{"cells":[{"metadata":{"_uuid":"6567b426-67d6-4a54-89c5-4d8637f55a18","_cell_guid":"62ff1e40-4e43-4baf-bdc8-023200965586","trusted":true},"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\nlstFilesDCM_train = dict()\nlstFilesDCM_test = dict()\nimport os\n\nfor dirname, _, filenames in os.walk('/kaggle/input'):\n    for filename in filenames:\n        if \".dcm\" in filename.lower(): \n            print(os.path.join(dirname, filename))\n            if 'train' in dirname:\n                \n                #lstFilesDCM.append(os.path.join('/kaggle/input',dirname,filename))\n                lstFilesDCM_train.setdefault(dirname.split('/')[-1],[]).append(os.path.join('/kaggle/input',dirname,filename))\n            else:\n                lstFilesDCM_test.setdefault(dirname.split('/')[-1],[]).append(os.path.join('/kaggle/input',dirname,filename))\n                \n\n# You can write up to 5GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"a3f04d4c-b8e9-4ade-b213-78902b88bb34","_cell_guid":"31eaba56-6e13-4758-bcc5-979c79cbfacc","trusted":true},"cell_type":"code","source":"from pydicom import dcmread\nfrom pydicom.data import get_testdata_files\nimport matplotlib.pyplot as plt\nimport random\nimport cv2\n\nfrom skimage import measure\nfrom skimage import morphology\nfrom sklearn.cluster import KMeans\nfrom pydicom.pixel_data_handlers.util import apply_color_lut\neditmode=False","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"bf8d8741-c6e1-4caa-a857-8b121d630226","_cell_guid":"c1947c66-f587-4235-b716-57655489e438","trusted":true},"cell_type":"code","source":"def smokeprocess(thiscat):\n    smocat=['Currently smokes', 'Ex-smoker', 'Never smoked']\n    rtoh=list(np.zeros(len(smocat),dtype=np.int64))\n    rtoh[smocat.index(thiscat)]=1\n    return list(rtoh)\ndef sexprocess(thiscat):\n    smocat=['Female', 'Male']\n    rtoh=list(np.zeros(len(smocat),dtype=np.int64))\n    rtoh[smocat.index(thiscat)]=1\n    return list(rtoh)\n\ndef fillna(train_df,col,typ):\n    if typ=='cat':\n        return train_df.groupby(col).count().idxmax()[0]\n    else:\n        return train_df[col].median()\nfrom sklearn.preprocessing import RobustScaler       \n\ntrain_df=pd.read_csv('../input/osic-pulmonary-fibrosis-progression/train.csv')\nprint(train_df.columns)\nmiss=0\nX=[]\nXi=[]\nY=[]\nXr=[]\nsexna=fillna(train_df,'Sex','cat')\nagena=fillna(train_df,'Age','')\nbool_series = pd.isnull(train_df['Sex'])\ntrain_df[bool_series]=sexna\nbool_series = pd.isnull(train_df['Age'])\ntrain_df[bool_series]=agena\nidc=0\nuse_data=0\nfor dcmfk in list(lstFilesDCM_train.keys()):\n    idc=idc+1\n    if idc>20 and editmode:\n        break\n    pt=True\n    for dcmf in lstFilesDCM_train[dcmfk]:\n        #print(dcmf.split('/')[-1].split('.dcm')[0])\n        ds = dcmread(dcmf)\n\n\n        #print(dcmf.split('/')[-1].split('.dcm')[0])\n        #print(train_df[(train_df['Patient']==dcmfk) & (train_df['Weeks']==dcmf.split('/')[-1].split('.dcm')[0])])\n        try:\n            \"\"\"\n            if ds.pixel_array.shape!=(512,512):\n                res = cv2.resize(xx, dsize=(512, 512), interpolation=cv2.INTER_CUBIC)\n            else:\n                res = ds.pixel_array\n            \"\"\"\n            image_2d = ds.pixel_array.astype(float)\n            mean = np.mean(image_2d)\n            std = np.std(image_2d)\n            image_2d = image_2d - mean\n            image_2d = image_2d / std\n            image_2d = image_2d-image_2d.min()\n            \n            #image_2d = cv2.resize(image_2d, dsize=(224, 224), interpolation=cv2.INTER_LINEAR)\n            #image_2d_scaled = (np.maximum(image_2d,0) / image_2d.max()) * 255.0\n            image_2d_scaled = np.clip(image_2d / np.quantile(image_2d,0.99),0,1) #* 255.0\n            a0=image_2d_scaled.shape[0]\n            a1=image_2d_scaled.shape[1]\n            kmeans = KMeans(n_clusters=2).fit(image_2d_scaled[int(a0/2)-100:int(a0/2)+100:,int(a1/2)-100:int(a1/2)+100].reshape(-1,1))\n            centers = sorted(kmeans.cluster_centers_.flatten())\n            threshold = np.mean(centers)\n            image_2d_scaled = np.where(image_2d_scaled < threshold, 1.0, 0.0)            \n\n            eroded = morphology.erosion(image_2d_scaled, np.ones([2, 2]))\n            dilation = morphology.dilation(image_2d_scaled, np.ones([4, 4]))     \n\n            \n\n            labels = measure.label(dilation)  # Different labels are displayed in different colors\n            label_vals = np.unique(labels)\n            regions = measure.regionprops(labels)            \n            good_labels = []\n            for prop in regions:\n                B = prop.bbox\n                if B[2] - B[0] < image_2d_scaled.shape[0] / 10 * 9 and B[3] - B[1] <  image_2d_scaled.shape[1] / 10 * 9 and B[0] >  image_2d_scaled.shape[0] / 5 and B[2] <  image_2d_scaled.shape[1] / 5 * 4:\n                    good_labels.append(prop.label)\n            mask = np.ndarray([image_2d_scaled.shape[0], image_2d_scaled.shape[1]], dtype=np.int8)\n            mask[:] = 0\n            \n\n            #\n            #  After just the lungs are left, we do another large dilation\n            #  in order to fill in and out the lung mask\n            #\n            for N in good_labels:\n                mask = mask + np.where(labels == N, 1, 0)\n            mask = morphology.dilation(mask, np.ones([10, 10]))            \n            image_2d_scaled=image_2d_scaled*mask\n            res = cv2.resize(image_2d_scaled, dsize=(224, 224), interpolation=cv2.INTER_LINEAR)\n            \"\"\"\n            if pt:\n                plt.figure()\n                plt.pcolor(res)\n                pt=False\n            \"\"\"\n            #res = cv2.resize(ds.pixel_array, dsize=(128, 128), interpolation=cv2.INTER_CUBIC)\n            #res=RobustScaler().fit_transform(res)\n            meta=train_df[(train_df['Patient']==dcmfk) & (train_df['Weeks']==int(dcmf.split('/')[-1].split('.dcm')[0]))]\n            succ=0\n            try:\n                \n                Y.append(train_df[(train_df['Patient']==dcmfk) & (train_df['Weeks']==int(dcmf.split('/')[-1].split('.dcm')[0]))].FVC.values[0])\n                X.append(res)\n                Xi.append(sexprocess(meta['Sex'].values[0])+[meta['Age'].values[0]]+smokeprocess(meta['SmokingStatus'].values[0]))\n                Xr.append([dcmfk,int(dcmf.split('/')[-1].split('.dcm')[0])])\n                use_data=use_data+1\n\n                for i0 in range(0): \n                    img_flip=random.choice([0,-1])\n                    res= cv2.flip(res, img_flip)\n                    res= cv2.resize(res, dsize=(236, 236), interpolation=cv2.INTER_CUBIC)\n                    for i in range(1): \n\n                        shift=random.randint(1,12)\n                        shift1=random.randint(1,12)\n                        resn=res[shift:224+shift,shift1:224+shift1]\n                        Y.append(train_df[(train_df['Patient']==dcmfk) & (train_df['Weeks']==int(dcmf.split('/')[-1].split('.dcm')[0]))].FVC.values[0])\n                        X.append(resn)\n                        Xi.append(sexprocess(meta['Sex'].values[0])+[meta['Age'].values[0]]+smokeprocess(meta['SmokingStatus'].values[0]))\n                print('ID',idc,'USEDATA',use_data,'times3')\n            except:\n\n                miss=miss+1\n        except:\n            1==1\n\n\n        # 列出所有後設資料（metadata）\n        # print(ds)\n        #print(ds.PatientName)\n        # 以 matplotlib 繪製影像\n        #plt.imshow(ds.pixel_array)\n        #plt.show()","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"5b48c3c8-d0b0-4dd8-ae6d-4e1016522874","_cell_guid":"13f5f823-02f0-4b44-8e22-c919f131b42b","trusted":true},"cell_type":"code","source":"plt.figure()\n\n\nplt.pcolor(ds.pixel_array.astype(float))\n\n\"\"\"\nX\nXi\nY\nXr\n\"\"\"","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"2af2f895-a3fc-4bf4-a32c-565f2030e706","_cell_guid":"3f5db9c2-0afb-4d20-afe1-68c852ea5d80","trusted":true},"cell_type":"code","source":"\ntest_df=pd.read_csv('../input/osic-pulmonary-fibrosis-progression/test.csv')\nprint(test_df.columns)\nmiss=0\nX_test=[]\nXi_test=[]\nXr_test=[]\nY_test=[]\n\nX_test_no=[]\nXi_test_no=[]\nXr_test_no=[]\n\nbool_series = pd.isnull(test_df['Sex'])\ntest_df[bool_series]=sexna\nbool_series = pd.isnull(test_df['Age'])\ntest_df[bool_series]=agena\nidc=0\nuse_data=0\nfor dcmfk in list(lstFilesDCM_test.keys()):\n    idc=idc+1\n    if idc>10 and editmode:\n        break\n    for dcmf in lstFilesDCM_test[dcmfk]:\n        ds = dcmread(dcmf)\n\n        try:\n            image_2d = ds.pixel_array.astype(float)\n            mean = np.mean(image_2d)\n            std = np.std(image_2d)\n            image_2d = image_2d - mean\n            image_2d = image_2d / std\n            image_2d = image_2d-image_2d.min()\n            \n            #image_2d = cv2.resize(image_2d, dsize=(224, 224), interpolation=cv2.INTER_LINEAR)\n            #image_2d_scaled = (np.maximum(image_2d,0) / image_2d.max()) * 255.0\n            image_2d_scaled = np.clip(image_2d / np.quantile(image_2d,0.99),0,1) #* 255.0\n            a0=image_2d_scaled.shape[0]\n            a1=image_2d_scaled.shape[1]\n            kmeans = KMeans(n_clusters=2).fit(image_2d_scaled[int(a0/2)-100:int(a0/2)+100:,int(a1/2)-100:int(a1/2)+100].reshape(-1,1))\n            centers = sorted(kmeans.cluster_centers_.flatten())\n            threshold = np.mean(centers)\n            image_2d_scaled = np.where(image_2d_scaled < threshold, 1.0, 0.0)            \n\n            eroded = morphology.erosion(image_2d_scaled, np.ones([2, 2]))\n            dilation = morphology.dilation(image_2d_scaled, np.ones([4, 4]))     \n\n            \n\n            labels = measure.label(dilation)  # Different labels are displayed in different colors\n            label_vals = np.unique(labels)\n            regions = measure.regionprops(labels)            \n            good_labels = []\n            for prop in regions:\n                B = prop.bbox\n                if B[2] - B[0] < image_2d_scaled.shape[0] / 10 * 9 and B[3] - B[1] <  image_2d_scaled.shape[1] / 10 * 9 and B[0] >  image_2d_scaled.shape[0] / 5 and B[2] <  image_2d_scaled.shape[1] / 5 * 4:\n                    good_labels.append(prop.label)\n            mask = np.ndarray([image_2d_scaled.shape[0], image_2d_scaled.shape[1]], dtype=np.int8)\n            mask[:] = 0\n            \n\n            #\n            #  After just the lungs are left, we do another large dilation\n            #  in order to fill in and out the lung mask\n            #\n            for N in good_labels:\n                mask = mask + np.where(labels == N, 1, 0)\n            mask = morphology.dilation(mask, np.ones([10, 10]))            \n            image_2d_scaled=image_2d_scaled*mask\n            res = cv2.resize(image_2d_scaled, dsize=(224, 224), interpolation=cv2.INTER_LINEAR)\n            #res = cv2.resize(ds.pixel_array, dsize=(224, 224), interpolation=cv2.INTER_CUBIC)\n            #res=RobustScaler().fit_transform(res)\n            meta=test_df[(test_df['Patient']==dcmfk) & (test_df['Weeks']==int(dcmf.split('/')[-1].split('.dcm')[0]))]\n            succ=0\n            try:\n                Y_test.append(test_df[(test_df['Patient']==dcmfk) & (test_df['Weeks']==int(dcmf.split('/')[-1].split('.dcm')[0]))].FVC.values[0])\n                X_test.append(res)\n                Xi_test.append(sexprocess(meta['Sex'].values[0])+[meta['Age'].values[0]]+smokeprocess(meta['SmokingStatus'].values[0]))\n                Xr_test.append([dcmfk,int(dcmf.split('/')[-1].split('.dcm')[0])])\n            except:\n                miss=miss+1\n                meta=pd.DataFrame(test_df[(test_df['Patient']==dcmfk)].iloc[0,:]).T\n                X_test_no.append(res)\n                Xi_test_no.append(sexprocess(meta['Sex'].values[0])+[meta['Age'].values[0]]+smokeprocess(meta['SmokingStatus'].values[0]))\n                Xr_test_no.append([dcmfk,int(dcmf.split('/')[-1].split('.dcm')[0])])\n        except:\n            1==1","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"e24588bd-88df-4241-9993-e40ad852e9ec","_cell_guid":"7ecefab5-aa5b-4c6e-a46d-c103e81d1d83","trusted":true},"cell_type":"code","source":"\"\"\"\nrgb_batch_train = np.repeat(np.array(X)[..., np.newaxis], 3, -1)\nrgb_batch_test = np.repeat(np.array(X_test)[..., np.newaxis], 3, -1)\nrgb_batch_val = np.repeat(np.array(X_test_no)[..., np.newaxis], 3, -1)\nprint(rgb_batch_train.shape,rgb_batch_test.shape,rgb_batch_val.shape)\n\nrgb_batch_lab=np.append(rgb_batch_train,rgb_batch_test,axis=0)\nrgb_batch=np.append(rgb_batch_lab,rgb_batch_val,axis=0)\nrgb_batch.shape\n\"\"\"\nfrom sklearn.preprocessing import MinMaxScaler\nrgb_batch_Y=np.append(np.array(Y).reshape(-1,1),np.array(Y_test).reshape(-1,1),axis=0)\nscaler = MinMaxScaler()\nscaler.fit(rgb_batch_Y)\nprint(scaler.data_max_)\nYt=scaler.transform(rgb_batch_Y)\nYt.shape\nplt.hist(Yt)\nplt.hist(rgb_batch_Y)\nYt=rgb_batch_Y\n\n\"\"\"\nbins = np.linspace(min(np.unique(np.round(Y))), max(np.unique(np.round(Y))), 200)\ndigitized = np.digitize(Y, bins)\nfrom sklearn.preprocessing import LabelBinarizer\ny = LabelBinarizer().fit_transform(digitized)\n\"\"\"","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"0fe55adc-c600-405d-bd50-947ec9799d10","_cell_guid":"cb175cb7-cb98-4e93-9025-416321f8ce60","trusted":true},"cell_type":"code","source":"rgb_batch_lab=np.append(X,X_test,axis=0)\nrgb_batch=np.append(rgb_batch_lab,X_test_no,axis=0)\nrgb_batch=rgb_batch.reshape(-1,224,224,1)\nrgb_batch.shape","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"51338ca2-2fc3-4234-bd3b-0c3de7bf308c","_cell_guid":"eea723bc-088a-41a6-baaa-3370184d2487","trusted":true},"cell_type":"code","source":"import os\nimport random\nimport pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\nplt.style.use(\"ggplot\")\n%matplotlib inline\n\nfrom tqdm import tqdm_notebook, tnrange\nfrom itertools import chain\nfrom skimage.io import imread, imshow, concatenate_images\nfrom skimage.transform import resize\nfrom skimage.morphology import label\nfrom sklearn.model_selection import train_test_split\n\nimport tensorflow as tf\n\nfrom keras.models import Model, load_model\nfrom keras.layers import Input, BatchNormalization, Activation, Dense, Dropout\nfrom keras.layers.core import Lambda, RepeatVector, Reshape\nfrom keras.layers.convolutional import Conv2D, Conv2DTranspose\nfrom keras.layers.pooling import MaxPooling2D, GlobalMaxPool2D\nfrom keras.layers.merge import concatenate, add\nfrom keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau\nfrom keras.optimizers import Adam\nfrom keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"69c0d4a5-02bc-4ba0-8642-0801855dba5e","_cell_guid":"851bc9ae-5fb3-40ef-8a2a-98b1805bd995","trusted":true},"cell_type":"code","source":"def conv2d_block(input_tensor, n_filters, kernel_size=3, batchnorm=True):\n    # first layer\n    x = Conv2D(filters=n_filters, kernel_size=(kernel_size, kernel_size), kernel_initializer=\"he_normal\",\n               padding=\"same\")(input_tensor)\n    if batchnorm:\n        x = BatchNormalization()(x)\n    x = Activation(\"relu\")(x)\n    # second layer\n    x = Conv2D(filters=n_filters, kernel_size=(kernel_size, kernel_size), kernel_initializer=\"he_normal\",\n               padding=\"same\")(x)\n    if batchnorm:\n        x = BatchNormalization()(x)\n    x = Activation(\"relu\")(x)\n    return x","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"3a3c7d69-282c-42e6-8063-6e3e99e11663","_cell_guid":"73966f94-39e6-4631-b79f-efa175b01a37","trusted":true},"cell_type":"code","source":"def get_unet(input_img, n_filters=16, dropout=0.5, batchnorm=True):\n    # contracting path\n    c1 = conv2d_block(input_img, n_filters=n_filters*1, kernel_size=3, batchnorm=batchnorm)\n    p1 = MaxPooling2D((2, 2)) (c1)\n    p1 = Dropout(dropout*0.5)(p1)\n\n    c2 = conv2d_block(p1, n_filters=n_filters*2, kernel_size=3, batchnorm=batchnorm)\n    p2 = MaxPooling2D((2, 2)) (c2)\n    p2 = Dropout(dropout)(p2)\n\n    c3 = conv2d_block(p2, n_filters=n_filters*4, kernel_size=3, batchnorm=batchnorm)\n    p3 = MaxPooling2D((2, 2)) (c3)\n    p3 = Dropout(dropout)(p3)\n\n    c4 = conv2d_block(p3, n_filters=n_filters*8, kernel_size=3, batchnorm=batchnorm)\n    p4 = MaxPooling2D(pool_size=(2, 2)) (c4)\n    p4 = Dropout(dropout)(p4)\n    \n    c5 = conv2d_block(p4, n_filters=n_filters*16, kernel_size=3, batchnorm=batchnorm)\n    \n    # expansive path\n    u6 = Conv2DTranspose(n_filters*8, (3, 3), strides=(2, 2), padding='same') (c5)\n    u6 = concatenate([u6, c4])\n    u6 = Dropout(dropout)(u6)\n    c6 = conv2d_block(u6, n_filters=n_filters*8, kernel_size=3, batchnorm=batchnorm)\n\n    u7 = Conv2DTranspose(n_filters*4, (3, 3), strides=(2, 2), padding='same') (c6)\n    u7 = concatenate([u7, c3])\n    u7 = Dropout(dropout)(u7)\n    c7 = conv2d_block(u7, n_filters=n_filters*4, kernel_size=3, batchnorm=batchnorm)\n\n    u8 = Conv2DTranspose(n_filters*2, (3, 3), strides=(2, 2), padding='same') (c7)\n    u8 = concatenate([u8, c2])\n    u8 = Dropout(dropout)(u8)\n    c8 = conv2d_block(u8, n_filters=n_filters*2, kernel_size=3, batchnorm=batchnorm)\n\n    u9 = Conv2DTranspose(n_filters*1, (3, 3), strides=(2, 2), padding='same') (c8)\n    u9 = concatenate([u9, c1], axis=3)\n    u9 = Dropout(dropout)(u9)\n    c9 = conv2d_block(u9, n_filters=n_filters*1, kernel_size=3, batchnorm=batchnorm)\n    \n    outputs = Conv2D(1, (1, 1), activation='sigmoid') (c9)\n    model = Model(inputs=[input_img], outputs=[outputs])\n    return model","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"1f2640f2-64aa-482f-b350-91c31225e934","_cell_guid":"984c313f-5d34-4c32-b747-c21f248d07d1","trusted":true},"cell_type":"code","source":"input_img = Input((224, 224, 1), name='img')\nmodel = get_unet(input_img, n_filters=16, dropout=0.05, batchnorm=True)\n\nmodel.compile(optimizer=Adam(), loss=\"mse\", metrics=[\"mae\"])\nmodel.summary()\ncallbacks = [\n    EarlyStopping(patience=10, verbose=1),\n    ReduceLROnPlateau(factor=0.1, patience=3, min_lr=0.00001, verbose=1),\n    ModelCheckpoint('model-tgs-salt.h5', verbose=1, save_best_only=True, save_weights_only=True)\n]","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"8b95b363-fbaf-4fa6-b36d-bfc657795efb","_cell_guid":"cbc3c13f-430e-4d43-993f-1295927c3f8b","trusted":true},"cell_type":"code","source":"results = model.fit(rgb_batch, rgb_batch, batch_size=32, epochs=50, callbacks=callbacks)","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"41eb0a5a-501f-4d09-a4e7-ef8480418072","_cell_guid":"3476c511-bc21-4ef3-84cc-6dc60596fbab","trusted":true},"cell_type":"code","source":"import keras\nlayer_model = keras.Model(inputs=model.input,outputs=model.output)\n\npres=layer_model.predict(res.reshape(1,224,224,1))\nplt.figure()\nplt.pcolor(res)\nplt.colorbar()\nplt.figure()\nplt.pcolor(pres.reshape(224,224))\nplt.colorbar()","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"0e9b4016-20e9-4af1-93af-62e250198e7e","_cell_guid":"cb9c4704-ca2f-49fe-899d-a7ea01d57dc0","trusted":true},"cell_type":"code","source":"import keras\nintermediate_layer_model = keras.Model(inputs=model.input,outputs=model.get_layer('max_pooling2d_3').output)\nintermediate_output = intermediate_layer_model.predict(rgb_batch)\n\nblock4_pool_features=intermediate_output.reshape(intermediate_output.shape[0],-1)","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"3e188325-f567-4a26-8b3d-54395211c82b","_cell_guid":"852c0bf9-efa4-4fac-bd60-4dbe5169fb5b","trusted":true},"cell_type":"code","source":"\"\"\"\nimport numpy as np\nimport tensorflow as tf\nfrom keras.applications.resnet50 import ResNet50\nfrom tensorflow.python.keras.models import Sequential\nfrom tensorflow.python.keras.layers import Dense, GlobalAveragePooling2D, Dropout, Concatenate, Input\nfrom keras import Model\nfrom keras import optimizers\nbase_model = ResNet50(weights='imagenet', include_top=False, input_shape= (224, 224,3),pooling='max')\nblock4_pool_features = base_model.predict(np.array(rgb_batch).reshape(-1,224, 224,3))\nblock4_pool_features.shape\n\"\"\"\nfrom sklearn.manifold import TSNE\n\nX_embedded = TSNE(n_components=3,perplexity=30).fit_transform(block4_pool_features)\nX_embedded.shape\n#plt.scatter(X_embedded[:1241,0],X_embedded[:1241,1],c='b')\n#plt.scatter(X_embedded[1241:,0],X_embedded[1241:,1],c='g')\n\nXi=np.array(Xi)\nXi_test=np.array(Xi_test)\nXi_test_no=np.array(Xi_test_no)\nXi_temp=np.append(Xi,Xi_test,axis=0)\nXi_em2=np.append(Xi_temp,Xi_test_no,axis=0)\nXi_em2[:,2]=Xi_em2[:,2]/100\nXi_em2.shape\nXx=pd.concat([pd.DataFrame(X_embedded),pd.DataFrame(Xi_em2.reshape(-1,6))],axis=1)\nXx.shape\n\nXr=np.array(Xr)\nXr_test=np.array(Xr_test)\nXr_test_no=np.array(Xr_test_no)\nXr_temp=np.append(Xr,Xr_test,axis=0)\nXr_em2=np.append(Xr_temp,Xr_test_no,axis=0)\nXr_em2.shape\nplt.hist(Xr_em2[:,1].astype(float))\n#model = Model(inputs=base_model.input, outputs=base_model.get_layer('block4_pool').output)\n#block4_pool_features = model.predict(np.array(rgb_batch).reshape(-1,224, 224,3))","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"22a1ae4a-37ba-479d-a685-12c76b6fde8d","_cell_guid":"4bbd081f-a1ea-4df1-a916-d6aae3da40e7","trusted":true},"cell_type":"code","source":"datapkg_pidw=dict()\nfor l in range(len(Yt)):\n    pid=Xr_em2[l][0]\n    week=float(Xr_em2[l][1])\n    try:\n        if datapkg_pidw[pid]>week:\n            datapkg_pidw.update({pid:week})\n    except:\n        datapkg_pidw.update({pid:week})\n\n            \ndatapkg_test_pidw=dict()\n\nfor it in range(len(Yt),len(X_embedded))   :\n    pid=Xr_em2[it][0]\n    week=float(Xr_em2[it][1])\n    try:\n        if datapkg_test_pidw[pid]>week:\n            datapkg_test_pidw.update({pid:week})\n    except:\n        datapkg_test_pidw.update({pid:week})","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"b9db4a50-4105-4c97-b748-a26e3f93114c","_cell_guid":"092f68aa-a4bf-4c53-baf5-a0f05af4cc42","trusted":true},"cell_type":"code","source":"datapkg=dict()\nfor l in range(len(Yt)):\n    pid=Xr_em2[l][0]\n    week=Xr_em2[l][1]\n    fw=datapkg_pidw[pid]\n    datapkg.setdefault(pid,[]).append([list(X_embedded[l])+list(Xi_em2[l])+[(float(week)-fw)/500,Yt[l][0]]])\ndatapkg_test=dict()\n\nfor it in range(len(Yt),len(X_embedded))   :\n    pid=Xr_em2[it][0]\n    week=Xr_em2[it][1]\n    fw=datapkg_test_pidw[pid]\n    datapkg_test.setdefault(pid,[]).append([list(X_embedded[it])+list(Xi_em2[it])+[(float(week)-fw)/500]])\n    print(week,fw,(float(week)-fw)/500)","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"5775749a-e425-4e14-9a9b-0094150c1a00","_cell_guid":"cb3798e8-71d8-46fd-8435-d05060cf3949","trusted":true},"cell_type":"code","source":"import numpy as np\nfrom matplotlib import pyplot as plt\n\nfrom sklearn.gaussian_process import GaussianProcessRegressor\nfrom sklearn.gaussian_process.kernels import RBF, ConstantKernel as C\ndef tgp(Xx, Yt):\n    kernel = C(1.0, (1e-3, 1e3)) * RBF(1, (1e-2, 1e2))\n    gp = GaussianProcessRegressor(kernel=kernel)\n\n    # Fit to data using Maximum Likelihood Estimation of the parameters\n    gp.fit(Xx, Yt)\n    ppp=gp.predict(Xx)\n    return gp,ppp\ngpd=dict()\nfor pid in datapkg.keys():\n    iii=np.array(datapkg[pid]).reshape(-1,11)\n    ix=iii[:,:10]\n    iy=iii[:,10]\n\n    igp,ppp=tgp(ix, iy)\n    if np.mean(100*(abs(ppp-iy)/iy))<=0.5:\n\n        gpd.setdefault(pid,[]).append(igp)\n    else:\n        print(pid,'out')\n\npdd=dict()\nfor ppid in  datapkg_test.keys():\n    iii2=np.array(datapkg_test[ppid]).reshape(-1,10)\n    for ie in iii2:\n        fw=datapkg_test_pidw[ppid]\n        pdk=ppid+'_-'+str(int(ie[-1]*500+fw))\n        allpd=[]\n        for mid in gpd.keys():\n            gp=gpd[mid][0]\n            y_pred, sigma = gp.predict([ie], return_std=True)\n            y_pred_t11, sigma_t11 = gp.predict([np.array(list(ie[:3]*1.1)+list(ie[3:]))], return_std=True)\n            y_pred_t09, sigma_t09 = gp.predict([np.array(list(ie[:3]*0.9)+list(ie[3:]))], return_std=True)\n            if not y_pred_t11!=y_pred or y_pred_t09!=y_pred:\n                allpd.append([y_pred[0],sigma[0]])\n        \n        pdd.setdefault(pdk,[]).append(allpd)\npdd.keys()\npdd['ID00426637202313170790466_-402']\nimport pickle\n\nfile = open('/kaggle/working/op.pkl', 'wb')\npickle.dump(pdd, file)\nfile.close()\n\nfinaloutput=[]\nfor ik in np.sort(list(pdd.keys())):\n\n    pmean=np.array(np.array(pdd[ik])[0])[:,0]\n    \n    \n    _f=np.array(np.array(pdd[ik])[0])[pmean>=1000,:]\n\n    bins = range(0, max(Yt)[0]+3000,3000)\n    digitized = np.digitize(_f[:,0], bins)\n    counts = np.bincount(digitized)\n\n    ll=np.array(bins)[np.argmax(counts)]-3000\n    ul=np.array(bins)[np.argmax(counts)]\n    finaloutput.append([ik]+list(np.mean(_f[(_f[:,0]>=ll) & (_f[:,0]<=ul),:],axis=0)))\n    \"\"\"\n    plt.figure()\n    plt.hist(_f[:,0])\n    \"\"\"\nopdf=pd.DataFrame(finaloutput)    \nopdf.columns=['Patient_Week','FVC','Confidence']\nopdf.to_csv('/kaggle/working/submission.csv', index=False)\n\"\"\"\n\n# Make the prediction on the meshed x-axis (ask for MSE as well)\ny_pred, sigma = gp.predict(Xx, return_std=True)\nplt.plot(Yt)\nplt.plot(y_pred)\nplt.ylim([0,1])\nscaler.inverse_transform(y_pred)\n\"\"\"\n\n\n\n\n\n\n\n\n\n\n\n\n\"\"\"\n# Plot the function, the prediction and the 95% confidence interval based on\n# the MSE\nplt.figure()\n#plt.plot(x, f(x), 'r:', label=r'$f(x) = x\\,\\sin(x)$')\nplt.plot(X, yt, 'r.', markersize=10, label='Observations')\nplt.plot(Xx, y_pred, 'b-', label='Prediction')\nplt.fill(np.concatenate([x, x[::-1]]),\n         np.concatenate([y_pred - 1.9600 * sigma,\n                        (y_pred + 1.9600 * sigma)[::-1]]),\n         alpha=.5, fc='b', ec='None', label='95% confidence interval')\nplt.xlabel('$x$')\nplt.ylabel('$f(x)$')\nplt.ylim(-10, 20)\nplt.legend(loc='upper left')\n\"\"\"","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"d47dca63-47cb-4f7a-853a-0e96993d6325","_cell_guid":"332ddce1-f4a9-4e82-b0b6-18cd69972fc9","trusted":true},"cell_type":"code","source":"\"\"\"\n\nimport tensorflow as tf\nfrom keras.applications.resnet50 import ResNet50\nfrom tensorflow.python.keras.models import Sequential\nfrom tensorflow.python.keras.layers import Dense, GlobalAveragePooling2D, Dropout, Concatenate, Input\nfrom keras import Model\nfrom keras import optimizers\nbase_model = ResNet50(weights='imagenet', include_top=False, input_shape= (224, 224,3))\n#base_model.layers.pop()\n#base_model.outputs = []\n#x = base_model.output\n#base_model.summary()\nx = base_model.layers[-2].output\nx = GlobalAveragePooling2D()(x)\n###x = Dropout(0.7)(x)\nprepredictions = Dense(128, activation= 'relu')(x)\ninp2 = Input(shape=(6,))\nconatenated = Concatenate(axis=1)([prepredictions, inp2])\npredictions = Dense(154, activation= 'softmax')(conatenated)\nmodel = Model(inputs = [ base_model.input, inp2], outputs = predictions)\nmodel.summary()\nAdam=optimizers.Adam\ncallback = tf.keras.callbacks.EarlyStopping(monitor='val_accuracy', patience=3, mode='max')\nmodel.compile(optimizer=Adam(lr=5e-4, decay=5e-4 / 40) ,loss='categorical_crossentropy', metrics=['categorical_crossentropy'])\nhistory = model.fit([np.array(rgb_batch).reshape(-1,224, 224,3),np.array(Xi).reshape(-1,6)], np.array(y), batch_size=16, epochs=40,validation_split=0.1, callbacks=[callback])\n\"\"\"","execution_count":null,"outputs":[]},{"metadata":{"_uuid":"33397316-ded1-4aa2-ba1b-ca2cdc9dba02","_cell_guid":"f49151f4-c99a-4021-bc4f-67e04371301e","trusted":true},"cell_type":"code","source":"\"\"\"\nrgb_batch_test = np.repeat(np.array(X_test)[..., np.newaxis], 3, -1)\n\nYp=model.predict([np.array(rgb_batch_test).reshape(-1,224, 224,3),np.array(Xi_test).reshape(-1,6)],batch_size=16)\nYp_trans=np.argmax(Yp,axis=1)\n\ndigitized_test = np.digitize(Y_test, bins)\nprint(digitized_test,Yp_trans)\n\"\"\"","execution_count":null,"outputs":[]}],"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":4,"nbformat_minor":4}