Reading data
","metadata":{}},{"cell_type":"code","source":"\nif ('train' not in os.listdir('/kaggle/working')):\n print ('extraction...')\n with zipfile.ZipFile('/kaggle/input/dogs-vs-cats/train.zip', 'r') as ztrain:\n ztrain.extractall('/kaggle/working')\nif ('test1' not in os.listdir('/kaggle/working')):\n print ('extraction...')\n with zipfile.ZipFile('/kaggle/input/dogs-vs-cats/test1.zip', 'r') as ztest:\n ztest.extractall('/kaggle/working')\n \n","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:15:55.386681Z","iopub.execute_input":"2023-11-14T12:15:55.387714Z","iopub.status.idle":"2023-11-14T12:16:09.747690Z","shell.execute_reply.started":"2023-11-14T12:15:55.387675Z","shell.execute_reply":"2023-11-14T12:16:09.746862Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#constructing dataframe of pictures' paths and their labels in traint folder \nweather={'sunny':0 ,'snowy':1 ,'rainy':2, 'foggy':3}\ncategory={'dog':0 , 'cat':1}\ndef images_in_folder(folder ,weather_stat): \n data=pd.DataFrame({'path': [], 'root':[], 'picture_name':[] , 'label': [], 'stat':[]}) \n data = data.set_index('path')\n root = folder\n for picture_name in os.listdir(folder):\n if picture_name.split('.')[0] =='dog':\n data.loc[folder+picture_name]= [root,picture_name ,0 , weather[weather_stat]]\n elif picture_name.split('.')[0] =='cat':\n data.loc[folder+picture_name]= [root,picture_name ,1 , weather[weather_stat]]\n else:\n data.loc[folder+picture_name]= [root,picture_name ,None , weather[weather_stat]] \n return(data)\n\ndata=images_in_folder('/kaggle/working/train/' ,weather_stat = 'sunny')\ntest_data=images_in_folder('/kaggle/working/test1/' , weather_stat = 'sunny')\ndata.head()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T21:09:31.426523Z","iopub.execute_input":"2023-11-14T21:09:31.426909Z","iopub.status.idle":"2023-11-14T21:12:32.468770Z","shell.execute_reply.started":"2023-11-14T21:09:31.426877Z","shell.execute_reply":"2023-11-14T21:12:32.467133Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_data.head()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T21:13:30.882512Z","iopub.execute_input":"2023-11-14T21:13:30.882945Z","iopub.status.idle":"2023-11-14T21:13:30.898472Z","shell.execute_reply.started":"2023-11-14T21:13:30.882911Z","shell.execute_reply":"2023-11-14T21:13:30.896439Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fig, axes = plt.subplots(1, 5, figsize=(12, 3))\nfor i, path in enumerate(data.head(5).index):\n image = Image.open(path)\n axes[i].imshow(image)\n axes[i].axis('off')\n axes[i].set_title(data.loc[path].label)\nplt.tight_layout()\n\nfig, axes = plt.subplots(1, 5, figsize=(12, 3))\nfor i, path in enumerate(data.tail(5).index):\n image = Image.open(path)\n axes[i].imshow(image)\n axes[i].axis('off')\n axes[i].set_title(data.loc[path].label) \nplt.tight_layout()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:19:07.149442Z","iopub.execute_input":"2023-11-14T12:19:07.149845Z","iopub.status.idle":"2023-11-14T12:19:08.464181Z","shell.execute_reply.started":"2023-11-14T12:19:07.149816Z","shell.execute_reply":"2023-11-14T12:19:08.462879Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data.label.value_counts().plot(kind='bar')","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:19:08.465314Z","iopub.execute_input":"2023-11-14T12:19:08.466473Z","iopub.status.idle":"2023-11-14T12:19:08.657400Z","shell.execute_reply.started":"2023-11-14T12:19:08.466437Z","shell.execute_reply":"2023-11-14T12:19:08.656237Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Check picture Sizes
","metadata":{}},{"cell_type":"code","source":"def get_dims(file): \n image = cv2.imread(file) \n h, w, d = image.shape\n return [h, w, d] \n\nfor path in data.index: \n data.loc[path ,['h','w', 'd'] ]= get_dims(path)","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:19:08.658827Z","iopub.execute_input":"2023-11-14T12:19:08.659475Z","iopub.status.idle":"2023-11-14T12:20:23.178638Z","shell.execute_reply.started":"2023-11-14T12:19:08.659438Z","shell.execute_reply":"2023-11-14T12:20:23.176861Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:20:23.180007Z","iopub.execute_input":"2023-11-14T12:20:23.180347Z","iopub.status.idle":"2023-11-14T12:20:23.201109Z","shell.execute_reply.started":"2023-11-14T12:20:23.180316Z","shell.execute_reply":"2023-11-14T12:20:23.199642Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fig = px.scatter(data, x='h', y='w', color='label', \n labels={'Height':'h', 'Width':'w'},\n title=\"Scatter plot of Image Dimensions by Class\")\n\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:20:23.202367Z","iopub.execute_input":"2023-11-14T12:20:23.202669Z","iopub.status.idle":"2023-11-14T12:20:25.452809Z","shell.execute_reply.started":"2023-11-14T12:20:23.202642Z","shell.execute_reply":"2023-11-14T12:20:25.451357Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"TARGET_SIZE=(224,224)\ndata = data[(data.h>=200)&(data.h>=200)]\nfor path in data[(data.h<200)&(data.h<200) ].index:\n os.remove(path)","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:20:25.457062Z","iopub.execute_input":"2023-11-14T12:20:25.457801Z","iopub.status.idle":"2023-11-14T12:20:25.469819Z","shell.execute_reply.started":"2023-11-14T12:20:25.457765Z","shell.execute_reply":"2023-11-14T12:20:25.468678Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Check contrast
","metadata":{"execution":{"iopub.status.busy":"2023-11-04T15:42:01.160161Z","iopub.execute_input":"2023-11-04T15:42:01.160471Z","iopub.status.idle":"2023-11-04T15:42:01.165833Z","shell.execute_reply.started":"2023-11-04T15:42:01.160450Z","shell.execute_reply":"2023-11-04T15:42:01.164770Z"}}},{"cell_type":"code","source":"def compute_mean_histogram(paths):\n histograms = {'red': [], 'green': [], 'blue': []}\n for path in paths:\n image = cv2.imread(path) # Load image in color mode\n for i, color in enumerate(('blue', 'green', 'red')):\n hist = cv2.calcHist([image], [i], None, [256], [0, 256])\n histograms[color].append(hist)\n mean_histograms = {color: np.mean(hists, axis=0) for color, hists in histograms.items()}\n return mean_histograms\n\nmean_histograms_per_class = {}\nmean_histograms_per_class[0] = compute_mean_histogram(list(data.loc[data['label']==0].index ))\nmean_histograms_per_class[1] = compute_mean_histogram(list(data.loc[data['label']==1].index ))\n\n#Plotting the histogram \nfig = go.Figure()\nfor class_label, mean_histograms in mean_histograms_per_class.items():\n for color, mean_histogram in mean_histograms.items():\n fig.add_trace(go.Scatter(x=np.arange(256), y=mean_histogram.flatten(), mode='lines', name=f'{class_label} - {color}'))\n\nfig.update_layout(\n title='Mean Histograms per Class per Channel',\n xaxis_title='Pixel Value',\n yaxis_title='Frequency',\n legend_title='Class - Channel'\n)\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:20:25.471004Z","iopub.execute_input":"2023-11-14T12:20:25.471895Z","iopub.status.idle":"2023-11-14T12:21:25.023587Z","shell.execute_reply.started":"2023-11-14T12:20:25.471869Z","shell.execute_reply":"2023-11-14T12:21:25.021711Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Check Standard Deviation And Noise Level
","metadata":{"execution":{"iopub.status.busy":"2023-11-04T15:42:51.387273Z","iopub.execute_input":"2023-11-04T15:42:51.387555Z","iopub.status.idle":"2023-11-04T15:42:51.392967Z","shell.execute_reply.started":"2023-11-04T15:42:51.387535Z","shell.execute_reply":"2023-11-04T15:42:51.391655Z"}}},{"cell_type":"code","source":"def compute_avg_std(image_paths):\n stds = {'red': [], 'green': [], 'blue': []}\n for image_path in image_paths:\n image = cv2.imread(image_path) # Load image in color mode\n for i, color in enumerate(('blue', 'green', 'red')):\n std = np.std(image[:,:,i])\n stds[color].append(std)\n avg_stds = {color: np.mean(values) for color, values in stds.items()}\n return avg_stds\n\n\navg_stds_per_class = {}\navg_stds_per_class [0] = compute_avg_std(list(data.loc[data['label']==0 ].index ))\navg_stds_per_class [1] = compute_avg_std(list(data.loc[data['label']==1].index ))\n ","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:21:25.025408Z","iopub.execute_input":"2023-11-14T12:21:25.026443Z","iopub.status.idle":"2023-11-14T12:22:37.627074Z","shell.execute_reply.started":"2023-11-14T12:21:25.026398Z","shell.execute_reply":"2023-11-14T12:22:37.625675Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Plotting\ncategories = list(avg_stds_per_class.keys())\nfig = go.Figure()\nfor color in ['red', 'green', 'blue']:\n values = [avg_stds_per_class[class_label][color] for class_label in categories]\n fig.add_trace(go.Bar(x=categories, y=values, name=f'{color} channel'))\n\nfig.update_layout(\n title='Average Standard Deviation of Pixel Values per Class per Channel',\n xaxis_title='Class',\n yaxis_title='Average Standard Deviation',\n legend_title='Channel'\n)\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:22:37.628371Z","iopub.execute_input":"2023-11-14T12:22:37.628690Z","iopub.status.idle":"2023-11-14T12:22:37.652849Z","shell.execute_reply.started":"2023-11-14T12:22:37.628661Z","shell.execute_reply":"2023-11-14T12:22:37.651358Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def estimate_noise(image_paths):\n noise_levels = []\n for image_path in image_paths:\n image = cv2.imread(image_path, cv2.IMREAD_GRAYSCALE) # Load image in grayscale mode\n laplacian = cv2.Laplacian(image, cv2.CV_64F)\n std_dev = np.std(laplacian)\n noise_levels.append(std_dev)\n mean_noise_level = np.mean(noise_levels)\n return mean_noise_level\n \nmean_noise_levels = {}\nmean_noise_levels [0] = estimate_noise(list(data.loc[data['label']==0 ].index ))\nmean_noise_levels [1] = estimate_noise(list(data.loc[data['label']==1 ].index ))\n ","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:22:37.654960Z","iopub.execute_input":"2023-11-14T12:22:37.655322Z","iopub.status.idle":"2023-11-14T12:23:13.646966Z","shell.execute_reply.started":"2023-11-14T12:22:37.655290Z","shell.execute_reply":"2023-11-14T12:23:13.645356Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Plotting\nfig = go.Figure()\nclass_labels = list(mean_noise_levels.keys())\nnoise_levels = list(mean_noise_levels.values())\nfig.add_trace(go.Bar(x=class_labels, y=noise_levels))\nfig.update_layout(\n title='Estimated Noise Levels per Class',\n xaxis_title='Class',\n yaxis_title='Estimated Noise Level',\n legend_title='Class'\n)\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:23:13.648472Z","iopub.execute_input":"2023-11-14T12:23:13.648797Z","iopub.status.idle":"2023-11-14T12:23:13.666661Z","shell.execute_reply.started":"2023-11-14T12:23:13.648771Z","shell.execute_reply":"2023-11-14T12:23:13.664902Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"PreProcess and Data Augmentation
","metadata":{}},{"cell_type":"markdown","source":"Data Augmentation
","metadata":{}},{"cell_type":"code","source":"\ndef produce_save_augmented_filters(image_paths , plot_some = True ): \n for count, path in enumerate(image_paths):\n #Snow\n image = cv2.imread(path)\n image_name = path.rsplit('/', 1)[-1]\n lbl =data.loc[path , 'label']\n aug = iaa.FastSnowyLandscape(lightness_threshold=140,lightness_multiplier=2.5)\n snowy_image = aug.augment_image(image)\n snow_path = '/kaggle/working/train/1.'+image_name \n data.loc[snow_path ,'label'] = lbl\n data.loc[snow_path , 'picture_name']= \"1.\"+image_name\n data.loc[snow_path ,'stat'] = 1 # stat=1 means it is snowy\n cv2.imwrite(snow_path, snowy_image) \n # Rain\n aug = aug = iaa.Rain(speed=(0.1, 0.3))\n rainy_image = aug.augment_image(image)\n rain_path = '/kaggle/working/train/2.'+ image_name\n data.loc[rain_path,'label'] = lbl\n data.loc[rain_path , 'picture_name']= \"2.\"+image_name\n data.loc[rain_path ,'stat'] = 2 # stat=2 means it is rainy \n cv2.imwrite(rain_path, rainy_image)\n #fog\n aug = iaa.Fog()\n foggy_image = aug.augment_image(image)\n fog_path = '/kaggle/working/train/3.' +image_name\n data.loc[fog_path,'label'] = lbl\n data.loc[fog_path , 'picture_name']= \"3.\"+image_name\n data.loc[fog_path ,'stat'] = 3 # stat=3 means it is foggy\n cv2.imwrite(fog_path, foggy_image)\n #Plot some augmented images\n if plot_some == True: \n if count<5:\n count += 1\n fig, ax = plt.subplots(1,4, figsize=(12, 3))\n ax[0].imshow(image)\n ax[0].set_title(\"Intact\")\n ax[0].axis('off')\n ax[1].imshow(snowy_image)\n ax[1].set_title(\"Snowy \")\n ax[1].axis('off')\n ax[2].imshow(rainy_image)\n ax[2].set_title(\"Rainy \")\n ax[2].axis('off')\n ax[3].imshow(foggy_image)\n ax[3].set_title(\"Foggy \")\n ax[3].axis('off')\n \n ","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:23:13.668510Z","iopub.execute_input":"2023-11-14T12:23:13.668885Z","iopub.status.idle":"2023-11-14T12:23:13.682665Z","shell.execute_reply.started":"2023-11-14T12:23:13.668854Z","shell.execute_reply":"2023-11-14T12:23:13.680963Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"produce_save_augmented_filters(random.choices(list(data.index) ,k=len(data)//100),\n plot_some = True)\n ","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:23:13.684898Z","iopub.execute_input":"2023-11-14T12:23:13.685624Z","iopub.status.idle":"2023-11-14T12:23:36.372688Z","shell.execute_reply.started":"2023-11-14T12:23:13.685588Z","shell.execute_reply":"2023-11-14T12:23:36.370928Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Size Correction and Contrast Correction Using Clahe
","metadata":{}},{"cell_type":"code","source":"from pathlib import Path\n \ndef save_clahe_img(data , plot_some = True): \n save_folder = '/kaggle/working/train/'\n \n image_paths=list(data.index)\n for count, image_path in enumerate(image_paths ):\n image = cv2.imread(image_path)\n image = cv2.resize(image,TARGET_SIZE) \n lab = cv2.cvtColor(image, cv2.COLOR_BGR2LAB)\n # Split Lab\n lab_planes = list(cv2.split(lab)) \n # Apply Histogram Equlization\n clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))\n lab_planes[0] = clahe.apply(lab_planes[0])\n lab = cv2.merge(lab_planes)\n cla = cv2.cvtColor(lab, cv2.COLOR_LAB2BGR) \n cv2.imwrite(image_path, cla)\n #plot first 5 images with their corrected images for instance\n if plot_some == True: \n if count<5:\n count += 1\n fig, ax = plt.subplots(1,2, figsize=(12, 3))\n ax[0].imshow(image)\n ax[0].set_title(\"Intact\")\n ax[0].axis('off')\n ax[1].imshow(cla)\n ax[1].set_title(\"Corrected contrast & resized\") \n ax[1].axis('off') \n \nsave_clahe_img(data , plot_some = True)\ndata.to_csv('/kaggle/working/data_df.csv')","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:23:36.374633Z","iopub.execute_input":"2023-11-14T12:23:36.375464Z","iopub.status.idle":"2023-11-14T12:25:24.167604Z","shell.execute_reply.started":"2023-11-14T12:23:36.375437Z","shell.execute_reply":"2023-11-14T12:25:24.166120Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Using Augmented and PreProcessed data
","metadata":{}},{"cell_type":"markdown","source":"Train, Validation and Test Data
","metadata":{}},{"cell_type":"code","source":"data.head()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:25:24.168966Z","iopub.execute_input":"2023-11-14T12:25:24.169330Z","iopub.status.idle":"2023-11-14T12:25:24.185442Z","shell.execute_reply.started":"2023-11-14T12:25:24.169296Z","shell.execute_reply":"2023-11-14T12:25:24.184075Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data=data.reset_index()\ntest_data=test_data.reset_index()\ndata.label=data.label.astype(str)\ndata.head()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T21:14:59.431998Z","iopub.execute_input":"2023-11-14T21:14:59.432364Z","iopub.status.idle":"2023-11-14T21:14:59.455556Z","shell.execute_reply.started":"2023-11-14T21:14:59.432337Z","shell.execute_reply":"2023-11-14T21:14:59.454108Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_data, val_data = train_test_split(data[['path' , 'label']], test_size=0.2, random_state=14) \ntrain_data.label.value_counts().plot(kind='bar')","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:25:24.222457Z","iopub.execute_input":"2023-11-14T12:25:24.222753Z","iopub.status.idle":"2023-11-14T12:25:24.414043Z","shell.execute_reply.started":"2023-11-14T12:25:24.222728Z","shell.execute_reply":"2023-11-14T12:25:24.412409Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"batch_size = 16\nsize=224\ntrain_datagen = ImageDataGenerator(rescale=1/255)\ntrain_generator = train_datagen.flow_from_dataframe(\n dataframe = train_data,\n x_col='path',\n y_col='label', \n batch_size=batch_size,\n class_mode=\"categorical\"\n )\nvalid_datagen = ImageDataGenerator(rescale=1/255)\nvalid_generator = train_datagen.flow_from_dataframe(\n dataframe=val_data,\n x_col='path',\n y_col='label', \n batch_size=batch_size,\n class_mode=\"categorical\"\n )\ntest_datagen = ImageDataGenerator(rescale=1./255)\ntest_generator = test_datagen.flow_from_dataframe(\n dataframe = test_data, \n x_col='path',\n y_col=None,\n class_mode=None,\n batch_size=batch_size,\n shuffle=False)","metadata":{"execution":{"iopub.status.busy":"2023-11-14T21:15:16.885384Z","iopub.execute_input":"2023-11-14T21:15:16.885745Z","iopub.status.idle":"2023-11-14T21:15:17.182829Z","shell.execute_reply.started":"2023-11-14T21:15:16.885714Z","shell.execute_reply":"2023-11-14T21:15:17.181056Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Transfer learning using VGG16 model
","metadata":{}},{"cell_type":"code","source":"from keras.models import Sequential\nfrom keras import layers\nfrom keras.layers import Conv2D, MaxPooling2D, Dropout, Flatten, Dense, Activation,GlobalMaxPooling2D\nfrom keras import applications\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras import optimizers\nfrom keras.applications import VGG16\nfrom keras.models import Model\n \ninput_shape = (224, 224, 3)\n\nvgg16_epochs = 5\nbatch_size = 16\n\npre_trained_model = VGG16(input_shape=input_shape, include_top=False, weights=\"imagenet\")\n \nfor layer in pre_trained_model.layers[:-1]:\n layer.trainable = False\npre_trained_model.layers[-1].trainable = True\n \nlast_layer = pre_trained_model.get_layer('block5_pool')\nlast_output = last_layer.output\n \n \nx = GlobalMaxPooling2D()(last_output) \nx = Dense(512, activation='relu')(x) \nx = Dropout(0.5)(x)\nx = Dense(128, activation='relu')(x) \nx = Dropout(0.5)(x) \nx = layers.Dense(2, activation='softmax')(x)\n\nmodel = Model(pre_trained_model.input, x)\n\nmodel.compile(loss='categorical_crossentropy',\n optimizer='adam',\n #optimizer=optimizers.SGD(lr=1e-4, momentum=0.9),\n metrics=['accuracy'])\n\nmodel.summary()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:25:24.708768Z","iopub.execute_input":"2023-11-14T12:25:24.709347Z","iopub.status.idle":"2023-11-14T12:25:25.889740Z","shell.execute_reply.started":"2023-11-14T12:25:24.709317Z","shell.execute_reply":"2023-11-14T12:25:25.888355Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plot_model(model)","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:25:25.891165Z","iopub.execute_input":"2023-11-14T12:25:25.891596Z","iopub.status.idle":"2023-11-14T12:25:26.226801Z","shell.execute_reply.started":"2023-11-14T12:25:25.891561Z","shell.execute_reply":"2023-11-14T12:25:26.226119Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.callbacks import ModelCheckpoint , EarlyStopping\nmch = ModelCheckpoint('model.h5', monitor='val_loss', mode='min', save_best_only=True)\nes = EarlyStopping(monitor='val_loss', mode='min', patience=3)","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:25:26.227774Z","iopub.execute_input":"2023-11-14T12:25:26.228416Z","iopub.status.idle":"2023-11-14T12:25:26.233723Z","shell.execute_reply.started":"2023-11-14T12:25:26.228389Z","shell.execute_reply":"2023-11-14T12:25:26.232756Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%time\nhistory = model.fit(\n train_generator,\n epochs=4,\n steps_per_epoch = train_generator.samples // batch_size,\n validation_data = valid_generator, \n validation_steps = valid_generator.samples // batch_size, \n callbacks=[es, mch]\n)\n ","metadata":{"execution":{"iopub.status.busy":"2023-11-14T12:25:26.235202Z","iopub.execute_input":"2023-11-14T12:25:26.235680Z","iopub.status.idle":"2023-11-14T19:09:17.679762Z","shell.execute_reply.started":"2023-11-14T12:25:26.235653Z","shell.execute_reply":"2023-11-14T19:09:17.678230Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.save('/kaggle/working/')","metadata":{"execution":{"iopub.status.busy":"2023-11-14T19:09:21.575112Z","iopub.execute_input":"2023-11-14T19:09:21.575415Z","iopub.status.idle":"2023-11-14T19:09:24.437752Z","shell.execute_reply.started":"2023-11-14T19:09:21.575389Z","shell.execute_reply":"2023-11-14T19:09:24.436316Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Model Evaluation
","metadata":{}},{"cell_type":"code","source":"def plot_model_history(model_history, param1='accuracy', param2='val_accuracy'):\n fig, axs = plt.subplots(1,2,figsize=(15,5))\n axs[0].plot(range(1,len(model_history.history[param1])+1),model_history.history[param1])\n axs[0].plot(range(1,len(model_history.history[param2])+1),model_history.history[param2])\n axs[0].set_title('Accuracy with Feature extractor VGG')\n axs[0].set_xlabel('Epoch') \n axs[0].legend(['train', 'val'], loc='best')\n axs[1].plot(range(1,len(model_history.history['loss'])+1),model_history.history['loss'])\n axs[1].plot(range(1,len(model_history.history['val_loss'])+1),model_history.history['val_loss'])\n axs[1].set_title('Model Loss with Feature extractor VGG') \n axs[1].set_xlabel('Epoch') \n axs[1].legend(['train', 'val'], loc='best')\n plt.show()\n \nplot_model_history(history, param1='accuracy', param2='val_accuracy')","metadata":{"execution":{"iopub.status.busy":"2023-11-14T19:09:24.441150Z","iopub.execute_input":"2023-11-14T19:09:24.441560Z","iopub.status.idle":"2023-11-14T19:09:24.933190Z","shell.execute_reply.started":"2023-11-14T19:09:24.441525Z","shell.execute_reply":"2023-11-14T19:09:24.931887Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"validation set prediction
","metadata":{}},{"cell_type":"code","source":"y_val = pd.to_numeric(val_data['label'] ,downcast=\"integer\")\ny_pred = model.predict(valid_generator) \ny_pred_val = np.where(y_pred < 0.5, 1,0)\n","metadata":{"execution":{"iopub.status.busy":"2023-11-14T23:31:58.670867Z","iopub.execute_input":"2023-11-14T23:31:58.672106Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\n\nconfusion_mtx = confusion_matrix(y_val, y_pred[:,1]) \nf,ax = plt.subplots(figsize=(8, 8))\nsns.heatmap(confusion_mtx, annot=True,ax=ax)\nplt.xlabel(\"Predicted label\")\nplt.ylabel(\"Real label\")\nplt.title(\"Confusion Matrixwith Feature extractor VGG\")\nplt.show()\n\nprint(classification_report(y_val, y_pred))\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Test set prediction
","metadata":{}},{"cell_type":"code","source":"y_pred_test = model.predict(test_generator) ","metadata":{"execution":{"iopub.status.busy":"2023-11-14T21:31:37.296852Z","iopub.execute_input":"2023-11-14T21:31:37.297258Z","iopub.status.idle":"2023-11-14T22:25:59.305166Z","shell.execute_reply.started":"2023-11-14T21:31:37.297226Z","shell.execute_reply":"2023-11-14T22:25:59.304035Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred__test = np.where(y_pred_test < 0.5, 1,0)\ntest_data['label'] = y_pred__test\n\n","metadata":{"execution":{"iopub.status.busy":"2023-11-14T22:49:27.460491Z","iopub.execute_input":"2023-11-14T22:49:27.460841Z","iopub.status.idle":"2023-11-14T22:49:27.465933Z","shell.execute_reply.started":"2023-11-14T22:49:27.460813Z","shell.execute_reply":"2023-11-14T22:49:27.464877Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_data.to_csv('/kaggle/working/results_test_set.csv')","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"category= {0:'dog' , 1:'cat'}\nsample_test = test_data.sample(n=16).reset_index() \nplt.figure(figsize=(12, 12))\nfor index, row in sample_test.iterrows():\n filename = row['path']\n label = row['label']\n img = load_img(filename, target_size=(224, 224))\n plt.subplot(4, 4, index+1)\n plt.imshow(img) \n plt.title(category[label])\n plt.axis('off')\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-11-14T22:51:53.243420Z","iopub.execute_input":"2023-11-14T22:51:53.243781Z","iopub.status.idle":"2023-11-14T22:51:54.664336Z","shell.execute_reply.started":"2023-11-14T22:51:53.243754Z","shell.execute_reply":"2023-11-14T22:51:54.662256Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Combination of Feature Extractor with ML
","metadata":{}},{"cell_type":"markdown","source":"Feature Extraction
","metadata":{}},{"cell_type":"code","source":"train_data=train_data.set_index('path')\ntest_data=test_data.set_index('path')\ntrain_data","metadata":{"execution":{"iopub.status.busy":"2023-11-14T23:01:27.810822Z","iopub.execute_input":"2023-11-14T23:01:27.811187Z","iopub.status.idle":"2023-11-14T23:01:27.827869Z","shell.execute_reply.started":"2023-11-14T23:01:27.811163Z","shell.execute_reply":"2023-11-14T23:01:27.826204Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"!pip install keras-cv-attention-models","metadata":{"execution":{"iopub.status.busy":"2023-11-14T23:23:40.015734Z","iopub.execute_input":"2023-11-14T23:23:40.016145Z","iopub.status.idle":"2023-11-14T23:23:52.756911Z","shell.execute_reply.started":"2023-11-14T23:23:40.016116Z","shell.execute_reply":"2023-11-14T23:23:52.755592Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\nfrom keras_cv_attention_models import moganet\nmoganet_model = moganet.MogaNetSmall(input_shape=(224, 224, 3), pretrained=\"imagenet\")\nbase_model = keras.Model(\n inputs=moganet_model.inputs,\n outputs=moganet_model.get_layer(name=\"avg_pool\").output,\n)","metadata":{"execution":{"iopub.status.busy":"2023-11-14T23:24:56.554239Z","iopub.execute_input":"2023-11-14T23:24:56.554833Z","iopub.status.idle":"2023-11-14T23:25:02.829343Z","shell.execute_reply.started":"2023-11-14T23:24:56.554797Z","shell.execute_reply":"2023-11-14T23:25:02.828310Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":" \ndef extract_features(image_path):\n img = image.load_img(image_path, target_size=(224, 224))\n x = image.img_to_array(img)\n x = np.expand_dims(x, axis=0)\n x = preprocess_input(x)\n features = base_model.predict(x)\n return features.flatten()\n\ntrain_features_df = pd.DataFrame(columns= [*range(512) , 'label'])\nfor image_path in (train_data.index): \n train_features_df.loc[image_path] = [*extract_features(image_path) , train_data.loc[image_path,'label'] ]\n \ntest_features_df = pd.DataFrame(columns= [*range(512) , 'label']) \nfor image_path in test_data.path: \n test_features_df.loc[image_path] = [*extract_features(image_path) , test_data.loc[image_path,'label'] ]\n \ntrain_features_df , test_features_df","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"ML model
","metadata":{}},{"cell_type":"code","source":"from sklearn.preprocessing import MinMaxScaler\nfrom sklearn.svm import SVC\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import accuracy_score\n\nscaler = MinMaxScaler()\ntrain_features_df[[0:512]] = scaler.fit_transform(train_features_df[[0:512]] )\nX_train, X_test, y_train, y_test = train_test_split(train_features_df, train_labels, test_size=0.2, random_state=42)\n \nsvm_classifier = SVC()\n \nsvm_classifier.fit(X_train, y_train)\n \ny_pred = svm_classifier.predict(X_test)\n \naccuracy = accuracy_score(y_test, y_pred)\nprint(\"Accuracy:\", accuracy)\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(classification_report(y_test, y_pred))","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}