{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"nvidiaTeslaT4","dataSources":[{"sourceId":50160,"databundleVersionId":7921029,"sourceType":"competition"}],"dockerImageVersionId":30699,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import sys\nfrom pathlib import Path\nimport subprocess\nimport os\nimport gc\nfrom glob import glob\n\nimport numpy as np\nimport pandas as pd\nimport polars as pl\nfrom datetime import datetime\nimport seaborn as sns\nimport matplotlib.pyplot as plt\n\nimport warnings\nwarnings.filterwarnings('ignore')\n\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\n\nimport lightgbm as lgb\nfrom sklearn.model_selection import StratifiedGroupKFold\nfrom sklearn.metrics import roc_auc_score \nfrom sklearn.impute import KNNImputer\nfrom sklearn.preprocessing import OrdinalEncoder, StandardScaler, RobustScaler\nfrom sklearn.decomposition import PCA\nfrom sklearn.feature_selection import VarianceThreshold\nfrom sklearn.base import BaseEstimator, RegressorMixin\nfrom imblearn.over_sampling import SMOTE","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2024-06-02T14:36:02.882628Z","iopub.execute_input":"2024-06-02T14:36:02.883471Z","iopub.status.idle":"2024-06-02T14:36:02.890490Z","shell.execute_reply.started":"2024-06-02T14:36:02.883438Z","shell.execute_reply":"2024-06-02T14:36:02.889592Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ROOT            = Path(\"/kaggle/input/home-credit-credit-risk-model-stability\")\n\nTRAIN_DIR       = ROOT / \"parquet_files\" / \"train\"\nTEST_DIR        = ROOT / \"parquet_files\" / \"test\"","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:36:02.892473Z","iopub.execute_input":"2024-06-02T14:36:02.892761Z","iopub.status.idle":"2024-06-02T14:36:02.900395Z","shell.execute_reply.started":"2024-06-02T14:36:02.892738Z","shell.execute_reply":"2024-06-02T14:36:02.899445Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Loading data","metadata":{}},{"cell_type":"code","source":"class Pipeline:\n    def set_table_dtypes(df: pl.DataFrame) -> pl.DataFrame:\n        for col in df.columns:\n            if col in [\"case_id\", \"WEEK_NUM\", \"num_group1\", \"num_group2\"]:\n                df = df.with_columns(pl.col(col).cast(pl.Int64))\n            elif col in [\"date_decision\"]:\n                df = df.with_columns(pl.col(col).cast(pl.Date))\n            elif col[-1] in (\"P\", \"A\"):\n                df = df.with_columns(pl.col(col).cast(pl.Float64))\n            elif col[-1] in (\"M\",):\n                df = df.with_columns(pl.col(col).cast(pl.String))\n            elif col[-1] in (\"D\",):\n                df = df.with_columns(pl.col(col).cast(pl.Date))\n        return df\n\n    def handle_dates(df):\n        for col in df.columns:\n            if col[-1] in (\"D\",):\n                df = df.with_columns(pl.col(col) - pl.col(\"date_decision\"))  #!!?\n                df = df.with_columns(pl.col(col).dt.total_days()) # t - t-1\n        df = df.drop(\"date_decision\", \"MONTH\")\n        return df\n\n    def filter_cols(df):\n        for col in df.columns:\n            if col not in [\"target\", \"case_id\", \"WEEK_NUM\"]:\n                isnull = df[col].is_null().mean()\n                if isnull > 0.7:\n                    df = df.drop(col)\n\n        for col in df.columns:\n            if (col not in [\"target\", \"case_id\", \"WEEK_NUM\"]) & (df[col].dtype == pl.String):\n                freq = df[col].n_unique()\n                if (freq == 1) | (freq > 200):\n                    df = df.drop(col)\n\n        return df\n","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:36:02.901518Z","iopub.execute_input":"2024-06-02T14:36:02.901796Z","iopub.status.idle":"2024-06-02T14:36:02.916423Z","shell.execute_reply.started":"2024-06-02T14:36:02.901772Z","shell.execute_reply":"2024-06-02T14:36:02.915510Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Aggregator:\n    #Please add or subtract features yourself, be aware that too many features will take up too much space.\n    def num_expr(df):\n        cols = [col for col in df.columns if col[-1] in (\"P\", \"A\")]\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n        \n        expr_last = [pl.last(col).alias(f\"last_{col}\") for col in cols]\n        #expr_first = [pl.first(col).alias(f\"first_{col}\") for col in cols]\n        expr_mean = [pl.mean(col).alias(f\"mean_{col}\") for col in cols]\n        return expr_max +expr_last+expr_mean\n    \n    def date_expr(df):\n        cols = [col for col in df.columns if col[-1] in (\"D\")]\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n        #expr_min = [pl.min(col).alias(f\"min_{col}\") for col in cols]\n        expr_last = [pl.last(col).alias(f\"last_{col}\") for col in cols]\n        #expr_first = [pl.first(col).alias(f\"first_{col}\") for col in cols]\n        expr_mean = [pl.mean(col).alias(f\"mean_{col}\") for col in cols]\n        return  expr_max +expr_last+expr_mean\n    \n    def str_expr(df):\n        cols = [col for col in df.columns if col[-1] in (\"M\",)]\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n        #expr_min = [pl.min(col).alias(f\"min_{col}\") for col in cols]\n        expr_last = [pl.last(col).alias(f\"last_{col}\") for col in cols]\n        #expr_first = [pl.first(col).alias(f\"first_{col}\") for col in cols]\n        #expr_count = [pl.count(col).alias(f\"count_{col}\") for col in cols]\n        return  expr_max +expr_last#+expr_count\n    \n    def other_expr(df):\n        cols = [col for col in df.columns if col[-1] in (\"T\", \"L\")]\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n        #expr_min = [pl.min(col).alias(f\"min_{col}\") for col in cols]\n        expr_last = [pl.last(col).alias(f\"last_{col}\") for col in cols]\n        #expr_first = [pl.first(col).alias(f\"first_{col}\") for col in cols]\n        return  expr_max +expr_last\n    \n    def count_expr(df):\n        cols = [col for col in df.columns if \"num_group\" in col]\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols] \n        #expr_min = [pl.min(col).alias(f\"min_{col}\") for col in cols]\n        expr_last = [pl.last(col).alias(f\"last_{col}\") for col in cols]\n        #expr_first = [pl.first(col).alias(f\"first_{col}\") for col in cols]\n        return  expr_max +expr_last\n    \n    def get_exprs(df):\n        exprs = Aggregator.num_expr(df) + \\\n                Aggregator.date_expr(df) + \\\n                Aggregator.str_expr(df) + \\\n                Aggregator.other_expr(df) + \\\n                Aggregator.count_expr(df)\n\n        return exprs\n\ndef read_file(path, depth=None):\n    df = pl.read_parquet(path)\n    df = df.pipe(Pipeline.set_table_dtypes)\n    if depth in [1,2]:\n        df = df.group_by(\"case_id\").agg(Aggregator.get_exprs(df)) \n    return df\n\ndef read_files(regex_path, depth=None):\n    chunks = []\n    \n    for path in glob(str(regex_path)):\n        df = pl.read_parquet(path)\n        df = df.pipe(Pipeline.set_table_dtypes)\n        if depth in [1, 2]:\n            df = df.group_by(\"case_id\").agg(Aggregator.get_exprs(df))\n        chunks.append(df)\n    \n    df = pl.concat(chunks, how=\"vertical_relaxed\")\n    df = df.unique(subset=[\"case_id\"])\n    return df\n\ndef feature_eng(df_base, depth_0, depth_1, depth_2):\n    df_base = (\n        df_base\n        .with_columns(\n            month_decision = pl.col(\"date_decision\").dt.month(),\n            weekday_decision = pl.col(\"date_decision\").dt.weekday(),\n        )\n    )\n    for i, df in enumerate(depth_0 + depth_1 + depth_2):\n        df_base = df_base.join(df, how=\"left\", on=\"case_id\", suffix=f\"_{i}\")\n    df_base = df_base.pipe(Pipeline.handle_dates)\n    return df_base\n\ndef to_pandas(df_data, cat_cols=None):\n    df_data = df_data.to_pandas()\n    if cat_cols is None:\n        cat_cols = list(df_data.select_dtypes(\"object\").columns)\n    df_data[cat_cols] = df_data[cat_cols].astype(\"category\")\n    return df_data, cat_cols\n\ndef reduce_mem_usage(df):\n    \"\"\" iterate through all the columns of a dataframe and modify the data type\n        to reduce memory usage.        \n    \"\"\"\n    start_mem = df.memory_usage().sum() / 1024**2\n    print('Memory usage of dataframe is {:.2f} MB'.format(start_mem))\n    \n    for col in df.columns:\n        col_type = df[col].dtype\n        if str(col_type)==\"category\":\n            continue\n        \n        if col_type != object:\n            c_min = df[col].min()\n            c_max = df[col].max()\n            if str(col_type)[:3] == 'int':\n                if c_min > np.iinfo(np.int8).min and c_max < np.iinfo(np.int8).max:\n                    df[col] = df[col].astype(np.int8)\n                elif c_min > np.iinfo(np.int16).min and c_max < np.iinfo(np.int16).max:\n                    df[col] = df[col].astype(np.int16)\n                elif c_min > np.iinfo(np.int32).min and c_max < np.iinfo(np.int32).max:\n                    df[col] = df[col].astype(np.int32)\n                elif c_min > np.iinfo(np.int64).min and c_max < np.iinfo(np.int64).max:\n                    df[col] = df[col].astype(np.int64)  \n            else:\n                if c_min > np.finfo(np.float16).min and c_max < np.finfo(np.float16).max:\n                    df[col] = df[col].astype(np.float16)\n                elif c_min > np.finfo(np.float32).min and c_max < np.finfo(np.float32).max:\n                    df[col] = df[col].astype(np.float32)\n                else:\n                    df[col] = df[col].astype(np.float64)\n        else:\n            continue\n    end_mem = df.memory_usage().sum() / 1024**2\n    print('Memory usage after optimization is: {:.2f} MB'.format(end_mem))\n    print('Decreased by {:.1f}%'.format(100 * (start_mem - end_mem) / start_mem))\n    \n    return df","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:36:02.917722Z","iopub.execute_input":"2024-06-02T14:36:02.917978Z","iopub.status.idle":"2024-06-02T14:36:02.949957Z","shell.execute_reply.started":"2024-06-02T14:36:02.917956Z","shell.execute_reply":"2024-06-02T14:36:02.949085Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### Training data","metadata":{}},{"cell_type":"code","source":"%%time\ndata_store = {\n    \"df_base\": read_file(TRAIN_DIR / \"train_base.parquet\"),\n    \"depth_0\": [\n        read_file(TRAIN_DIR / \"train_static_cb_0.parquet\"),\n        read_files(TRAIN_DIR / \"train_static_0_*.parquet\"),\n    ],\n    \"depth_1\": [\n        read_files(TRAIN_DIR / \"train_applprev_1_*.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_tax_registry_a_1.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_tax_registry_b_1.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_tax_registry_c_1.parquet\", 1),\n        read_files(TRAIN_DIR / \"train_credit_bureau_a_1_*.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_credit_bureau_b_1.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_other_1.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_person_1.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_deposit_1.parquet\", 1),\n        read_file(TRAIN_DIR / \"train_debitcard_1.parquet\", 1),\n    ],\n    \"depth_2\": [\n        read_file(TRAIN_DIR / \"train_credit_bureau_b_2.parquet\", 2),\n        read_files(TRAIN_DIR / \"train_credit_bureau_a_2_*.parquet\", 2),\n        read_file(TRAIN_DIR / \"train_applprev_2.parquet\", 2),\n        read_file(TRAIN_DIR / \"train_person_2.parquet\", 2)\n    ]\n}","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:36:02.979269Z","iopub.execute_input":"2024-06-02T14:36:02.979545Z","iopub.status.idle":"2024-06-02T14:38:14.891351Z","shell.execute_reply.started":"2024-06-02T14:36:02.979521Z","shell.execute_reply":"2024-06-02T14:38:14.890354Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%time\ndf_train = feature_eng(**data_store)\ndel data_store\ngc.collect()\ndf_train = df_train.pipe(Pipeline.filter_cols)\ndf_train, cat_cols = to_pandas(df_train)\ndf_train = reduce_mem_usage(df_train)\nnums=df_train.select_dtypes(exclude='category').columns\nfrom itertools import combinations, permutations\nnans_df = df_train[nums].isna()\nnans_groups={}\nfor col in nums:\n    cur_group = nans_df[col].sum()\n    try:\n        nans_groups[cur_group].append(col)\n    except:\n        nans_groups[cur_group]=[col]\ndel nans_df; x=gc.collect()\n\ndef reduce_group(grps):\n    use = []\n    for g in grps:\n        mx = 0; vx = g[0]\n        for gg in g:\n            n = df_train[gg].nunique()\n            if n>mx:\n                mx = n\n                vx = gg\n        use.append(vx)\n    return use\n\ndef group_columns_by_correlation(matrix, threshold=0.8):\n    correlation_matrix = matrix.corr()\n    groups = []\n    remaining_cols = list(matrix.columns)\n    while remaining_cols:\n        col = remaining_cols.pop(0)\n        group = [col]\n        correlated_cols = [col]\n        for c in remaining_cols:\n            if correlation_matrix.loc[col, c] >= threshold:\n                group.append(c)\n                correlated_cols.append(c)\n        groups.append(group)\n        remaining_cols = [c for c in remaining_cols if c not in correlated_cols]\n    \n    return groups\n\nuses=[]\nfor k,v in nans_groups.items():\n    if len(v)>1:\n            Vs = nans_groups[k]\n            grps= group_columns_by_correlation(df_train[Vs], threshold=0.8)\n            use=reduce_group(grps)\n            uses=uses+use\n    else:\n        uses=uses+v\n# print(uses)\nprint(len(uses))\nuses=uses+list(df_train.select_dtypes(include='category').columns)\nprint(len(uses))\ndf_train=df_train[uses]","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:38:14.893388Z","iopub.execute_input":"2024-06-02T14:38:14.893687Z","iopub.status.idle":"2024-06-02T14:39:53.069609Z","shell.execute_reply.started":"2024-06-02T14:38:14.893661Z","shell.execute_reply":"2024-06-02T14:39:53.068522Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Test data","metadata":{}},{"cell_type":"code","source":"data_store = {\n    \"df_base\": read_file(TEST_DIR / \"test_base.parquet\"),\n    \"depth_0\": [\n        read_file(TEST_DIR / \"test_static_cb_0.parquet\"),\n        read_files(TEST_DIR / \"test_static_0_*.parquet\"),\n    ],\n    \"depth_1\": [\n        read_files(TEST_DIR / \"test_applprev_1_*.parquet\", 1),\n        read_file(TEST_DIR / \"test_tax_registry_a_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_tax_registry_b_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_tax_registry_c_1.parquet\", 1),\n        read_files(TEST_DIR / \"test_credit_bureau_a_1_*.parquet\", 1),\n        read_file(TEST_DIR / \"test_credit_bureau_b_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_other_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_person_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_deposit_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_debitcard_1.parquet\", 1),\n    ],\n    \"depth_2\": [\n        read_file(TEST_DIR / \"test_credit_bureau_b_2.parquet\", 2),\n        read_files(TEST_DIR / \"test_credit_bureau_a_2_*.parquet\", 2),\n        read_file(TEST_DIR / \"test_applprev_2.parquet\", 2),\n        read_file(TEST_DIR / \"test_person_2.parquet\", 2)\n    ]\n}","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:39:53.071320Z","iopub.execute_input":"2024-06-02T14:39:53.071669Z","iopub.status.idle":"2024-06-02T14:39:53.378297Z","shell.execute_reply.started":"2024-06-02T14:39:53.071638Z","shell.execute_reply":"2024-06-02T14:39:53.377510Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test = feature_eng(**data_store)\nprint(\"test data shape:\\t\", df_test.shape)\ndel data_store\ngc.collect()\ndf_test = df_test.select([col for col in df_train.columns if col != \"target\"])\nprint(\"train data shape:\\t\", df_train.shape)\nprint(\"test data shape:\\t\", df_test.shape)\n\ndf_test, cat_cols = to_pandas(df_test, cat_cols)\ndf_test = reduce_mem_usage(df_test)\nweek_num = list(df_test[\"WEEK_NUM\"])\ngc.collect()","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:39:53.379753Z","iopub.execute_input":"2024-06-02T14:39:53.380290Z","iopub.status.idle":"2024-06-02T14:39:53.893426Z","shell.execute_reply.started":"2024-06-02T14:39:53.380254Z","shell.execute_reply":"2024-06-02T14:39:53.892528Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Preprocessing","metadata":{}},{"cell_type":"markdown","source":"- Drop rows whose total of nan is above 20%\n- Replace inf with nan\n- Fill nan with means / bfill ffill\n- Ordinal encoder fitted on train and used to transform test\n    + Unseen category is mapped to -1\n- Scaler fitted on train and used to transform test\n- Variance threshold","metadata":{}},{"cell_type":"code","source":"#df_train[cat_cols] = df_train[cat_cols].astype(str)\n#df_test[cat_cols] = df_test[cat_cols].astype(str)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:39:53.896881Z","iopub.execute_input":"2024-06-02T14:39:53.897333Z","iopub.status.idle":"2024-06-02T14:39:53.901377Z","shell.execute_reply.started":"2024-06-02T14:39:53.897297Z","shell.execute_reply":"2024-06-02T14:39:53.900530Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Preprocess:\n    def drop_na_cols(df: pd.DataFrame, threshold = 0.6):\n        # Drop columns whose nan counts is above a percentage threshold\n        # done in reading file with th 0.7\n        to_drop = []\n        preserve_df = df[[\"target\", \"case_id\", \"WEEK_NUM\"]].copy()\n        for col in df.columns:\n            if df[col].isna().sum() / len(df) > threshold or df[col].isnull().sum() / len(df) > threshold:\n                to_drop.append(col)\n        df = df.drop(columns=to_drop)\n        df[[\"target\", \"case_id\", \"WEEK_NUM\"]] = preserve_df\n        return df\n    \n    def drop_na_rows(df: pd.DataFrame, threshold=0.8) -> pd.DataFrame:\n        min_non_nan_values = int((threshold) * df.shape[1])\n        df = df.dropna(thresh=min_non_nan_values)\n        return df\n    \n    def replace_inf(df: pd.DataFrame) -> pd.DataFrame:\n        numeric_cols = df.select_dtypes(include=[np.number]).columns\n        df[numeric_cols] = df[numeric_cols].applymap(lambda x: np.nan if np.isinf(x) else x)\n        return df\n    def variance_drop(df: pd.DataFrame) -> pd.DataFrame:\n        var_thresh = VarianceThreshold(0.7)\n        df = var_thresh.fit_transform(df)\n        return df\n\ndef transform_with_unseen(encoder, train_df, test_df, column_name):\n    known_categories = encoder.categories_[0]\n    category_mapping = {category: idx for idx, category in enumerate(known_categories)}\n    test_df[column_name] = test_df[column_name].map(category_mapping)\n    \n    for cat_feature in cat_cols:\n        all_categories = list(set(train_df[cat_feature].unique()) | set(test_df[cat_feature].unique()))\n        train_df[cat_feature] = pd.Categorical(train_df[cat_feature], categories=all_categories)\n        test_df[cat_feature] = pd.Categorical(test_df[cat_feature], categories=all_categories)\n    \n    return test_df\n\ndef fill_df(df: pd.DataFrame, df_test: pd.DataFrame) -> pd.DataFrame:\n        for col in df.columns:\n            if df[col].dtype != 'object' and df[col].dtype != 'string' and df[col].dtype != 'category':\n                if not np.isnan(df[col].mean()):\n                    df[col] = df[col].fillna(df[col].mean().astype(df[col].dtype))\n                    df_test[col] = df_test[col].fillna(df_test[col].mean().astype(df_test[col].dtype))\n                else:\n                    df[col] = df[col].fillna(0)\n                    df_test[col] = df_test[col].fillna(0)\n            else:\n                df[col] = df[col].bfill().ffill()\n                df_test[col] = df_test[col].bfill().ffill()\n        return df, df_test\n    \ndef label_encode(df: pd.DataFrame, df_test: pd.DataFrame) -> pd.DataFrame:\n    enc = OrdinalEncoder(dtype=np.int16, handle_unknown='use_encoded_value', unknown_value=-1)\n    #imputer = SimpleImputer(strategy='constant', fill_value='unseen')\n    # Learn the encoding of train set -> apply to test set\n    for col in df.columns:\n        if df[col].dtype == 'object' or df[col].dtype == 'string' or df[col].dtype == 'category':\n            df[col] = enc.fit_transform(df[col].values.reshape(-1, 1)).flatten()\n            #imputer.fit_transform(df_test[col])\n            df_test[col] = enc.transform(df_test[col].values.reshape(-1, 1)).flatten()\n    return df, df_test\n    \ndef scale_numeric(df: pd.DataFrame, df_test: pd.DataFrame) -> pd.DataFrame:\n    scaler = RobustScaler()\n    # Learn the scaling parameter of train set -> apply to test set\n    for col in df.columns:\n        if df[col].dtype != 'object' and df[col].dtype != 'string' and df[col].dtype != 'category':\n            threshold = df[col].quantile(0.95)\n            df[col] = np.where(df[col] > threshold, threshold, df[col])\n            df[col] = scaler.fit_transform(df[col].values.reshape(-1, 1)).flatten().astype(df[col].dtype)\n            df_test[col] = scaler.transform(df_test[col].values.reshape(-1, 1)).flatten()\n    return df, df_test","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:39:53.902723Z","iopub.execute_input":"2024-06-02T14:39:53.903078Z","iopub.status.idle":"2024-06-02T14:39:53.933828Z","shell.execute_reply.started":"2024-06-02T14:39:53.903047Z","shell.execute_reply":"2024-06-02T14:39:53.932958Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%time\ndf_train = (\n                df_train.pipe(Preprocess.drop_na_rows)\n                    .pipe(Preprocess.drop_na_cols)\n                    .pipe(Preprocess.replace_inf)\n            )\ndf_train = df_train.reset_index(drop=True)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:39:53.935404Z","iopub.execute_input":"2024-06-02T14:39:53.935664Z","iopub.status.idle":"2024-06-02T14:47:10.802131Z","shell.execute_reply.started":"2024-06-02T14:39:53.935642Z","shell.execute_reply":"2024-06-02T14:47:10.801028Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train = reduce_mem_usage(df_train)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:47:10.803647Z","iopub.execute_input":"2024-06-02T14:47:10.804461Z","iopub.status.idle":"2024-06-02T14:47:13.887040Z","shell.execute_reply.started":"2024-06-02T14:47:10.804421Z","shell.execute_reply":"2024-06-02T14:47:13.886051Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y = df_train[\"target\"]\nweeks = df_train[\"WEEK_NUM\"]\ndf_train= df_train.drop(columns=[\"target\", \"case_id\", \"WEEK_NUM\"])\ncv = StratifiedGroupKFold(n_splits=6, shuffle=False)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:47:13.888511Z","iopub.execute_input":"2024-06-02T14:47:13.889195Z","iopub.status.idle":"2024-06-02T14:47:14.603466Z","shell.execute_reply.started":"2024-06-02T14:47:13.889138Z","shell.execute_reply":"2024-06-02T14:47:14.602657Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%time\ndf_test = df_test[df_train.columns]\ndf_train, df_test = fill_df(df_train, df_test)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:47:14.604728Z","iopub.execute_input":"2024-06-02T14:47:14.605032Z","iopub.status.idle":"2024-06-02T14:47:18.291822Z","shell.execute_reply.started":"2024-06-02T14:47:14.605006Z","shell.execute_reply":"2024-06-02T14:47:18.290816Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"> Sau khi fill NA ở df_test vẫn còn cột full na -> xử lý ở encoder","metadata":{}},{"cell_type":"code","source":"df_test = df_test.pipe(Preprocess.replace_inf)\ndf_train, df_test = scale_numeric(df_train, df_test)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:47:18.293351Z","iopub.execute_input":"2024-06-02T14:47:18.294052Z","iopub.status.idle":"2024-06-02T14:47:39.638582Z","shell.execute_reply.started":"2024-06-02T14:47:18.294011Z","shell.execute_reply":"2024-06-02T14:47:39.637507Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test = reduce_mem_usage(df_test)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:47:39.639791Z","iopub.execute_input":"2024-06-02T14:47:39.640163Z","iopub.status.idle":"2024-06-02T14:47:39.748498Z","shell.execute_reply.started":"2024-06-02T14:47:39.640127Z","shell.execute_reply":"2024-06-02T14:47:39.747408Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train, df_test = label_encode(df_train, df_test)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:47:39.749712Z","iopub.execute_input":"2024-06-02T14:47:39.749994Z","iopub.status.idle":"2024-06-02T14:48:04.299316Z","shell.execute_reply.started":"2024-06-02T14:47:39.749968Z","shell.execute_reply":"2024-06-02T14:48:04.298465Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cat_cols = [col for col in cat_cols if col in df_train.columns]","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:48:04.304371Z","iopub.execute_input":"2024-06-02T14:48:04.304672Z","iopub.status.idle":"2024-06-02T14:48:04.309381Z","shell.execute_reply.started":"2024-06-02T14:48:04.304646Z","shell.execute_reply":"2024-06-02T14:48:04.308405Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## PCA","metadata":{}},{"cell_type":"markdown","source":"#### SMOTE for imbalance handling\n> No improvement","metadata":{}},{"cell_type":"markdown","source":"## Mutual information\n> not working","metadata":{"_kg_hide-input":true}},{"cell_type":"markdown","source":"# Training with baseline model (LightGBM)","metadata":{}},{"cell_type":"code","source":"from sklearn import metrics\ndef stability_metric(y_true, y_pred, weeks_to_score):\n    \"\"\"\n    Custom metric for model optimization during training\n    \"\"\"\n    gini_in_time = []\n    for week in sorted(weeks_to_score.unique()):\n        week_idx = weeks_to_score.eq(week)\n        if (np.unique(y_true[week_idx]).shape[0] <= 1):\n            continue\n        gini = np.array(\n            2 * metrics.roc_auc_score(y_true[week_idx], y_pred[week_idx]) - 1\n        )\n        gini_in_time.append(gini)\n\n    # hyperparams\n    w_fallingrate = 88.0\n    w_resstd = -0.5\n\n    x = np.arange(len(gini_in_time))\n    y = np.array(gini_in_time)\n\n    avg_gini = np.mean(y)\n    \n    if x.shape[0] <= 1:\n        return avg_gini\n    \n    a, b = np.polyfit(x, y, 1)\n    y_hat = a * x + b\n    residuals = y - y_hat\n    res_std = np.std(residuals)\n\n    stability_score = avg_gini + w_fallingrate * min(0, a) + w_resstd * res_std\n\n    return stability_score\n","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:48:04.310570Z","iopub.execute_input":"2024-06-02T14:48:04.310869Z","iopub.status.idle":"2024-06-02T14:48:04.320913Z","shell.execute_reply.started":"2024-06-02T14:48:04.310845Z","shell.execute_reply":"2024-06-02T14:48:04.319926Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"params = [{\n    \"boosting_type\": \"gbdt\",\n    \"objective\": \"binary\",\n    \"metric\": \"auc\",\n    \"max_depth\": 8,\n    \"num_leaves\": 32,\n    \"learning_rate\": 0.03,\n    \"n_estimators\": 1000,\n    \"reg_alpha\": 0.1,\n    \"reg_lambda\": 10,\n    \"extra_trees\": True,\n    \"num_boost_round\": 300,\n    \"verbose\": -1,\n    \"device\": \"gpu\",\n    \"missing\": -1\n}, {\n    \"boosting_type\": \"gbdt\",\n    \"objective\": \"binary\",\n    \"metric\": \"auc\",\n    \"max_depth\": 16,\n    \"num_leaves\": 64,\n    \"learning_rate\": 0.05,\n    \"n_estimators\": 1500,\n    \"reg_alpha\": 0.1,\n    \"reg_lambda\": 10,\n    \"extra_trees\": True,\n    \"num_boost_round\": 300,\n    \"verbose\": -1,\n    \"device\": \"gpu\",\n    \"missing\": -1\n}]","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:48:04.322367Z","iopub.execute_input":"2024-06-02T14:48:04.322722Z","iopub.status.idle":"2024-06-02T14:48:04.330417Z","shell.execute_reply.started":"2024-06-02T14:48:04.322688Z","shell.execute_reply":"2024-06-02T14:48:04.329478Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"type(weeks)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:48:04.331807Z","iopub.execute_input":"2024-06-02T14:48:04.332066Z","iopub.status.idle":"2024-06-02T14:48:04.343053Z","shell.execute_reply.started":"2024-06-02T14:48:04.332044Z","shell.execute_reply":"2024-06-02T14:48:04.342137Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"stability_metric(np.array([1, 0, 1, 0, 1, 1]), np.array([1, 0, 1, 0, 1, 1]), pd.Series([1, 1, 1, 1, 1, 2]))","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:48:04.344490Z","iopub.execute_input":"2024-06-02T14:48:04.344757Z","iopub.status.idle":"2024-06-02T14:48:04.354734Z","shell.execute_reply.started":"2024-06-02T14:48:04.344735Z","shell.execute_reply":"2024-06-02T14:48:04.353734Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### LGB + CatBoost","metadata":{}},{"cell_type":"code","source":"n_est=1200","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:48:04.355779Z","iopub.execute_input":"2024-06-02T14:48:04.356523Z","iopub.status.idle":"2024-06-02T14:48:04.360877Z","shell.execute_reply.started":"2024-06-02T14:48:04.356489Z","shell.execute_reply":"2024-06-02T14:48:04.359953Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%time\nfrom catboost import CatBoostClassifier, Pool\n\nfitted_models_cat = []\nfitted_models_lgb = []\n\ncv_scores_cat = []\ncv_scores_lgb = []\ncv_stab_scores_cat = []\ncv_stab_scores_lgb = []\nN = df_train.shape[0]\niterator = 0\nfor idx_train, idx_valid in cv.split(df_train.iloc[:N], y.iloc[:N], groups=weeks.iloc[:N]):#\n    X_train, y_train = df_train.iloc[:N].iloc[idx_train], y.iloc[:N].iloc[idx_train]# \n    X_valid, y_valid = df_train.iloc[:N].iloc[idx_valid], y.iloc[:N].iloc[idx_valid]\n    \n    weeks_val = weeks.iloc[:N].iloc[idx_valid]\n    def eval_funk(y_true, y_pred): \n        return \"stability_score\", stability_metric(y_true, y_pred, weeks_val), True\n    \n    class CB_stability_metric(object):\n        def __init__(self, weeks_val):\n            self.weeks_val = weeks_val\n        \n        def is_max_optimal(self):\n            return True\n\n        def evaluate(self, approxes, target, weight):\n            return stability_metric(target, approxes, self.weeks_val), sum(weight)\n\n        def get_final_error(self, error, weight):\n            return error / (weight + 1e-38)\n\n    \n    train_pool = Pool(X_train, y_train,cat_features=cat_cols)\n    val_pool = Pool(X_valid, y_valid,cat_features=cat_cols)\n    clf = CatBoostClassifier(\n        eval_metric=\"AUC\",\n        task_type='GPU',\n        learning_rate=0.03,\n        iterations=n_est)\n    random_seed=28\n    clf.fit(train_pool, eval_set=val_pool,verbose=300)\n    fitted_models_cat.append(clf)\n    y_pred_valid = clf.predict_proba(X_valid)[:,1]\n    auc_score = roc_auc_score(y_valid, y_pred_valid)\n    cv_scores_cat.append(auc_score)\n    stab_score = eval_funk(y_valid, y_pred_valid)[1]\n    cv_stab_scores_cat.append(stab_score)\n    \n    \n    X_train[cat_cols] = X_train[cat_cols].astype(\"category\")\n    X_valid[cat_cols] = X_valid[cat_cols].astype(\"category\")\n    \n    \n    if iterator % 2 == 0:\n        model = lgb.LGBMClassifier(**params[0])\n    else:\n        model = lgb.LGBMClassifier(**params[1])\n    model.fit(\n        X_train, y_train,\n        eval_set = [(X_valid, y_valid)],\n        eval_metric = eval_funk,\n        callbacks = [lgb.log_evaluation(300), lgb.early_stopping(100)] )\n    \n    fitted_models_lgb.append(model)\n    y_pred_valid = model.predict_proba(X_valid)[:,1]\n    auc_score = roc_auc_score(y_valid, y_pred_valid)\n    cv_scores_lgb.append(auc_score)\n    stab_score = eval_funk(y_valid, y_pred_valid)[1]\n    cv_stab_scores_lgb.append(stab_score)\n    \n    iterator += 1\n    \n    ","metadata":{"execution":{"iopub.status.busy":"2024-06-02T14:48:04.362660Z","iopub.execute_input":"2024-06-02T14:48:04.362994Z","iopub.status.idle":"2024-06-02T15:43:01.283039Z","shell.execute_reply.started":"2024-06-02T14:48:04.362963Z","shell.execute_reply":"2024-06-02T15:43:01.281967Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\nprint(\"CV AUC scores cat: \", cv_scores_cat)\nprint(\"Maximum CV AUC score cat: \", max(cv_scores_cat))\n\nprint(\"CV AUC scores lgb: \", cv_scores_lgb)\nprint(\"Maximum CV AUC score lgb: \", max(cv_scores_lgb))\n\nprint()\n\nprint(\"CV stability scores cat: \", cv_stab_scores_cat)\nprint(\"Maximum CV stability score cat: \", max(cv_stab_scores_cat))\n\nprint(\"CV stability scores lgb: \", cv_stab_scores_lgb)\nprint(\"Maximum CV stability score lgb: \", max(cv_stab_scores_lgb))","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:01.284357Z","iopub.execute_input":"2024-06-02T15:43:01.284664Z","iopub.status.idle":"2024-06-02T15:43:01.292135Z","shell.execute_reply.started":"2024-06-02T15:43:01.284637Z","shell.execute_reply":"2024-06-02T15:43:01.291084Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Ensemble","metadata":{}},{"cell_type":"code","source":"class EnsembleModel(BaseEstimator, RegressorMixin):\n    def __init__(self, estimators):\n        super().__init__()\n        self.estimators = estimators\n        \n    def fit(self, X, y=None):\n        return self\n    \n    def predict(self, X):\n        y_preds = [estimator.predict(X) for estimator in self.estimators]\n        return np.mean(y_preds, axis=0)\n    \n    def predict_proba(self, X):\n        y_preds = [estimator.predict_proba(X) for estimator in self.estimators]\n        #X[cat_cols] = X[cat_cols].astype(\"category\")\n        y_preds += [estimator.predict_proba(X) for estimator in self.estimators[6:]]\n        \n        return np.mean(y_preds, axis=0)\n\nmodel = EnsembleModel(fitted_models_cat+fitted_models_lgb)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:01.293453Z","iopub.execute_input":"2024-06-02T15:43:01.293803Z","iopub.status.idle":"2024-06-02T15:43:01.303304Z","shell.execute_reply.started":"2024-06-02T15:43:01.293770Z","shell.execute_reply":"2024-06-02T15:43:01.302267Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Submission","metadata":{}},{"cell_type":"code","source":"for cat_feature in cat_cols:\n    all_categories = list(set(df_train[cat_feature].unique()) | set(df_test[cat_feature].unique()))\n    df_test[cat_feature] = pd.Categorical(df_test[cat_feature], categories=all_categories)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:01.304702Z","iopub.execute_input":"2024-06-02T15:43:01.305039Z","iopub.status.idle":"2024-06-02T15:43:01.799043Z","shell.execute_reply.started":"2024-06-02T15:43:01.305014Z","shell.execute_reply":"2024-06-02T15:43:01.798247Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred = model.predict_proba(df_test)[:, 1]\ny_pred","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:01.800151Z","iopub.execute_input":"2024-06-02T15:43:01.800465Z","iopub.status.idle":"2024-06-02T15:43:02.707533Z","shell.execute_reply.started":"2024-06-02T15:43:01.800439Z","shell.execute_reply":"2024-06-02T15:43:02.706632Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"subm_df = pd.read_csv(ROOT / \"sample_submission.csv\")\nsubm_df = subm_df.set_index(\"case_id\")","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:02.709155Z","iopub.execute_input":"2024-06-02T15:43:02.709502Z","iopub.status.idle":"2024-06-02T15:43:02.725462Z","shell.execute_reply.started":"2024-06-02T15:43:02.709475Z","shell.execute_reply":"2024-06-02T15:43:02.724480Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"subm_df[\"score\"] = y_pred\nsubm_df[\"WEEK_NUM\"] = week_num\nprint(\"Check null: \", subm_df[\"score\"].isnull().any())","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:02.726661Z","iopub.execute_input":"2024-06-02T15:43:02.726936Z","iopub.status.idle":"2024-06-02T15:43:02.733635Z","shell.execute_reply.started":"2024-06-02T15:43:02.726912Z","shell.execute_reply":"2024-06-02T15:43:02.732555Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"condition = subm_df[\"WEEK_NUM\"] < (subm_df[\"WEEK_NUM\"].max() - subm_df[\"WEEK_NUM\"].min())/2 + subm_df[\"WEEK_NUM\"].min() \nSHIFT = 0.0718\nsubm_df.loc[condition, 'score'] = (subm_df.loc[condition, 'score'] - SHIFT).clip(0)","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:02.735271Z","iopub.execute_input":"2024-06-02T15:43:02.735777Z","iopub.status.idle":"2024-06-02T15:43:02.746725Z","shell.execute_reply.started":"2024-06-02T15:43:02.735737Z","shell.execute_reply":"2024-06-02T15:43:02.745811Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"del subm_df[\"WEEK_NUM\"]\nsubm_df.to_csv(\"submission.csv\")\nsubm_df","metadata":{"execution":{"iopub.status.busy":"2024-06-02T15:43:02.748034Z","iopub.execute_input":"2024-06-02T15:43:02.748600Z","iopub.status.idle":"2024-06-02T15:43:02.768451Z","shell.execute_reply.started":"2024-06-02T15:43:02.748565Z","shell.execute_reply":"2024-06-02T15:43:02.767534Z"},"trusted":true},"execution_count":null,"outputs":[]}]}