokenizer(num_words = 20000, filters='!"#$%&()*+,-./:;<=>?@[\\]^`{|}~\t\n') tokenizer.fit_on_texts(sentence_train) X_train = tokenizer.texts_to_sequences(sentence_train) X_test = tokenizer.texts_to_sequences(sentence_test)

# Adding 1 because of reserved 0 index for padding vocab_size = len(tokenizer.word_index) + 1 print('Found %s unique tokens.' % vocab_size)

x_train = pad_sequences(X_train, padding = 'post', maxlen = 1000) x_test = pad_sequences(X_test, padding = 'post', maxlen = 1000)

import pickle with open('/content/drive/My Drive/DT Data set/glove_vectors', 'rb') as f: model = pickle.load(f) gloves_words = set(model.keys())

#https://machinelearningmastery.com/use-word-embedding-layers-deep-learning-keras/ # create a weight matrix for words in training docs embedding_matrix = np.zeros((vocab_size, 300)) for word, i in tokenizer.word_index.items(): embedding_vector = model.get(word) if embedding_vector is not None: embedding_matrix[i] = embedding_vector

from tensorflow.keras.models import Sequential from tensorflow.keras import layers from tensorflow import keras import tensorflow as tf from tensorflow.keras.callbacks import EarlyStopping from tensorflow.keras.callbacks import ModelCheckpoint import datetime from tensorflow.keras.utils import plot_model

import numpy as np from sklearn.metrics import confusion_matrix, f1_score class f1Score(tf.keras.callbacks.Callback): def __init__(self,training_data,validation_data): self.x_train = training_data[0] self.y_train = training_data[1] self.x_val = validation_data[0] self.y_val = validation_data[1] def on_train_begin(self, logs={}): self.val_f1s = [] def on_epoch_end(self, epoch, logs={}): val_predict = (np.asarray(self.model.predict(self.x_val))).round() val_targ = self.y_val val_micro_f1 = f1_score(val_targ, val_predict, average = 'micro') self.val_f1s.append(val_micro_f1) print(" val_micro_f1: %f " %(val_micro_f1)) return

# define model #https://blog.keras.io/using-pre-trained-word-embeddings-in-a-keras-model.html embedding_layer = layers.Embedding(vocab_size, 300, input_length = 1000, weights=[embedding_matrix], trainable=False) sequence_input = keras.Input(shape=(1000,)) embedded_sequences = embedding_layer(sequence_input) conv1d_with_filter_size_m = layers.Conv1D(16, 5, activation='relu')(embedded_sequences) conv1d_with_filter_size_n = layers.Conv1D(16, 5, activation='relu')(embedded_sequences) conv1d_with_filter_size_o = layers.Conv1D(16, 5, activation='relu')(embedded_sequences) concatenate1 = layers.Concatenate()([conv1d_with_filter_size_m, conv1d_with_filter_size_n, conv1d_with_filter_size_o]) MaxPoolLayer1 = layers.MaxPooling1D(5)(concatenate1) conv1d_with_filter_size_i = layers.Conv1D(15, 3, activation='relu')(MaxPoolLayer1) conv1d_with_filter_size_j = layers.Conv1D(15, 3, activation='relu')(MaxPoolLayer1) conv1d_with_filter_size_k = layers.Conv1D(15, 3, activation='relu')(MaxPoolLayer1) concatenate2 = layers.Concatenate()([conv1d_with_filter_size_i, conv1d_with_filter_size_j, conv1d_with_filter_size_k]) MaxPoolLayer2 = layers.MaxPooling1D(5)(concatenate2) conv1d_with_filter_size_p = layers.Conv1D(16, 3, activation='relu')(MaxPoolLayer2) flatten = layers.Flatten()(conv1d_with_filter_size_p) dropout = layers.Dropout(0.30)(flatten) layer1 = layers.Dense(64, activation='relu')(dropout) layer2 = layers.Dense(32, activation='relu')(layer1) output = layers.Dense(20, activation='softmax')(layer2) model1 = keras.Model(inputs = sequence_input, outputs = output) model1.summary()

plot_model(model1, to_file='model1.png', show_shapes=True)

%load_ext tensorboard

# Clear any logs from previous runs !rm -rf ./logs/

micro_f1_score_callback = f1Score(training_data=(x_train, y_train), validation_data=(x_test, y_test)) #file path, it saves the model in the 'model_save' folder and we are naming model with epoch number #and val acc to differtiate with other models #you have to create model_save folder before running the code. filepath="model_save/weights-{epoch:02d}-{val_accuracy:.4f}.hdf5" checkpoint = ModelCheckpoint(filepath=filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='auto') earlystop = EarlyStopping(monitor='val_loss', patience=3, verbose=1) optimizer = tf.keras.optimizers.Adam(learning_rate = 0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-07) log_dir = "logs/fit" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S") tensorboard_callbacks = tf.keras.callbacks.TensorBoard(log_dir = log_dir, histogram_freq = 1, write_graph = True, write_grads = True) model1.compile(optimizer = optimizer, loss = 'categorical_crossentropy', metrics=['accuracy']) callbacks_list = [micro_f1_score_callback, checkpoint, earlystop, tensorboard_callbacks] model1.fit(x_train, y_train, epochs=25, validation_data=(x_test,y_test), batch_size=16, callbacks=callbacks_list)

%tensorboard --logdir logs/fit20210610-095052/

tokenizer2 = Tokenizer(filters='!"#$%&()*+,-./:;<=>?@[\\]^`{|}~\t\n', char_level=True) tokenizer2.fit_on_texts(sentence_train) X_train2 = tokenizer2.texts_to_sequences(sentence_train) X_test2 = tokenizer2.texts_to_sequences(sentence_test)

# Adding 1 because of reserved 0 index for padding vocab_size2 = len(tokenizer2.word_index) + 1 print('Found %s unique tokens.' % vocab_size2)

x_train2 = pad_sequences(X_train2, padding = 'post', maxlen = 1500) x_test2 = pad_sequences(X_test2, padding = 'post', maxlen = 1500)

char_glove = {} with open('/content/drive/My Drive/AppliedAi/glove.840B.300d-char.txt') as f: for line in f: word, *vector = line.split() char_glove[word] = np.array(vector, dtype='float32') f.close()

char_embedding_matrix = np.zeros((vocab_size2, 300)) for char, i in tokenizer2.word_index.items(): embedding_vector = char_glove.get(char) if embedding_vector is not None: char_embedding_matrix[i] = embedding_vector

print(x_train2.shape) print(x_test2.shape)

Embedding_Layer_to_get_char_embedded = layers.Embedding(vocab_size2, 300, input_length=300, weights = [char_embedding_matrix], trainable = True) input = keras.Input(shape=(1500,)) embedding_sequence = Embedding_Layer_to_get_char_embedded(input) conv1d_with_filter_size_n = layers.Conv1D(20, 7, activation='relu')(embedding_sequence) conv1d_with_filter_size_m = layers.Conv1D(20, 7, activation='relu')(conv1d_with_filter_size_n) MaxPoolLayer_1 = layers.MaxPooling1D(3)(conv1d_with_filter_size_m) conv1d_with_filter_size_k = layers.Conv1D(20, 3, activation='relu')(MaxPoolLayer_1) conv1d_with_filter_size_t = layers.Conv1D(20, 3, activation='relu')(conv1d_with_filter_size_k) MaxPoolLayer_2 = layers.MaxPooling1D(3)(conv1d_with_filter_size_t) flatten = layers.Flatten()(MaxPoolLayer_2) dropout = layers.Dropout(0.5)(flatten) dense1 = layers.Dense(256, activation='relu')(dropout) output = layers.Dense(20, activation='softmax')(dense1) model2 = keras.Model( inputs = input, outputs = output) model2.summary()

plot_model( model=model2, to_file = 'model2.png', show_shapes= True)

%load_ext tensorboard

#clear logs from previous run !rm -rf ./logs/

micro_f1_score_callback = f1Score( training_data = (x_train2, y_train), validation_data=(x_test2, y_test)) #file path, it saves the model in the 'model2_save' folder and we are naming model with epoch number #and val acc to differtiate with other models #you have to create model_save folder before running the code. filepath="model2_save/weights-{epoch:02d}-{val_accuracy:.4f}.hdf5" checkpoint = ModelCheckpoint(filepath=filepath, monitor='val_loss', verbose=1, save_best_only=True, mode='auto') earlystoping = EarlyStopping(monitor='val_loss', patience=3, verbose=1) optimizer = tf.keras.optimizers.Adam(learning_rate = 0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-07) log_dir = "logs/fit" + datetime.datetime.now().strftime("%Y%m%d-%H%M%S") tensorboard_callbacks = tf.keras.callbacks.TensorBoard(log_dir = log_dir, histogram_freq = 1, write_graph = True, write_grads = True) model2.compile(optimizer = optimizer, loss = 'categorical_crossentropy', metrics=['accuracy']) callbacks_list = [micro_f1_score_callback, checkpoint, earlystoping, tensorboard_callbacks] model2.fit(x_train2, y_train, epochs=25, validation_data=(x_test2,y_test), batch_size=16, callbacks=callbacks_list)

from prettytable import PrettyTable x = PrettyTable() x.field_names = ["Model", "Train Accuracy", "Test Accuracy"] x.add_row(["Model1", 0.8043, 0.7085]) x.add_row(["Model2", 0.3479, 0.1372]) print(x)