2daysmodel.py

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

import os
import PIL
import glob
import cv2
import numpy as np
import pathlib
import matplotlib.pyplot as plt


import tensorflow as tf  # version 2.5
from tensorflow import keras
from tensorflow.keras.layers import Softmax
from tensorflow.keras.layers import Dense, Flatten, Dropout, Activation
from tensorflow.keras.models import Sequential
from tensorflow.keras.optimizers import Adam

dir = '/content/drive/MyDrive/2dayprediction'
dataset = pathlib.Path(dir)

positives = (list(dataset.glob('2daynegative/*')))
negatives = (list(dataset.glob('2daypositive/*')))

"""Training and Testing + some preprocessing"""

# Splitting the dataset into training and testing and doing some preproessing
#training part 

height = 512
width = 512
batch_size = 4
learning_rate_base = 0.001
epochs = 25

#training and validation split, 80% is training, 20% for validation
#first subset is training for training
#label mode is categorical because this requires category as in positive or negative
#equal all the images resized to this dimnention

train_dataset = tf.keras.preprocessing.image_dataset_from_directory(dataset,validation_split = 0.2, subset = "training", seed = 123, label_mode = 'categorical', image_size = (height, width), batch_size = batch_size)

#code for the validation part

#same code, with diff subset 
#subset is validation 

val_dataset = tf.keras.preprocessing.image_dataset_from_directory(dataset, validation_split = 0.2,  subset = "validation", seed = 123, label_mode = "categorical", image_size = (height, width), batch_size = batch_size )

#printing the different classes inside the datset

class_names = train_dataset.class_names
print(class_names)

#names of all the folders in the dataset

#getting the model and tuning it

resnet50_structure = Sequential() #creating an empty neural network model, initilizing this by calling sequential function.
#this makes sure that layers added will be added in a sequence

#calling resnet50

#top = false means that i am going to use my own input and output layer, why? bc the original resnset model might have been trained in diff image dimention, unlike mine.

model_pretrained = tf.keras.applications.ResNet50(include_top = False, input_shape = (height,width,3), pooling = 'avg', classes = 2, weights = 'imagenet') #using imagenet for it's weight 

#we can obtain this code from keras application documentation, under ResNet50

for each_layer in model_pretrained.layers: 
  each_layer.trainable = False #for every layer inside that pretrained model is false bc the resnset50 model has learned do not learn the weights again. 

resnet50_structure.add(model_pretrained) #adding the pretrained model to the resnet model that we just created
resnet50_structure.add(Flatten())#flattening whatever the model gave into one dimention 
resnet50_structure.add(Dense(512,activation='relu')) #using fully connected layer here at the end
#512 neurons for learning
#The final output will have 2 neurons for the 2 output
resnet50_structure.add(Dense(2, activation = 'softmax'))

resnet50_structure.summary()

#when the model is ready, now I have to compile it. I provider an optimizer of adam. 


resnet50_structure.compile(optimizer = Adam(lr=learning_rate_base), loss = 'categorical_crossentropy', metrics = ['accuracy'])

#now I have to fit the model
#provide the training and val data seta, and specify the # of epocs
#monitoring the accuracy, validation and loss.

history = resnet50_structure.fit(train_dataset, validation_data = val_dataset, epochs = epochs)

fig1 = plt.gcf()
plt.plot(history.history['accuracy']) #store the models' logs in a history variable. 
plt.plot(history.history['val_accuracy'])
plt.axis(ymin=0.4, ymax=1)
plt.grid() # labeling
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epochs')
plt.legend(['Train', 'Validation'])
plt.show()

plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.grid()
plt.title('Model Loss')
plt.xlabel('loss')
plt.ylabel('Epochs')
plt.legend(['train', 'validation'])
plt.show()

#Model is ready

import random

import cv2 #used to read images
image = cv2.imread(str(negatives[0])) 
resize_img = cv2.resize(image, (height, width)) #resizing it to 512x512
image = np.expand_dims(resize_img, axis = 0) #expanding the dimention of the image to 4D to input into the CNN
#print(image.shape)

pred = resnet50_structure.predict(image) #passing image to model to test
output_Class = class_names[np.argmax(pred)]

print("The predicted classification for this image is: ",  output_Class)