{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"

Detección de fracturas de la columna cervical mediante técnicas de aprendizaje automático

\n\n
\n\n

TRABAJO FIN DE GRADO

\n

2022-2023

\n\n
\n\n

Domingo José Caballero Navarro

\n\n
\n\n

Tutorizado por:

\n\n

José Antonio Gámez Martín

\n

Juan Carlos Alfaro Jiménez

","metadata":{}},{"cell_type":"markdown","source":"---","metadata":{}},{"cell_type":"markdown","source":"## Índice\n1. **[Datos necesarios](#datos)**\n \n2. **[Diseño del modelo](#modelo)**\n\n3. **[Resultados y discusión](#resultados)**\n\n---","metadata":{}},{"cell_type":"markdown","source":"# 1. Datos necesarios \n
Hola
\n","metadata":{}},{"cell_type":"code","source":"! pip install python-gdcm\n! pip install pylibjpeg pylibjpeg-libjpeg pydicom","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:18.60846Z","iopub.execute_input":"2023-05-05T09:24:18.608936Z","iopub.status.idle":"2023-05-05T09:24:40.020517Z","shell.execute_reply.started":"2023-05-05T09:24:18.608852Z","shell.execute_reply":"2023-05-05T09:24:40.01932Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"seed = 270221","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:40.022898Z","iopub.execute_input":"2023-05-05T09:24:40.023522Z","iopub.status.idle":"2023-05-05T09:24:40.028837Z","shell.execute_reply.started":"2023-05-05T09:24:40.023478Z","shell.execute_reply":"2023-05-05T09:24:40.027913Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import tensorflow as tf","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:40.030508Z","iopub.execute_input":"2023-05-05T09:24:40.030857Z","iopub.status.idle":"2023-05-05T09:24:44.495206Z","shell.execute_reply.started":"2023-05-05T09:24:40.03082Z","shell.execute_reply":"2023-05-05T09:24:44.494174Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"device_type = \"GPU\" # The device type\ndevices = tf.config.list_physical_devices(device_type)\n\nif not devices:\n # Raise an informative message when there are no devices in the host runtime\n raise RuntimeError(f\"No {device_type} devices are used in the host.\")","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:44.497999Z","iopub.execute_input":"2023-05-05T09:24:44.498652Z","iopub.status.idle":"2023-05-05T09:24:44.675444Z","shell.execute_reply.started":"2023-05-05T09:24:44.498614Z","shell.execute_reply":"2023-05-05T09:24:44.673738Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import pandas as pd\n\ndata_df = pd.read_csv(\"../input/rsna-2022-cervical-spine-fracture-detection/train.csv\")","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:44.676892Z","iopub.execute_input":"2023-05-05T09:24:44.677327Z","iopub.status.idle":"2023-05-05T09:24:44.697059Z","shell.execute_reply.started":"2023-05-05T09:24:44.677285Z","shell.execute_reply":"2023-05-05T09:24:44.696219Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.model_selection import train_test_split\n\ntrain_df, test_df = train_test_split(data_df, test_size=0.1, stratify=data_df.patient_overall, random_state=seed)","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:44.698404Z","iopub.execute_input":"2023-05-05T09:24:44.698767Z","iopub.status.idle":"2023-05-05T09:24:44.977122Z","shell.execute_reply.started":"2023-05-05T09:24:44.698732Z","shell.execute_reply":"2023-05-05T09:24:44.976167Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"Train: \", train_df.shape)\nprint(\"Test: \", test_df.shape)","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:44.979261Z","iopub.execute_input":"2023-05-05T09:24:44.980023Z","iopub.status.idle":"2023-05-05T09:24:44.986528Z","shell.execute_reply.started":"2023-05-05T09:24:44.979982Z","shell.execute_reply":"2023-05-05T09:24:44.98523Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df_t1 = train_df[:360]","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df_t1, val_df_t1 = train_test_split(train_df_t1, test_size=0.1, stratify=train_df_t1.patient_overall, random_state=seed)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df_t1.head()","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:44.988198Z","iopub.execute_input":"2023-05-05T09:24:44.989255Z","iopub.status.idle":"2023-05-05T09:24:45.008065Z","shell.execute_reply.started":"2023-05-05T09:24:44.989218Z","shell.execute_reply":"2023-05-05T09:24:45.007217Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os \nfrom os import listdir\n\npath = \"../input/rsna-2022-cervical-spine-fracture-detection\"\npath_train_images = os.path.join(path, \"train_images\")\npath_segmentations = os.path.join(path, \"segmentations\")","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:45.010322Z","iopub.execute_input":"2023-05-05T09:24:45.011309Z","iopub.status.idle":"2023-05-05T09:24:45.01695Z","shell.execute_reply.started":"2023-05-05T09:24:45.011274Z","shell.execute_reply":"2023-05-05T09:24:45.01605Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"metadata_df = pd.read_csv(\"../input/explicacionrsna/metadata_df.csv\")","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:45.022105Z","iopub.execute_input":"2023-05-05T09:24:45.022404Z","iopub.status.idle":"2023-05-05T09:24:47.40235Z","shell.execute_reply.started":"2023-05-05T09:24:45.022379Z","shell.execute_reply":"2023-05-05T09:24:47.401359Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"metadata_df.head()","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:47.403963Z","iopub.execute_input":"2023-05-05T09:24:47.404334Z","iopub.status.idle":"2023-05-05T09:24:47.418819Z","shell.execute_reply.started":"2023-05-05T09:24:47.404294Z","shell.execute_reply":"2023-05-05T09:24:47.417596Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"
En este caso podemos apreciar la forma de la vértebra, en este caso podemos apreciar cómo esta mínimamente desplaza hacía arriba. Teniendo esto en cuenta, deberemos comprobar con otros ejemplos si esto sucede también en los demás pacientes dado que de ser negativo tendríamos que centrar todas las imágenes. Por lo tanto, comprobaremos el segundo paciente del conjunto de entrenamiento.
","metadata":{}},{"cell_type":"code","source":"train_df_t1.head()","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:47.420207Z","iopub.execute_input":"2023-05-05T09:24:47.421452Z","iopub.status.idle":"2023-05-05T09:24:47.436178Z","shell.execute_reply.started":"2023-05-05T09:24:47.421415Z","shell.execute_reply":"2023-05-05T09:24:47.435231Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Preprocesamiento","metadata":{}},{"cell_type":"code","source":"train_segmented = pd.read_csv(\"../input/segmentation/train_segmented.csv\")","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:47.437565Z","iopub.execute_input":"2023-05-05T09:24:47.437922Z","iopub.status.idle":"2023-05-05T09:24:52.051029Z","shell.execute_reply.started":"2023-05-05T09:24:47.437887Z","shell.execute_reply":"2023-05-05T09:24:52.049955Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_segmented = train_segmented[['SOPInstanceUID','C1','C2','C3','C4','C5','C6','C7']]\ntrain_segmented.head()","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_segmented[['C1', 'C2', 'C3', 'C4', 'C5', 'C6', 'C7']] = train_segmented[['C1', 'C2', 'C3', 'C4', 'C5', 'C6', 'C7']].applymap(lambda x: 1 if x >= 0.5 else 0)\ntrain_segmented.head()","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:52.05249Z","iopub.execute_input":"2023-05-05T09:24:52.052891Z","iopub.status.idle":"2023-05-05T09:24:52.072536Z","shell.execute_reply.started":"2023-05-05T09:24:52.052853Z","shell.execute_reply":"2023-05-05T09:24:52.07113Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_segmented[train_segmented['SOPInstanceUID']=='1.2.826.0.1.3680043.14723.1.100'][['C1','C2','C3','C4','C5','C6','C7']]","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:52.074336Z","iopub.execute_input":"2023-05-05T09:24:52.07504Z","iopub.status.idle":"2023-05-05T09:24:52.133016Z","shell.execute_reply.started":"2023-05-05T09:24:52.075001Z","shell.execute_reply":"2023-05-05T09:24:52.132003Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# 2. Diseño del modelo \n
Modelo
","metadata":{}},{"cell_type":"markdown","source":"# RSNA Efficient-net Baseline","metadata":{}},{"cell_type":"code","source":"import os\nimport cv2\nimport glob\nimport traceback\nimport numpy as np\nimport pandas as pd\nfrom path import Path\nfrom tqdm import tqdm\nimport nibabel as nib\nimport pydicom\nimport tensorflow as tf\nfrom keras import layers\nfrom pydicom import dcmread\nfrom tensorflow import keras\nimport tensorflow_hub as hub\nimport matplotlib.pyplot as plt\nfrom tensorflow.keras import backend as K\nfrom pydicom.data import get_testdata_files\nfrom tensorflow.keras.utils import to_categorical\nfrom sklearn.model_selection import StratifiedKFold\nfrom pydicom.pixel_data_handlers.util import apply_voi_lut\nfrom tensorflow.keras.layers import Input, Dense, Flatten, Conv2D\nfrom tensorflow.keras.preprocessing.image import load_img, img_to_array","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:52.134663Z","iopub.execute_input":"2023-05-05T09:24:52.135032Z","iopub.status.idle":"2023-05-05T09:24:53.723581Z","shell.execute_reply.started":"2023-05-05T09:24:52.134998Z","shell.execute_reply":"2023-05-05T09:24:53.722602Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"path_train_png = '../input/images-preprocessed/images_preprocessed/train_images_preprocessed'","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.738787Z","iopub.execute_input":"2023-05-05T09:24:53.739574Z","iopub.status.idle":"2023-05-05T09:24:53.752736Z","shell.execute_reply.started":"2023-05-05T09:24:53.73952Z","shell.execute_reply":"2023-05-05T09:24:53.75176Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport cv2\nfrom PIL import Image","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.75584Z","iopub.execute_input":"2023-05-05T09:24:53.756096Z","iopub.status.idle":"2023-05-05T09:24:53.763672Z","shell.execute_reply.started":"2023-05-05T09:24:53.756072Z","shell.execute_reply":"2023-05-05T09:24:53.762766Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def load_dicom(path, size = 64):\n try:\n img=pydicom.dcmread(path)\n img.PhotometricInterpretation = 'YBR_FULL'\n data=img.pixel_array\n data=data-np.min(data)\n if np.max(data) != 0:\n data=data/np.max(data)\n data=(data*255).astype(np.uint8) \n return cv2.cvtColor(data.reshape(512, 512), cv2.COLOR_GRAY2RGB)\n except: \n return np.zeros((512, 512, 3))\n \ndef load_png(path, size = 64):\n img = Image.open(path).convert(\"RGB\")\n img = img.resize((size, size))\n img_array = np.array(img) / 255.0\n return img_array","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.765185Z","iopub.execute_input":"2023-05-05T09:24:53.765444Z","iopub.status.idle":"2023-05-05T09:24:53.776093Z","shell.execute_reply.started":"2023-05-05T09:24:53.76542Z","shell.execute_reply":"2023-05-05T09:24:53.775209Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import pydicom\nfrom pydicom import dcmread\nfrom pydicom.data import get_testdata_files\nfrom pydicom.pixel_data_handlers.util import apply_voi_lut\n\nimport tensorflow as tf\nfrom keras import layers\nfrom tensorflow import keras\nimport tensorflow_hub as hub\nfrom tensorflow.keras import backend as K\nfrom tensorflow.keras.utils import to_categorical\nfrom tensorflow.keras.layers import Input, Dense, Flatten, Conv2D\nfrom tensorflow.keras.preprocessing.image import load_img, img_to_array","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.790535Z","iopub.execute_input":"2023-05-05T09:24:53.79097Z","iopub.status.idle":"2023-05-05T09:24:53.800196Z","shell.execute_reply.started":"2023-05-05T09:24:53.790936Z","shell.execute_reply":"2023-05-05T09:24:53.799289Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_indices = train_df_t1.index.tolist()","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.817251Z","iopub.execute_input":"2023-05-05T09:24:53.817608Z","iopub.status.idle":"2023-05-05T09:24:53.825097Z","shell.execute_reply.started":"2023-05-05T09:24:53.817574Z","shell.execute_reply":"2023-05-05T09:24:53.824176Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def data_generator():\n \n for i in train_indices: \n \n idt = train_df_t1.loc[i, 'StudyInstanceUID']\n path_png = os.path.join(path_train_png, idt)\n \n for im in os.listdir(path_png):\n train_label = []\n img = load_png(os.path.join(path_png , im))\n \n sop_uid = idt+'.1.'+im\n sop_uid = sop_uid.split(\".png\")[0]\n \n if sop_uid in np.array(train_segmented['SOPInstanceUID']):\n vert_prob = np.array(train_segmented[train_segmented['SOPInstanceUID']==sop_uid][['C1','C2','C3','C4','C5','C6','C7']])\n vert_prob = vert_prob[0]\n vert_prob = np.array(vert_prob)\n \n train_label.extend([\n train_df_t1.loc[i, \"C1\"],\n train_df_t1.loc[i, \"C2\"],\n train_df_t1.loc[i, \"C3\"],\n train_df_t1.loc[i, \"C4\"],\n train_df_t1.loc[i, \"C5\"],\n train_df_t1.loc[i, \"C6\"],\n train_df_t1.loc[i, \"C7\"]\n ])\n\n train_label = np.array(train_label)\n\n fracture_label = vert_prob * train_label\n \n \n yield (img, vert_prob), fracture_label","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def data_generator_without_vert():\n \n for i in train_indices: \n \n idt = train_df_t1.loc[i, 'StudyInstanceUID']\n path_png = os.path.join(path_train_png, idt)\n \n for im in os.listdir(path_png):\n train_label = []\n\n img = load_png(os.path.join(path_png,im))\n \n img = cv2.resize(img, (64 , 64))\n image = img_to_array(img)\n image = image / 255.0\n \n train_label.extend([\n train_df_t1.loc[i, \"patient_overall\"],\n train_df_t1.loc[i, \"C1\"],\n train_df_t1.loc[i, \"C2\"],\n train_df_t1.loc[i, \"C3\"],\n train_df_t1.loc[i, \"C4\"],\n train_df_t1.loc[i, \"C5\"],\n train_df_t1.loc[i, \"C6\"],\n train_df_t1.loc[i, \"C7\"]\n ])\n \n yield image, train_label\n ","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.86319Z","iopub.execute_input":"2023-05-05T09:24:53.863621Z","iopub.status.idle":"2023-05-05T09:24:53.87399Z","shell.execute_reply.started":"2023-05-05T09:24:53.863586Z","shell.execute_reply":"2023-05-05T09:24:53.873373Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"val_indices = val_df_t1.index.tolist()","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def val_generator():\n \n for i in val_indices: \n \n idt = val_df_t1.loc[i, 'StudyInstanceUID']\n path_png = os.path.join(path_train_png, idt)\n \n for im in os.listdir(path_png):\n train_label = []\n img = load_png(os.path.join(path_png , im))\n \n sop_uid = idt+'.1.'+im\n sop_uid = sop_uid.split(\".png\")[0]\n \n if sop_uid in np.array(train_segmented['SOPInstanceUID']):\n vert_prob = np.array(train_segmented[train_segmented['SOPInstanceUID']==sop_uid][['C1','C2','C3','C4','C5','C6','C7']])\n vert_prob = vert_prob[0]\n vert_prob = np.array(vert_prob)\n \n train_label.extend([\n val_df_t1.loc[i, \"C1\"],\n val_df_t1.loc[i, \"C2\"],\n val_df_t1.loc[i, \"C3\"],\n val_df_t1.loc[i, \"C4\"],\n val_df_t1.loc[i, \"C5\"],\n val_df_t1.loc[i, \"C6\"],\n val_df_t1.loc[i, \"C7\"]\n ])\n\n train_label = np.array(train_label)\n\n fracture_label = vert_prob * train_label\n \n \n yield (img, vert_prob), fracture_label","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def val_generator_without_vert():\n \n for i in val_indices: \n \n idt = val_df_t1.loc[i, 'StudyInstanceUID']\n path_dcm = os.path.join(path_train_images, idt)\n path_png = os.path.join(path_train_png, idt)\n \n for im in os.listdir(path_png):\n train_label = []\n\n img = load_png(os.path.join(path_png,im))\n \n img = cv2.resize(img, (64 , 64))\n image = img_to_array(img)\n image = image / 255.0\n \n train_label.extend([\n val_df_t1.loc[i, \"patient_overall\"],\n val_df_t1.loc[i, \"C1\"],\n val_df_t1.loc[i, \"C2\"],\n val_df_t1.loc[i, \"C3\"],\n val_df_t1.loc[i, \"C4\"],\n val_df_t1.loc[i, \"C5\"],\n val_df_t1.loc[i, \"C6\"],\n val_df_t1.loc[i, \"C7\"]\n ])\n \n yield image, train_label\n ","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_model():\n inp1 = keras.layers.Input(shape=(None, None, 1), name='image')\n inp2 = keras.layers.Input(shape=(7,), name='vert_prob')\n \n x = Conv2D(3, 3, padding='SAME')(inp1)\n \n x = keras.applications.efficientnet.EfficientNetB5(include_top=False)(x)\n x = keras.layers.GlobalAveragePooling2D()(x)\n x = keras.layers.BatchNormalization()(x)\n \n x2 = keras.layers.Dense(32, activation='relu')(inp2)\n \n x = keras.layers.concatenate([x, x2])\n \n x = keras.layers.Dropout(0.2)(x)\n \n out = keras.layers.Dense(7, 'sigmoid')(x)\n \n model = keras.models.Model(inputs=[inp1, inp2], outputs=out)\n \n return model","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.909691Z","iopub.execute_input":"2023-05-05T09:24:53.909976Z","iopub.status.idle":"2023-05-05T09:24:53.920194Z","shell.execute_reply.started":"2023-05-05T09:24:53.909951Z","shell.execute_reply":"2023-05-05T09:24:53.91929Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_model_without_vert():\n inp = keras.layers.Input(shape=(None, None, 1), name='image')\n \n x = Conv2D(3, 3, padding='SAME')(inp)\n \n x = keras.applications.EfficientNetV2.pretrained()(x)\n \n x = keras.layers.GlobalAveragePooling2D()(x)\n \n x = keras.layers.Dense(16, activation='relu')(x)\n x = keras.layers.Dropout(0.2)(x)\n out = keras.layers.Dense(8, 'softmax')(x)\n \n model = keras.models.Model(inputs=inp, outputs=out)\n \n return model","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.92258Z","iopub.execute_input":"2023-05-05T09:24:53.923315Z","iopub.status.idle":"2023-05-05T09:24:53.931753Z","shell.execute_reply.started":"2023-05-05T09:24:53.923278Z","shell.execute_reply":"2023-05-05T09:24:53.930813Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import traceback\nfrom sklearn.model_selection import StratifiedKFold","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.945237Z","iopub.execute_input":"2023-05-05T09:24:53.945673Z","iopub.status.idle":"2023-05-05T09:24:53.953098Z","shell.execute_reply.started":"2023-05-05T09:24:53.945635Z","shell.execute_reply":"2023-05-05T09:24:53.952201Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def configure_for_performance(data):\n data = data.cache()\n data = data.batch(8)\n data = data.prefetch(buffer_size=tf.data.AUTOTUNE)\n return data","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:56.435974Z","iopub.execute_input":"2023-05-05T09:24:56.436324Z","iopub.status.idle":"2023-05-05T09:24:56.442659Z","shell.execute_reply.started":"2023-05-05T09:24:56.436289Z","shell.execute_reply":"2023-05-05T09:24:56.441588Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_data = tf.data.Dataset.from_generator(data_generator, ((tf.float32, tf.int8), tf.int8))","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.967304Z","iopub.execute_input":"2023-05-05T09:24:53.967666Z","iopub.status.idle":"2023-05-05T09:24:53.975995Z","shell.execute_reply.started":"2023-05-05T09:24:53.967632Z","shell.execute_reply":"2023-05-05T09:24:53.975071Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_data = configure_for_performance(train_data)","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:56.444612Z","iopub.execute_input":"2023-05-05T09:24:56.44534Z","iopub.status.idle":"2023-05-05T09:24:56.452288Z","shell.execute_reply.started":"2023-05-05T09:24:56.445303Z","shell.execute_reply":"2023-05-05T09:24:56.451343Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"val_data = tf.data.Dataset.from_generator(val_generator, ((tf.float32, tf.int8), tf.int8))","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"val_data = configure_for_performance(val_data)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#train_data_wv = tf.data.Dataset.from_generator(data_generator_without_vert, (tf.float32, tf.int8))","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:53.977694Z","iopub.execute_input":"2023-05-05T09:24:53.978035Z","iopub.status.idle":"2023-05-05T09:24:56.434629Z","shell.execute_reply.started":"2023-05-05T09:24:53.978001Z","shell.execute_reply":"2023-05-05T09:24:56.433675Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#train_data_wv = configure_for_performance(train_data_wv)","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:56.453942Z","iopub.execute_input":"2023-05-05T09:24:56.454351Z","iopub.status.idle":"2023-05-05T09:24:56.46764Z","shell.execute_reply.started":"2023-05-05T09:24:56.454316Z","shell.execute_reply":"2023-05-05T09:24:56.46675Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#val_data_wv = tf.data.Dataset.from_generator(val_generator_without_vert, (tf.float32, tf.int8))","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#val_data_wv = configure_for_performance(val_data_wv)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#model_wv = get_model_without_vert()\n#model_wv.summary()","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:24:56.469278Z","iopub.execute_input":"2023-05-05T09:24:56.469722Z","iopub.status.idle":"2023-05-05T09:25:06.095558Z","shell.execute_reply.started":"2023-05-05T09:24:56.469688Z","shell.execute_reply":"2023-05-05T09:25:06.094595Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.callbacks import ModelCheckpoint\nfrom tensorflow.keras.callbacks import EarlyStopping\n\ncheckpoint_filepath = 'p1_best_model_efficientnetb5.h5'\n\nmodel_checkpoint_callback = ModelCheckpoint(\n checkpoint_filepath,\n save_best_only=True,\n monitor='val_accuracy',\n mode='max',\n verbose=1)\n\nearly_stopping = EarlyStopping(monitor='val_loss', \n min_delta=0, \n patience=5, \n verbose=1)\n\ncallbacks = [early_stopping, model_checkpoint_callback]","metadata":{"execution":{"iopub.status.busy":"2023-05-06T06:50:55.680715Z","iopub.execute_input":"2023-05-06T06:50:55.681332Z","iopub.status.idle":"2023-05-06T06:51:01.359042Z","shell.execute_reply.started":"2023-05-06T06:50:55.681246Z","shell.execute_reply":"2023-05-06T06:51:01.358091Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\"\"\"model_wv.compile(optimizer=tf.keras.optimizers.Adam(), \n loss=tf.keras.losses.CategoricalCrossentropy(),\n metrics=[tf.keras.metrics.CategoricalAccuracy()]\n )\"\"\"","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:25:06.097032Z","iopub.execute_input":"2023-05-05T09:25:06.097389Z","iopub.status.idle":"2023-05-05T09:25:06.123834Z","shell.execute_reply.started":"2023-05-05T09:25:06.097343Z","shell.execute_reply":"2023-05-05T09:25:06.122958Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#hist = model_wv.fit(train_data_wv, validation_data=val_data_wv, epochs = 25, callbacks=callbacks)","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:25:06.125295Z","iopub.execute_input":"2023-05-05T09:25:06.125687Z","iopub.status.idle":"2023-05-05T09:34:47.511592Z","shell.execute_reply.started":"2023-05-05T09:25:06.125651Z","shell.execute_reply":"2023-05-05T09:34:47.510477Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#model_wv.fit(val_data_wv, epochs=2)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#model_wv.save('t3_model_without_vert_efficientnetb5.h5')","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#model = get_model()\n \n#hist = model.fit(train_data, steps_per_epoch=len(train_indices) // batch_size, epochs = 2)","metadata":{"execution":{"iopub.status.busy":"2023-05-05T09:34:47.513332Z","iopub.execute_input":"2023-05-05T09:34:47.513763Z","iopub.status.idle":"2023-05-05T09:34:47.518158Z","shell.execute_reply.started":"2023-05-05T09:34:47.513715Z","shell.execute_reply":"2023-05-05T09:34:47.517143Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model = get_model()","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.compile(optimizer=tf.keras.optimizers.Adam(), \n loss=tf.keras.losses.BinaryCrossentropy(),\n metrics=[tf.keras.metrics.Precision(), tf.keras.metrics.AUC(multi_label=True)]\n )","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.fit(train_data, validation_data=val_data, epochs = 25, callbacks=callbacks)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"try:\n model.load_weights('/kaggle/working/p1_best_model_efficientnetb5.h5')\nexcept:\n print(\"No se han cargado los pesos\")","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.fit(val_data, epochs=2)","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.save('p1_model_efficientnetb5.h5')","metadata":{},"execution_count":null,"outputs":[]}]}