04_d__FER_SVM_LBP_PCA.ipynb

{"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":"# Setup","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19"}},{"cell_type":"code","source":"# setup to see the execution time in each cell\n\n!pip install ipython-autotime\n!pip install directory_structure\n# !pip install wandb\n%load_ext autotime","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:03.171341Z","iopub.execute_input":"2022-05-09T04:57:03.172087Z","iopub.status.idle":"2022-05-09T04:57:21.693461Z","shell.execute_reply.started":"2022-05-09T04:57:03.172048Z","shell.execute_reply":"2022-05-09T04:57:21.692625Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import pandas as pd\nimport os\nimport glob\nimport PIL\nfrom PIL import Image\nimport numpy as np\nimport cupy as cp\nimport matplotlib.pyplot as plt\n# import wandb","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:21.695809Z","iopub.execute_input":"2022-05-09T04:57:21.696088Z","iopub.status.idle":"2022-05-09T04:57:21.701898Z","shell.execute_reply.started":"2022-05-09T04:57:21.696052Z","shell.execute_reply":"2022-05-09T04:57:21.700881Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"DRIVER_ROOT_DIR = \"/kaggle/input/\" # HC Directory\n# DRIVER_ROOT_DIR = \"/content/drive/MyDrive/DTSC 870/Code\" #MT Directory\n\n# no augmentation\n# DATASET_02_TRAIN = DRIVER_ROOT_DIR + \"fer2013/02_FER/train\"\n# DATASET_02_TEST = DRIVER_ROOT_DIR + \"fer2013/02_FER/test\"\n\n# augmentation v.1\n# DATASET_02_TRAIN = DRIVER_ROOT_DIR + \"fer2013-aug/Aug_train\"\n# DATASET_02_TEST = DRIVER_ROOT_DIR + \"fer2013-aug-test/Aug_test\"\n\n# augmentation v.2\nDATASET_02_TRAIN = DRIVER_ROOT_DIR + \"fer2013-aug-train-2/Aug_train_2\"\nDATASET_02_TEST = DRIVER_ROOT_DIR + \"fer2013-aug-test-2/Aug_test_2\"\n# DATASET_02_TEST = DRIVER_ROOT_DIR + \"d/datasets/huihenrychen/fer2013-aug-test-2/Aug_test_2\"","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:21.703236Z","iopub.execute_input":"2022-05-09T04:57:21.703479Z","iopub.status.idle":"2022-05-09T04:57:21.713028Z","shell.execute_reply.started":"2022-05-09T04:57:21.703447Z","shell.execute_reply":"2022-05-09T04:57:21.712239Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Utility Functions","metadata":{}},{"cell_type":"code","source":"def get_data_df(dir):\n    # modified code from: https://www.kaggle.com/namgalielei/simple-load-images-and-count-number-of-each-class\n\n    train_df = pd.DataFrame()\n\n    trainset = glob.glob(dir)\n\n    train_df['file'] = [img.split(\"/\")[-1] for img in trainset]\n    train_df['class'] = [img.split(\"/\")[-2] for img in trainset]\n\n    return train_df","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:21.715302Z","iopub.execute_input":"2022-05-09T04:57:21.715883Z","iopub.status.idle":"2022-05-09T04:57:21.72455Z","shell.execute_reply.started":"2022-05-09T04:57:21.715847Z","shell.execute_reply":"2022-05-09T04:57:21.723695Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def generate_set(df, dir, classes_):\n\n    # new_df = pd.DataFrame()\n    pixels = []\n    class_ = []\n\n    # trainset = glob.glob(dir)\n    for i in range(len(df.index)):\n        # get the absolute img path\n        # e.g., Brain_tumor_images/<train or test>/<class label>/<file name>\n        path = dir + \"/\" + df.iloc[i][\"class\"] + \"/\" +df.iloc[i][\"file\"]\n        # print(img)\n        img = Image.open(path)\n        # print(\"Img: {} \\tClass: {}\".format(np.array(img).flatten(), df.iloc[i][\"class\"]))\n        pixels.append(cp.asnumpy(cp.array(img)).flatten())\n        # pixels.append(np.array(img))\n\n        # y_true encoding here\n        class_.append(classes_.index(df.iloc[i][\"class\"]))\n\n        # end loop here\n\n    # return train_df\n    return pixels, class_","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:21.72615Z","iopub.execute_input":"2022-05-09T04:57:21.726593Z","iopub.status.idle":"2022-05-09T04:57:21.735294Z","shell.execute_reply.started":"2022-05-09T04:57:21.726556Z","shell.execute_reply":"2022-05-09T04:57:21.734388Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Generate the train and test sets","metadata":{}},{"cell_type":"code","source":"fer_df_train = get_data_df(DATASET_02_TRAIN+\"/*/*.jpg\")\nfer_df_test = get_data_df(DATASET_02_TEST+\"/*/*.jpg\")","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:21.736535Z","iopub.execute_input":"2022-05-09T04:57:21.736971Z","iopub.status.idle":"2022-05-09T04:57:24.113759Z","shell.execute_reply.started":"2022-05-09T04:57:21.736934Z","shell.execute_reply":"2022-05-09T04:57:24.112951Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"classes = fer_df_train[\"class\"].unique().tolist()\n# classes\n# classes.index(\"surprise\")","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:24.115004Z","iopub.execute_input":"2022-05-09T04:57:24.115284Z","iopub.status.idle":"2022-05-09T04:57:24.123795Z","shell.execute_reply.started":"2022-05-09T04:57:24.115247Z","shell.execute_reply":"2022-05-09T04:57:24.122907Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"x_train, y_train = generate_set(fer_df_train, DATASET_02_TRAIN, classes)\nx_test, y_test = generate_set(fer_df_test, DATASET_02_TEST, classes)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:57:24.125342Z","iopub.execute_input":"2022-05-09T04:57:24.125665Z","iopub.status.idle":"2022-05-09T05:01:38.657714Z","shell.execute_reply.started":"2022-05-09T04:57:24.125629Z","shell.execute_reply":"2022-05-09T05:01:38.656871Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### EDA","metadata":{}},{"cell_type":"code","source":"fer_df_test.sample(10)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.658866Z","iopub.execute_input":"2022-05-09T05:01:38.659101Z","iopub.status.idle":"2022-05-09T05:01:38.67579Z","shell.execute_reply.started":"2022-05-09T05:01:38.659068Z","shell.execute_reply":"2022-05-09T05:01:38.675078Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# fer_df_train.shape","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.679085Z","iopub.execute_input":"2022-05-09T05:01:38.679634Z","iopub.status.idle":"2022-05-09T05:01:38.68611Z","shell.execute_reply.started":"2022-05-09T05:01:38.679606Z","shell.execute_reply":"2022-05-09T05:01:38.684701Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"x_train_df = pd.DataFrame()\nx_train_df['class'] = y_train\n\n# x_train_df","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.687685Z","iopub.execute_input":"2022-05-09T05:01:38.688107Z","iopub.status.idle":"2022-05-09T05:01:38.730718Z","shell.execute_reply.started":"2022-05-09T05:01:38.688068Z","shell.execute_reply":"2022-05-09T05:01:38.730005Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# x_train_df[\"class\"].hist()","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.731867Z","iopub.execute_input":"2022-05-09T05:01:38.732105Z","iopub.status.idle":"2022-05-09T05:01:38.737969Z","shell.execute_reply.started":"2022-05-09T05:01:38.732072Z","shell.execute_reply":"2022-05-09T05:01:38.737093Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# print(np.array(x_train).shape)\n# print(np.array(x_test).shape)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.739438Z","iopub.execute_input":"2022-05-09T05:01:38.739855Z","iopub.status.idle":"2022-05-09T05:01:38.744582Z","shell.execute_reply.started":"2022-05-09T05:01:38.739821Z","shell.execute_reply":"2022-05-09T05:01:38.743765Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Feature Engineering","metadata":{}},{"cell_type":"code","source":"# from sklearn.decomposition import PCA\nfrom cuml.decomposition import PCA\nimport matplotlib.pyplot as plt\nimport plotly.express as px\nfrom sklearn.preprocessing import StandardScaler\nimport cv2\nfrom skimage.feature import local_binary_pattern\nfrom plotly.subplots import make_subplots\nimport plotly.graph_objects as go\nfrom matplotlib.pyplot import figure","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.746199Z","iopub.execute_input":"2022-05-09T05:01:38.74664Z","iopub.status.idle":"2022-05-09T05:01:38.754004Z","shell.execute_reply.started":"2022-05-09T05:01:38.746605Z","shell.execute_reply":"2022-05-09T05:01:38.752742Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## LBP & PCA","metadata":{}},{"cell_type":"code","source":"def generate_set2(df, dir, p, r, classes_, method_=\"uniform\"):\n  # Generate train and test sets through OpenCV with grayscale in all pixels\n\n    pixels = []\n    class_ = []\n    eps = 1e-7 # 0.0000001\n\n    # trainset = glob.glob(dir)\n    for i in range(len(df.index)):\n        # get the absolute img path\n        # e.g., Brain_tumor_images/<train or test>/<class label>/<file name>\n        path = dir + \"/\" + df.iloc[i][\"class\"] + \"/\" +df.iloc[i][\"file\"]\n        img = cv2.imread(path)\n\n        # convert the read img into grayscale\n        img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)\n        lbp_img = local_binary_pattern(img_gray, p, r, method=method_)\n        # pixels.append(np.array(lbp_img).flatten())\n\n        # compute the histogram\n        (hist, _) = np.histogram(\n            lbp_img.ravel(),\n            bins=np.arange(0, p + 3),\n            range=(0, p + 2)\n          )\n\n        hist = hist.astype(\"float\")\n        hist /= (hist.sum() + eps)\n        pixels.append(hist)\n\n        # y_true encoding here\n        class_.append(classes_.index(df.iloc[i][\"class\"]))\n        # end loop\n        \n    return pixels, class_","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.755026Z","iopub.execute_input":"2022-05-09T05:01:38.755256Z","iopub.status.idle":"2022-05-09T05:01:38.76854Z","shell.execute_reply.started":"2022-05-09T05:01:38.755224Z","shell.execute_reply":"2022-05-09T05:01:38.767705Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"selected_exp_var = [0.7, 0.8, 0.9, 0.95, 0.97, 0.99, 1.0]\n\n# none img augmentation\nselected_num_comp = [13, 32, 104, 256, 425, 904, 2304]\n\n# img augmentation\nselected_num_comp_aug = [14, 32, 108, 269, 446, 940, 2265]\n\n# combination of p and r values:\npr_list = [\n  {\"p\": 8, \"r\": 1},\n  {\"p\": 16, \"r\": 2},\n  {\"p\": 24, \"r\": 3},\n]","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.76964Z","iopub.execute_input":"2022-05-09T05:01:38.77047Z","iopub.status.idle":"2022-05-09T05:01:38.777837Z","shell.execute_reply.started":"2022-05-09T05:01:38.770433Z","shell.execute_reply":"2022-05-09T05:01:38.776977Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pr_index = 2\n\nx_train_fs, y_train_fs = generate_set2(fer_df_train, DATASET_02_TRAIN, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\nx_test_fs, y_test_fs = generate_set2(fer_df_test, DATASET_02_TEST, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:01:38.779026Z","iopub.execute_input":"2022-05-09T05:01:38.779345Z","iopub.status.idle":"2022-05-09T05:04:38.860289Z","shell.execute_reply.started":"2022-05-09T05:01:38.779309Z","shell.execute_reply":"2022-05-09T05:04:38.859525Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pca = PCA()\nx_train_pca = pca.fit_transform(np.array(x_train_fs))\nx_test_pca = pca.transform(np.array(x_test_fs))","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:38.861672Z","iopub.execute_input":"2022-05-09T05:04:38.862026Z","iopub.status.idle":"2022-05-09T05:04:38.940465Z","shell.execute_reply.started":"2022-05-09T05:04:38.861998Z","shell.execute_reply":"2022-05-09T05:04:38.939732Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"exp_var_cumul = np.cumsum(pca.explained_variance_ratio_)\nexp_var_cumul_round = np.round_(exp_var_cumul, decimals = 4)\n\nnum_comp = range(1, exp_var_cumul_round.shape[0] + 1)\n\nprint(exp_var_cumul_round)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:38.94167Z","iopub.execute_input":"2022-05-09T05:04:38.942363Z","iopub.status.idle":"2022-05-09T05:04:38.951581Z","shell.execute_reply.started":"2022-05-09T05:04:38.942326Z","shell.execute_reply":"2022-05-09T05:04:38.950875Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### PCA Visualization","metadata":{}},{"cell_type":"code","source":"labels = {\n    str(i): f\"PC {i+1} ({var:.1f}%)\"\n    for i, var in enumerate(pca.explained_variance_ratio_ * 100)\n}\n\nfig = px.scatter_matrix(\n    x_train_pca,\n    # x_train_df,\n    labels=labels,\n    # dimensions=(0, 1),\n    dimensions=range(4),\n    # color=x_train_df['class'],\n)\nfig.update_traces(diagonal_visible=False)\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:38.953971Z","iopub.execute_input":"2022-05-09T05:04:38.954373Z","iopub.status.idle":"2022-05-09T05:04:39.128813Z","shell.execute_reply.started":"2022-05-09T05:04:38.954345Z","shell.execute_reply":"2022-05-09T05:04:39.12806Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# get a list of # of components\nnum_comp = range(1, exp_var_cumul_round.shape[0] + 1)\n\nper_var = np.round(pca.explained_variance_ratio_* 100, decimals=1)\n\nfig = px.line(\n    x=num_comp,\n    y=exp_var_cumul_round,\n    labels={\"x\": \"# Components\", \"y\": \"Cumulative Explained Variance\"},\n    title = \"# of components V.S. variance\",\n    # markers=True\n)\n\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.130099Z","iopub.execute_input":"2022-05-09T05:04:39.130621Z","iopub.status.idle":"2022-05-09T05:04:39.208346Z","shell.execute_reply.started":"2022-05-09T05:04:39.130579Z","shell.execute_reply":"2022-05-09T05:04:39.207732Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# cumsum_var_243 = [0.8618, 0.9531, 0.9796, 0.9907, 1.0]\ncumsum_var_243_aug = [0.8611, 0.9525, 0.9793, 0.9905, 1.0]\n\nselectd_num_comp243 = [2, 7, 13, 18, 26]","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.209361Z","iopub.execute_input":"2022-05-09T05:04:39.210083Z","iopub.status.idle":"2022-05-09T05:04:39.21494Z","shell.execute_reply.started":"2022-05-09T05:04:39.210045Z","shell.execute_reply":"2022-05-09T05:04:39.21416Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"per_var = np.round(pca.explained_variance_ratio_* 100, decimals=1)\n# labels = ['PC' + str(x) for x in range(1, len(per_var)+1)]\n\nfig = px.bar(\n    x=range(1,len(per_var)+1), \n    y=per_var, \n    title='Scree Plot of # of components and % of explained variance',\n    labels={\"x\": \"# Components\", \"y\": \"% of Explained Variance\"},\n)\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.215913Z","iopub.execute_input":"2022-05-09T05:04:39.216575Z","iopub.status.idle":"2022-05-09T05:04:39.287255Z","shell.execute_reply.started":"2022-05-09T05:04:39.216542Z","shell.execute_reply":"2022-05-09T05:04:39.286575Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Model","metadata":{}},{"cell_type":"markdown","source":"## Setup","metadata":{}},{"cell_type":"code","source":"from sklearn.metrics import balanced_accuracy_score, accuracy_score\nfrom numpy import mean\n\nfrom cuml.svm import SVC","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.288538Z","iopub.execute_input":"2022-05-09T05:04:39.28896Z","iopub.status.idle":"2022-05-09T05:04:39.293746Z","shell.execute_reply.started":"2022-05-09T05:04:39.288923Z","shell.execute_reply":"2022-05-09T05:04:39.293028Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def generate_PCA_set(comp, x_train, x_test):\n    pca = PCA(n_components=comp)\n    __x_train__ = pca.fit_transform(x_train)\n    __x_test__ = pca.transform(x_test)\n    return __x_train__, __x_test__","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.295316Z","iopub.execute_input":"2022-05-09T05:04:39.295725Z","iopub.status.idle":"2022-05-09T05:04:39.304014Z","shell.execute_reply.started":"2022-05-09T05:04:39.295691Z","shell.execute_reply":"2022-05-09T05:04:39.303241Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_model():\n    return SVC(random_state=1, multiclass_strategy=\"ovr\")\n\ndef get_model2():\n    return OneVsRestClassifier(SVC(random_state=1, multiclass_strategy=\"ovr\"))","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.305432Z","iopub.execute_input":"2022-05-09T05:04:39.306536Z","iopub.status.idle":"2022-05-09T05:04:39.312415Z","shell.execute_reply.started":"2022-05-09T05:04:39.306446Z","shell.execute_reply":"2022-05-09T05:04:39.311732Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def feature_scale(_x_train_, _x_test_):\n    sc = StandardScaler()\n    \n    train_sc = sc.fit_transform(_x_train_)\n    test_sc = sc.transform(_x_test_)\n    \n    return train_sc, test_sc\n\n\ndef feature_scale2(_x_train_):\n    sc = StandardScaler()\n    \n    train_sc2 = sc.fit_transform(_x_train_)\n    \n    return train_sc2","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.31348Z","iopub.execute_input":"2022-05-09T05:04:39.314214Z","iopub.status.idle":"2022-05-09T05:04:39.323384Z","shell.execute_reply.started":"2022-05-09T05:04:39.314162Z","shell.execute_reply":"2022-05-09T05:04:39.322692Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Train","metadata":{}},{"cell_type":"code","source":"cumsum_var_81 = [0.8341, 0.9512, 0.9801, 0.9946, 1.0]\ncumsum_var_81_aug = [0.8344, 0.9512, 0.9801, 0.9949, 1.0]\nselectd_num_comp81 = [2, 4, 7, 8, 10]\nlbp_exp81 = list()\npr_index = 0\n\nx_train_, y_train_ = generate_set2(fer_df_train, DATASET_02_TRAIN, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\nx_test_, y_test_ = generate_set2(fer_df_test, DATASET_02_TEST, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\n\nfor i in range(len(selectd_num_comp81)):\n    x_train_pca, x_test_pca = generate_PCA_set(selectd_num_comp81[i], np.array(x_train_), np.array(x_test_))\n\n    lbp_exp81.append({\n        \"explained_variance_%\": cumsum_var_81_aug[i],\n        \"p_r\": (pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"]),\n        \"model\": get_model(),\n        \"x_train\": x_train_pca,\n        \"y_train\": y_train_,\n        \"x_test\": x_test_pca,\n        \"y_test\": y_test_,\n    })","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:04:39.327697Z","iopub.execute_input":"2022-05-09T05:04:39.328315Z","iopub.status.idle":"2022-05-09T05:06:51.785605Z","shell.execute_reply.started":"2022-05-09T05:04:39.328281Z","shell.execute_reply":"2022-05-09T05:06:51.78486Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cumsum_var_162 = [0.8335, 0.9501, 0.9779, 0.9899, 1.0]\ncumsum_var_162_aug = [0.8349, 0.9495, 0.9775, 0.9897, 1.0]\n\nselectd_num_comp162 = [2, 6, 10, 13, 18]\nlbp_exp162 = list()\npr_index = 1\n\nx_train_, y_train_ = generate_set2(fer_df_train, DATASET_02_TRAIN, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\nx_test_, y_test_ = generate_set2(fer_df_test, DATASET_02_TEST, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\n\nfor i in range(len(selectd_num_comp162)):    \n    x_train_pca, x_test_pca = generate_PCA_set(selectd_num_comp162[i], np.array(x_train_), np.array(x_test_))\n\n    lbp_exp162.append({\n        \"explained_variance_%\": cumsum_var_162_aug[i],\n        \"p_r\": (pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"]),\n        \"model\": get_model(),\n        \"x_train\": x_train_pca,\n        \"y_train\": y_train_,\n        \"x_test\": x_test_pca,\n        \"y_test\": y_test_,\n    })","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:06:51.786857Z","iopub.execute_input":"2022-05-09T05:06:51.78717Z","iopub.status.idle":"2022-05-09T05:09:33.179134Z","shell.execute_reply.started":"2022-05-09T05:06:51.787133Z","shell.execute_reply":"2022-05-09T05:09:33.178245Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cumsum_var_243 = [0.8618, 0.9531, 0.9796, 0.9907, 1.0]\ncumsum_var_243_aug = [0.8611, 0.9525, 0.9793, 0.9905, 1.0]\n\nselectd_num_comp243 = [2, 7, 13, 18, 26]\nlbp_exp243 = list()\npr_index = 2\n\nx_train_, y_train_ = generate_set2(fer_df_train, DATASET_02_TRAIN, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\nx_test_, y_test_ = generate_set2(fer_df_test, DATASET_02_TEST, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\n\nfor i in range(len(selectd_num_comp243)):\n    x_train_pca, x_test_pca = generate_PCA_set(selectd_num_comp243[i], np.array(x_train_), np.array(x_test_))\n\n    lbp_exp243.append({\n        \"explained_variance_%\": cumsum_var_243_aug[i],\n        \"p_r\": (pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"]),\n        \"model\": get_model(),\n        \"x_train\": x_train_pca,\n        \"y_train\": y_train_,\n        \"x_test\": x_test_pca,\n        \"y_test\": y_test_,\n    })\n    \n\n    \n    \n# x_train_, y_train_ = generate_set2(fer_df_train, DATASET_02_TRAIN, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\n# x_test_, y_test_ = generate_set2(fer_df_test, DATASET_02_TEST, pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"], classes)\n\n# i = -1\n# x_train_pca, x_test_pca = generate_PCA_set(selectd_num_comp243[i], x_train_, x_test_)\n\n# lbp_exp_best = {\n#     \"explained_variance_%\": cumsum_var_243_aug[i],\n#     \"p_r\": (pr_list[pr_index][\"p\"], pr_list[pr_index][\"r\"]),\n#     \"model\": get_model(),\n#     \"x_train\": x_train_pca,\n#     \"y_train\": y_train_,\n#     \"x_test\": x_test_pca,\n#     \"y_test\": y_test_,\n# }","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:09:33.183045Z","iopub.execute_input":"2022-05-09T05:09:33.183319Z","iopub.status.idle":"2022-05-09T05:12:33.067622Z","shell.execute_reply.started":"2022-05-09T05:09:33.183277Z","shell.execute_reply":"2022-05-09T05:12:33.066868Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Experiment Tracking","metadata":{}},{"cell_type":"code","source":"# wandb.init(\n#   project=\"LBP_81_PCA_SVM\",\n#   notes=\"LBP with PCA on 81 for pr value\",\n# )","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:12:33.06877Z","iopub.execute_input":"2022-05-09T05:12:33.069026Z","iopub.status.idle":"2022-05-09T05:12:33.073616Z","shell.execute_reply.started":"2022-05-09T05:12:33.068991Z","shell.execute_reply":"2022-05-09T05:12:33.072657Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.multiclass import OneVsRestClassifier","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:12:33.074863Z","iopub.execute_input":"2022-05-09T05:12:33.075741Z","iopub.status.idle":"2022-05-09T05:12:33.082323Z","shell.execute_reply.started":"2022-05-09T05:12:33.075704Z","shell.execute_reply":"2022-05-09T05:12:33.081532Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def svm_lbp_pca_training(arr_):\n    \n    uw_pcaLBP_acc_result, w_pcaLBP_acc_result = list(), list()\n    \n    for i in range(len(arr_)):\n        # scaling\n        x_train_sc, x_test_sc = feature_scale(arr_[i][\"x_train\"], arr_[i][\"x_test\"])\n\n        x_train_sc, x_test_sc = np.array(x_train_sc), np.array(x_test_sc)\n        y_train = np.array(arr_[i][\"y_train\"], dtype=np.float32)\n\n        # train the model\n        # model = SVC(C=1.0, kernel='poly', gamma=1.0, degree=3, random_state=1, multiclass_strategy=\"ovr\")\n        model = arr_[i][\"model\"]\n        # model = SVC(decision_function_shape=\"ovr\", kernel=\"rbf\", random_state=13)\n        model.fit(x_train_sc, y_train)\n\n        # predict the train set\n        y_trainHat = model.predict(x_train_sc)\n\n        # predict the train set\n        y_testHat = model.predict(x_test_sc)\n\n        # compute the unweighted accuracy\n        uw_acc = balanced_accuracy_score(arr_[i][\"y_test\"], y_testHat)\n        uw_pcaLBP_acc_result.append(uw_acc)\n        print(\"The unweighted accuracy: {}\".format(uw_acc))\n\n        # compute the weighted accuracy\n        w_acc = accuracy_score(arr_[i][\"y_test\"], y_testHat)\n        print(\"The weighted accuracy: {}\".format(w_acc))\n        w_pcaLBP_acc_result.append(w_acc)\n        # end loop\n        \n    return uw_pcaLBP_acc_result, w_pcaLBP_acc_result\n\n\n\ndef svm_lbp_pca_training2(arr_):\n    \n    # uw_pcaLBP_acc_result, w_pcaLBP_acc_result = list(), list()\n    \n    # for i in range(len(arr_)):\n    # scaling\n    x_train_sc, x_test_sc = feature_scale(arr_[\"x_train\"], arr_[\"x_test\"])\n\n    x_train_sc, x_test_sc = np.array(x_train_sc), np.array(x_test_sc)\n    y_train = np.array(arr_[\"y_train\"], dtype=np.float32)\n\n    # train the model\n    # model = SVC(C=1.0, kernel='poly', gamma=1.0, degree=3, random_state=1, multiclass_strategy=\"ovr\")\n    model = arr_[\"model\"]\n    # model = SVC(decision_function_shape=\"ovr\", kernel=\"rbf\", random_state=13)\n    model.fit(x_train_sc, y_train)\n\n    # predict the train set\n    y_trainHat = model.predict(x_train_sc)\n\n    # predict the train set\n    y_testHat = model.predict(x_test_sc)\n\n    # compute the unweighted accuracy\n    uw_acc = balanced_accuracy_score(arr_[\"y_test\"], y_testHat)\n    # uw_pcaLBP_acc_result.append(uw_acc)\n    print(\"The unweighted accuracy: {}\".format(uw_acc))\n\n    # compute the weighted accuracy\n    w_acc = accuracy_score(arr_[\"y_test\"], y_testHat)\n    print(\"The weighted accuracy: {}\".format(w_acc))\n    # w_pcaLBP_acc_result.append(w_acc)\n    # end loop\n\n    # return uw_pcaLBP_acc_result, w_pcaLBP_acc_result","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:12:33.083723Z","iopub.execute_input":"2022-05-09T05:12:33.084482Z","iopub.status.idle":"2022-05-09T05:12:33.098716Z","shell.execute_reply.started":"2022-05-09T05:12:33.084445Z","shell.execute_reply":"2022-05-09T05:12:33.097783Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"uw_acc81, w_acc81 = svm_lbp_pca_training(lbp_exp81)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:12:33.100051Z","iopub.execute_input":"2022-05-09T05:12:33.10051Z","iopub.status.idle":"2022-05-09T05:19:47.468004Z","shell.execute_reply.started":"2022-05-09T05:12:33.100476Z","shell.execute_reply":"2022-05-09T05:19:47.467244Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"uw_acc162, w_acc162 = svm_lbp_pca_training(lbp_exp162)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:19:47.469172Z","iopub.execute_input":"2022-05-09T05:19:47.46991Z","iopub.status.idle":"2022-05-09T05:25:17.122206Z","shell.execute_reply.started":"2022-05-09T05:19:47.469872Z","shell.execute_reply":"2022-05-09T05:25:17.120673Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"uw_acc243, w_acc243 = svm_lbp_pca_training(lbp_exp243)\n\n# svm_lbp_pca_training2(lbp_exp_best)","metadata":{"execution":{"iopub.status.busy":"2022-05-09T05:25:17.12388Z","iopub.execute_input":"2022-05-09T05:25:17.124131Z","iopub.status.idle":"2022-05-09T05:29:45.621921Z","shell.execute_reply.started":"2022-05-09T05:25:17.124097Z","shell.execute_reply":"2022-05-09T05:29:45.621224Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import plotly.graph_objects as go\n\n# x_label = zip(cumsum_var_81, selectd_num_comp81)\n\nfig = go.Figure()\nfig.add_trace(go.Line(\n    x=cumsum_var_81,\n    y=uw_acc81,\n    mode='lines',\n    name='162 Unweighted Test Accuracy',\n    marker=dict(\n        color='red',\n        size=10\n    ))\n)\n\nfig.add_trace(go.Line(\n    x=cumsum_var_81,\n    y=w_acc81,\n    mode='lines',\n    name='162 Weighted Test Accuracy',\n    marker=dict(\n        color='green',\n        size=10\n    ))\n)\n\n\nfig.update_layout(\n    title=\"Tuned parameter SVM with LBP (8,1) & PCA\",\n    xaxis_title=\"% of Explained Variance\",\n    yaxis_title=\"Accuracy (%)\"\n)\nfig.update_traces(mode='markers+lines')\n\n# wandb.log({\"LBP_162_PCA_SVM\": fig})\n\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:23:48.443241Z","iopub.execute_input":"2022-05-09T04:23:48.443784Z","iopub.status.idle":"2022-05-09T04:23:48.463388Z","shell.execute_reply.started":"2022-05-09T04:23:48.443744Z","shell.execute_reply":"2022-05-09T04:23:48.462693Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import plotly.graph_objects as go\n\n# x_label = zip(cumsum_var_81, selectd_num_comp81)\n\nfig = go.Figure()\nfig.add_trace(go.Line(\n    x=cumsum_var_81,\n    y=uw_acc162,\n    mode='lines',\n    name='162 Unweighted Test Accuracy',\n    marker=dict(\n        color='red',\n        size=10\n    ))\n)\n\nfig.add_trace(go.Line(\n    x=cumsum_var_81,\n    y=w_acc162,\n    mode='lines',\n    name='162 Weighted Test Accuracy',\n    marker=dict(\n        color='green',\n        size=10\n    ))\n)\n\n\nfig.update_layout(\n    title=\"Tuned parameter SVM with LBP (16,2) & PCA\",\n    xaxis_title=\"% of Explained Variance\",\n    yaxis_title=\"Accuracy (%)\"\n)\nfig.update_traces(mode='markers+lines')\n\n# wandb.log({\"LBP_162_PCA_SVM\": fig})\n\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:23:48.464544Z","iopub.execute_input":"2022-05-09T04:23:48.465274Z","iopub.status.idle":"2022-05-09T04:23:48.483535Z","shell.execute_reply.started":"2022-05-09T04:23:48.465239Z","shell.execute_reply":"2022-05-09T04:23:48.482635Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import plotly.graph_objects as go\n\n# x_label = zip(cumsum_var_81, selectd_num_comp81)\n\nfig = go.Figure()\nfig.add_trace(go.Line(\n    x=cumsum_var_81,\n    y=uw_acc243,\n    mode='lines',\n    name='162 Unweighted Test Accuracy',\n    marker=dict(\n        color='red',\n        size=10\n    ))\n)\n\nfig.add_trace(go.Line(\n    x=cumsum_var_81,\n    y=w_acc243,\n    mode='lines',\n    name='162 Weighted Test Accuracy',\n    marker=dict(\n        color='green',\n        size=10\n    ))\n)\n\n\nfig.update_layout(\n    title=\"Tuned parameter SVM with LBP (24,3) & PCA\",\n    xaxis_title=\"% of Explained Variance\",\n    yaxis_title=\"Accuracy (%)\"\n)\nfig.update_traces(mode='markers+lines')\n\n# wandb.log({\"LBP_162_PCA_SVM\": fig})\n\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-09T04:23:48.48475Z","iopub.execute_input":"2022-05-09T04:23:48.485045Z","iopub.status.idle":"2022-05-09T04:23:48.502573Z","shell.execute_reply.started":"2022-05-09T04:23:48.48501Z","shell.execute_reply":"2022-05-09T04:23:48.501895Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Analysis","metadata":{}},{"cell_type":"code","source":"from sklearn.metrics import plot_confusion_matrix\nfrom sklearn.metrics import classification_report\nfrom cuml.metrics import confusion_matrix\n\nimport plotly.express as px\n\nfrom sklearn.metrics import roc_curve, auc\nfrom matplotlib.pyplot import figure\nfrom sklearn.preprocessing import LabelBinarizer","metadata":{"execution":{"iopub.status.busy":"2022-04-27T12:12:54.176593Z","iopub.execute_input":"2022-04-27T12:12:54.177362Z","iopub.status.idle":"2022-04-27T12:12:54.187206Z","shell.execute_reply.started":"2022-04-27T12:12:54.17731Z","shell.execute_reply":"2022-04-27T12:12:54.18613Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"best_item = lbp_exp_best\n\nx_train_sc_, x_test_sc_ = feature_scale(best_item[\"x_train\"], best_item[\"x_test\"])\n\nx_train_sc_, x_test_sc_ = np.array(x_train_sc_), np.array(x_test_sc_)\ny_train_ = np.array(best_item[\"y_train\"], dtype=np.float32)\n\ny_pred_ = best_item[\"model\"].predict(x_test_sc_)\ny_pred_ = np.array(y_pred_, dtype=np.int32)\ny_test_ = np.array(best_item[\"y_test\"], dtype=np.int32)","metadata":{"execution":{"iopub.status.busy":"2022-04-27T12:12:54.18868Z","iopub.execute_input":"2022-04-27T12:12:54.189645Z","iopub.status.idle":"2022-04-27T12:12:54.357965Z","shell.execute_reply.started":"2022-04-27T12:12:54.189558Z","shell.execute_reply":"2022-04-27T12:12:54.356965Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"conf_matrix = cp.asnumpy(confusion_matrix(y_test_, y_pred_))\nconfM = conf_matrix\n\nt = \"confusion matrix for SVM on LBP (24, 3) with PCA (\" + str(best_item[\"explained_variance_%\"]) + \" %) var\"\nfig = px.imshow(conf_matrix, text_auto=True, title=t)\nfig.show()","metadata":{"execution":{"iopub.status.busy":"2022-04-27T12:13:57.267798Z","iopub.execute_input":"2022-04-27T12:13:57.268117Z","iopub.status.idle":"2022-04-27T12:13:57.367027Z","shell.execute_reply.started":"2022-04-27T12:13:57.268085Z","shell.execute_reply":"2022-04-27T12:13:57.365991Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"per_class_acc = list()\n\nfor i in range(len(classes)):\n    row_sum = np.array(confM[i]).sum()\n    acc = (conf_matrix[i][i]/row_sum)*100\n    per_class_acc.append(acc)\n    \nfor i in range(len(classes)):\n    print(\"[{}: {} - {}]\".format(i, classes[i], per_class_acc[i]), end=\", \")","metadata":{"execution":{"iopub.status.busy":"2022-04-27T12:14:00.4392Z","iopub.execute_input":"2022-04-27T12:14:00.439836Z","iopub.status.idle":"2022-04-27T12:14:00.449494Z","shell.execute_reply.started":"2022-04-27T12:14:00.4398Z","shell.execute_reply":"2022-04-27T12:14:00.448351Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred2 = np.array(y_pred_).astype(int)\n\nlb = LabelBinarizer()\n\nend_y_pred2 = lb.fit_transform(y_pred2)\nend_y_test = lb.fit_transform(y_test_)","metadata":{"execution":{"iopub.status.busy":"2022-04-27T12:14:01.219179Z","iopub.execute_input":"2022-04-27T12:14:01.219856Z","iopub.status.idle":"2022-04-27T12:14:01.232046Z","shell.execute_reply.started":"2022-04-27T12:14:01.219799Z","shell.execute_reply":"2022-04-27T12:14:01.230789Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Compute ROC curve and ROC area for each class\nfpr = dict()\ntpr = dict()\nroc_auc = dict()\nfor i in range(len(classes)):\n    fpr[i], tpr[i], _ = roc_curve(end_y_test[:,i], end_y_pred2[:, i])\n    roc_auc[i] = auc(fpr[i], tpr[i])\n\n# Compute micro-average ROC curve and ROC area\nfpr[\"micro\"], tpr[\"micro\"], _ = roc_curve(end_y_test.ravel(), end_y_pred2.ravel())\nroc_auc[\"micro\"] = auc(fpr[\"micro\"], tpr[\"micro\"])\n\n# Plot ROC curve\nplt.figure()\nplt.plot(fpr[\"micro\"], tpr[\"micro\"],\n         label='micro-average ROC curve (area = {0:0.2f})'\n               ''.format(roc_auc[\"micro\"]))\nfor i in range(len(classes)):\n    plt.plot(fpr[i], tpr[i], label='ROC curve of class {0} (area = {1:0.2f})'\n                                   ''.format(i, roc_auc[i]))\n\n# figure(figsize=(8, 6), dpi=80)\nplt.plot([0, 1], [0, 1], 'k--')\nplt.xlim([0.0, 1.0])\nplt.ylim([0.0, 1.05])\nplt.xlabel('False Positive Rate')\nplt.ylabel('True Positive Rate')\nplt.title('Some extension of Receiver operating characteristic to multi-class')\nplt.legend(bbox_to_anchor=(1.04,1), borderaxespad=0)\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-04-27T12:14:01.912027Z","iopub.execute_input":"2022-04-27T12:14:01.912509Z","iopub.status.idle":"2022-04-27T12:14:02.427807Z","shell.execute_reply.started":"2022-04-27T12:14:01.912476Z","shell.execute_reply":"2022-04-27T12:14:02.426809Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"Classification report for SVM on LBP (24, 3) with PCA (\" + str(best_item[\"explained_variance_%\"]) + \" %) var\")\nprint(classification_report(y_test, y_pred2, target_names=classes))","metadata":{"execution":{"iopub.status.busy":"2022-04-27T12:15:30.420505Z","iopub.execute_input":"2022-04-27T12:15:30.421323Z","iopub.status.idle":"2022-04-27T12:15:30.46195Z","shell.execute_reply.started":"2022-04-27T12:15:30.421272Z","shell.execute_reply":"2022-04-27T12:15:30.461012Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}