{"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":"none","dataSources":[{"sourceId":50160,"databundleVersionId":7921029,"sourceType":"competition"}],"dockerImageVersionId":30648,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import os\nimport gc\nfrom glob import glob\nfrom pathlib import Path\nfrom datetime import datetime\n\nimport numpy as np\nimport pandas as pd\nimport polars as pl\n\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nimport joblib\n\nimport warnings\nwarnings.simplefilter(action='ignore', category=FutureWarning)","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2024-04-28T11:51:07.790035Z","iopub.execute_input":"2024-04-28T11:51:07.791264Z","iopub.status.idle":"2024-04-28T11:51:09.052606Z","shell.execute_reply.started":"2024-04-28T11:51:07.791214Z","shell.execute_reply":"2024-04-28T11:51:09.051278Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Configuration","metadata":{}},{"cell_type":"code","source":"SEED = 42\n\n# data\nNULL_TRESHOLD = 0.7\nMIN_CARDINALITY = 1\nMAX_CARDINALITY = 200\nNORMALIZATION = 'Z_NORM' # [Z_NORM, MINMAX]\n\nTRAIN = False\nTRAIN_SIMPLE = False\nKFOLD = 5 if not TRAIN_SIMPLE else 2\nEARLY_STOPPING = 100 if not TRAIN_SIMPLE else 5","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.055490Z","iopub.execute_input":"2024-04-28T11:51:09.056138Z","iopub.status.idle":"2024-04-28T11:51:09.063550Z","shell.execute_reply.started":"2024-04-28T11:51:09.056082Z","shell.execute_reply":"2024-04-28T11:51:09.061816Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# LGBM parameters\nparams = {\n    \"boosting_type\": \"gbdt\",\n    \"objective\": \"binary\",\n    \"metric\": \"auc\",\n    \"max_depth\": 10,  \n    \"learning_rate\": 0.05,\n    \"n_estimators\": 2000,  \n    \"colsample_bytree\": 0.8,\n    \"colsample_bynode\": 0.8,\n    \"verbose\": -1,\n    \"random_state\": 42,\n    \"reg_alpha\": 0.1,\n    \"reg_lambda\": 10,\n    \"extra_trees\":True,\n    'num_leaves':64,\n    \"device\": 'gpu', \n    \"verbose\": -1,\n}\n\nif TRAIN_SIMPLE:\n    params['max_depth'] = 3\n    params['n_estimators'] = 3\n    params['device'] = 'cpu'","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.065139Z","iopub.execute_input":"2024-04-28T11:51:09.065510Z","iopub.status.idle":"2024-04-28T11:51:09.080267Z","shell.execute_reply.started":"2024-04-28T11:51:09.065477Z","shell.execute_reply":"2024-04-28T11:51:09.078842Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ROOT            = Path(\"/kaggle/input/home-credit-credit-risk-model-stability\")\nTRAIN_DIR       = ROOT / \"parquet_files\" / \"train\"\nTEST_DIR        = ROOT / \"parquet_files\" / \"test\"","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.084228Z","iopub.execute_input":"2024-04-28T11:51:09.084799Z","iopub.status.idle":"2024-04-28T11:51:09.096216Z","shell.execute_reply.started":"2024-04-28T11:51:09.084731Z","shell.execute_reply":"2024-04-28T11:51:09.094848Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Data Description","metadata":{}},{"cell_type":"code","source":"feat_def = pd.read_csv('/kaggle/input/home-credit-credit-risk-model-stability/feature_definitions.csv')\n\nprint(feat_def.shape)\nfeat_def.head()","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.097978Z","iopub.execute_input":"2024-04-28T11:51:09.098915Z","iopub.status.idle":"2024-04-28T11:51:09.126124Z","shell.execute_reply.started":"2024-04-28T11:51:09.098878Z","shell.execute_reply":"2024-04-28T11:51:09.124752Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def definition(variable):\n    print(feat_def[feat_def['Variable'] == variable]['Description'].values[0])","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.127588Z","iopub.execute_input":"2024-04-28T11:51:09.127968Z","iopub.status.idle":"2024-04-28T11:51:09.136161Z","shell.execute_reply.started":"2024-04-28T11:51:09.127935Z","shell.execute_reply":"2024-04-28T11:51:09.134182Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Pipeline & Aggregator","metadata":{}},{"cell_type":"code","source":"binary_columns = ['bankacctype_710L', 'isbidproduct_1095L', 'paytype1st_925L', 'typesuite_864L', 'actualdpdtolerance_344P', 'clientscnt12m_3712952L', 'clientscnt3m_3712950L', 'clientscnt_100L']","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.138165Z","iopub.execute_input":"2024-04-28T11:51:09.138572Z","iopub.status.idle":"2024-04-28T11:51:09.147779Z","shell.execute_reply.started":"2024-04-28T11:51:09.138542Z","shell.execute_reply":"2024-04-28T11:51:09.146069Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Pipeline:\n    def __init__(self, n=None, verbose=0):\n        self.n = n\n        self.verbose = verbose\n    \n    def set_table_dtypes(self, df):\n        if self.n:\n            df = df.sample(n=self.n)\n\n        for col in df.columns:\n            if col in [\"case_id\", \"WEEK_NUM\", \"num_group1\", \"num_group2\"]: # train_base columns\n                df = df.with_columns(pl.col(col).cast(pl.Int32))  \n            elif col in [\"date_decision\"]:\n                df = df.with_columns(pl.col(col).cast(pl.Date))\n            elif col in binary_columns:\n                df = df.with_columns(pl.col(col).cast(pl.String).cast(pl.Categorical))  \n            \n            # use descriptive column names\n            elif col.startswith(\"for\"):\n                df = df.with_columns(pl.col(col).cast(pl.Int16))\n            elif \"num\" in col or \"cnt\" in col:\n                df = df.with_columns(pl.col(col).cast(pl.Int32))\n            elif col.startswith(\"pct\") or 'rate' in col:\n                df = df.with_columns(pl.col(col).cast(pl.Float32))\n            \n            # use transformation definitions\n            elif col[-1] == 'D': \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\", \"L\", \"T\"):\n                if df[col].dtype == 'boolean':\n                    df = df.with_columns(pl.col(col).cast(pl.Boolean))\n                if df[col].dtype.is_float():\n                    df = df.with_columns(pl.col(col).cast(pl.Float64))\n                elif df[col].dtype.is_integer():\n                    df = df.with_columns(pl.col(col).cast(pl.Int64))\n                else:\n                    df = df.with_columns(pl.col(col).cast(pl.Categorical))          \n\n        return df\n\n    def filter_cols(self, df):\n        essential_cols = [\"target\", \"case_id\", \"WEEK_NUM\"]\n        \n        count = 0\n        for col in filter(lambda x: x not in essential_cols, df.columns):\n            isnull = df[col].is_null().mean()\n\n            if isnull > NULL_TRESHOLD:\n                df = df.drop(col)\n                count += 1\n        print(f'Drop {count} columns w/ too many null values')\n\n        count = 0\n        for col in df.columns:\n            if (col not in essential_cols) & (df[col].dtype == pl.Categorical):\n                freq = df[col].n_unique()\n\n                if (freq == MIN_CARDINALITY) | (freq > MAX_CARDINALITY):\n                    df = df.drop(col)\n                    count += 1\n        print(f'Drop {count} columns w/ too small/many categories')\n        \n        return df\n\n    def handle_dates(self, df):\n        df = df.with_columns (\n            month_decision = pl.col(\"date_decision\").dt.month(),\n            weekday_decision = pl.col(\"date_decision\").dt.weekday(),\n        )\n        \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())\n\n        df = df.drop(\"date_decision\", \"MONTH\")\n\n        return df\n\n    def remove_outliers(self, df):\n        essential_cols = ['target', 'case_id', 'WEEK_NUM', 'month_decision', 'weekday_decision']\n        \n        mask_agg = pl.Series([False] * len(df))\n        for col in filter(lambda x: df[x].dtype.is_numeric() and x not in essential_cols, df.columns):                \n            q1 = df[col].quantile(0.25)\n            q3 = df[col].quantile(0.75)\n            iqr = q3 - q1\n\n            mask = (df[col] < q3 - 1.5 * iqr) & (df[col] > q1 + 1.5 * iqr)\n            mask = mask.fill_null(False)\n            mask_agg |= mask\n\n        df = df.filter(~mask_agg)\n        print(f'Drop {mask_agg.sum()} row w/ outliers')\n                \n        return df\n    \n    def normalize(self, df):\n        count = 0\n        for col in df.columns:\n            if df[col].dtype.is_numeric() and \\\n                col not in ['target', 'case_id', 'WEEK_NUM', 'month_decision', 'weekday_decision'] and \\\n                (col[-1] != 'D' and 'days' not in col):\n                \n                if NORMALIZATION == 'Z_NORM':\n                    df = df.with_columns(pl.col(col)-pl.col(col).mean() / pl.col(col).std())\n                elif NORMALIZATION == 'MINMAX':\n                    df = df.with_columns((pl.col(col)-pl.col(col).min()) / (pl.col(col).max()-pl.col(col).min()))\n                count += 1\n        print(f'Normalize {count} columns')\n\n        return df\n    \n    def remove_columns(self, df, target):\n        target = list(set(target))\n        \n        return df.drop(target)","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.149907Z","iopub.execute_input":"2024-04-28T11:51:09.150384Z","iopub.status.idle":"2024-04-28T11:51:09.187394Z","shell.execute_reply.started":"2024-04-28T11:51:09.150342Z","shell.execute_reply":"2024-04-28T11:51:09.185752Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"class Aggregator:\n    def __init__(self, version=0, verbose=0):\n        self.version = version\n        self.verbose = verbose\n\n    def num_expr(self, df):\n        cols = [col for col in df.columns if \\\n                (df[col].dtype.is_numeric() and col not in ['case_id', 'num_group1', 'num_group2'])]\n        if self.verbose:\n            print(f'num: {cols}')\n\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        \n        expr_mean = [pl.mean(col).alias(f\"mean_{col}\") for col in cols]\n        expr_std = [pl.std(col).alias(f\"std_{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        \n        if self.version == 0:\n            return expr_max + expr_min + expr_mean + expr_std\n        elif self.version == 1:\n            return expr_max + expr_last + expr_mean\n    \n    def date_expr(self, df):\n        cols = [col for col in df.columns if df[col].dtype==pl.Date]\n        if self.verbose:\n            print(f'date: {cols}')\n            \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\n        expr_mean = [pl.mean(col).alias(f\"mean_{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        \n        if self.version == 0:\n            return expr_max + expr_min\n        elif self.version == 1:\n            return expr_max + expr_last + expr_mean\n\n    def str_expr(self, df):\n        cols = [col for col in df.columns if col[-1] in (\"M\",)]\n        if self.verbose:\n            print(f'str: {cols}')\n        \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        \n        return expr_max + expr_last#+expr_count\n    \n    def other_expr(self, df):\n        cols = [col for col in df.columns if \\\n                (not df[col].dtype.is_numeric() and col[-1] in (\"T\", \"L\"))]\n        if self.verbose:\n            print(f'other: {cols}')\n            \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        \n        return  expr_max + expr_last\n\n    def count_expr(self, df):\n        cols = [col for col in df.columns if \"num_group\" in col]\n        if self.verbose:\n            print(f'count: {cols}')\n            \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 cat_expr(self, df):\n        cols = [col for col in df.columns if df[col].dtype==pl.Categorical]\n        if self.verbose:\n            print(f'cat: {cols}')\n            \n        expr_mode = [pl.col(col).mode().first().alias(f'mode_{col}') for col in cols]\n        \n        return expr_mode\n        \n    def get_exprs(self, df):\n        if self.version == 0:\n            exprs = self.num_expr(df) + \\\n                    self.date_expr(df) + \\\n                    self.cat_expr(df)\n        elif self.version == 1:\n            exprs = self.num_expr(df) + \\\n                    self.date_expr(df) + \\\n                    self.str_expr(df) + \\\n                    self.other_expr(df) + \\\n                    self.count_expr(df)\n            \n        return exprs","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.189054Z","iopub.execute_input":"2024-04-28T11:51:09.189549Z","iopub.status.idle":"2024-04-28T11:51:09.221560Z","shell.execute_reply.started":"2024-04-28T11:51:09.189506Z","shell.execute_reply":"2024-04-28T11:51:09.220474Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### File I/O","metadata":{}},{"cell_type":"code","source":"def read_file(path, aggregator, pipeline, depth=None):\n    print(f'Read {path}')\n    df = pl.read_parquet(path).pipe(pipeline.set_table_dtypes)\n\n    if depth in [1, 2]:\n        df = df.group_by(\"case_id\").agg(aggregator.get_exprs(df))\n\n    return df\n\ndef read_files(regex_path, aggregator, pipeline, depth=None):\n    chunks = []\n    for path in glob(str(regex_path)):\n        print(f'Read {path}')\n        chunks.append(pl.read_parquet(path))\n\n    df = pl.concat(chunks, how=\"vertical_relaxed\")\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\n    return df","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.222912Z","iopub.execute_input":"2024-04-28T11:51:09.223261Z","iopub.status.idle":"2024-04-28T11:51:09.238090Z","shell.execute_reply.started":"2024-04-28T11:51:09.223230Z","shell.execute_reply":"2024-04-28T11:51:09.237031Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Insepect Dataframes","metadata":{}},{"cell_type":"markdown","source":"#### train_base","metadata":{}},{"cell_type":"code","source":"df_base = pl.read_parquet(TRAIN_DIR / \"train_base.parquet\")\n\nprint(df_base.shape)\nprint(df_base.head())","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.239250Z","iopub.execute_input":"2024-04-28T11:51:09.239668Z","iopub.status.idle":"2024-04-28T11:51:09.392000Z","shell.execute_reply.started":"2024-04-28T11:51:09.239636Z","shell.execute_reply":"2024-04-28T11:51:09.391000Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_target(case_id):\n    return df_base.filter(pl.col('case_id')==case_id)['target'][0]\n\nprint(get_target(3))\nprint(get_target(4))\n\ndef get_targets(case_id_list):\n    return df_base.filter(pl.col('case_id').is_in(case_id_list))['target']\n\ndisplay(get_targets([1, 2, 3]))","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.393114Z","iopub.execute_input":"2024-04-28T11:51:09.393626Z","iopub.status.idle":"2024-04-28T11:51:09.419213Z","shell.execute_reply.started":"2024-04-28T11:51:09.393597Z","shell.execute_reply":"2024-04-28T11:51:09.418046Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### Helper Functions","metadata":{}},{"cell_type":"code","source":"def dataframe_information(path):\n    df = pl.read_parquet(TRAIN_DIR / path)\n    \n    print(df.shape)\n    print(df.head())\n    \n    count = 0\n    unique_id = np.unique(df['case_id'])\n    targets = get_targets(unique_id)\n\n    print(f'{targets.sum()} / {len(unique_id)}, {targets.sum()/len(unique_id)*100:.2f}%')\n    \n    for col in df.columns:\n        if col in ['case_id', 'num_group1', 'num_group2']:\n            continue\n\n        print(col)\n        definition(col)\n\n    return df","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.425229Z","iopub.execute_input":"2024-04-28T11:51:09.425944Z","iopub.status.idle":"2024-04-28T11:51:09.434442Z","shell.execute_reply.started":"2024-04-28T11:51:09.425901Z","shell.execute_reply":"2024-04-28T11:51:09.433290Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# analyze correlation with target\ndef target_corr(data):\n    corr = data.with_columns(\n        get_targets(data['case_id']).alias('target')\n    ).corr()\n    \n    corr_df = pd.DataFrame({'features': data.columns + ['target'], 'corr': corr['target']})\n    return corr_df.sort_values('corr', ascending=False)\n\n# analyze & reduce columns with correlation\ndef group_columns_by_correlation(corr_matrix, threshold=0.8):\n    column_dict = {}\n    for i, c in enumerate(corr_matrix.columns):\n        column_dict[c] = i\n    \n    groups = []\n    remaining_cols = list(corr_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 corr_matrix.select(pl.col(col))[column_dict[c]].item() >= 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\ndef reduce_columns_by_correlation(data):\n    groups = group_columns_by_correlation(data.corr())\n    print(groups)\n    \n    select_cols = [group[0] for group in groups]\n\n    print(select_cols)\n    print(len(select_cols))\n    \n    return select_cols","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.436153Z","iopub.execute_input":"2024-04-28T11:51:09.436917Z","iopub.status.idle":"2024-04-28T11:51:09.451386Z","shell.execute_reply.started":"2024-04-28T11:51:09.436875Z","shell.execute_reply":"2024-04-28T11:51:09.450036Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### train_credit_bureau_b_2","metadata":{}},{"cell_type":"code","source":"df = dataframe_information(\"train_credit_bureau_b_2.parquet\")","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.453246Z","iopub.execute_input":"2024-04-28T11:51:09.453710Z","iopub.status.idle":"2024-04-28T11:51:09.642600Z","shell.execute_reply.started":"2024-04-28T11:51:09.453669Z","shell.execute_reply":"2024-04-28T11:51:09.641250Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df = df.with_columns(pl.col('pmts_date_1107D').cast(pl.Date))\n\ndata = df.group_by('case_id', 'num_group1', maintain_order=True).agg(\n    payment_count = pl.col('num_group2').count(),\n    payment_date_duration = pl.col('pmts_date_1107D').max() - pl.col('pmts_date_1107D').min(),\n    due_payment_sum = pl.col('pmts_dpdvalue_108P').sum(),\n    due_payment_max = pl.col('pmts_dpdvalue_108P').max(),\n    due_payment_last = pl.col('pmts_dpdvalue_108P').last(),\n    overdue_payment_sum = pl.col('pmts_pmtsoverdue_635A').sum(),\n    overdue_payment_max = pl.col('pmts_pmtsoverdue_635A').max(),\n    overdue_payment_last = pl.col('pmts_pmtsoverdue_635A').last(),\n)\n\ndata = data.fill_null(0.0)\ndata = data.fill_nan(0.0)\n\ndata = data.with_columns([\n    np.log(pl.col('due_payment_sum') + 1),\n    np.log(pl.col('overdue_payment_sum') + 1),\n    (pl.col('due_payment_sum') / pl.col('payment_date_duration')).alias('due_payment_mean1'),\n    (pl.col('due_payment_sum') / pl.col('payment_count')).alias('due_payment_mean2'),\n    (pl.col('overdue_payment_sum') / pl.col('payment_date_duration')).alias('overdue_payment_mean1'),\n    (pl.col('overdue_payment_sum') / pl.col('payment_count')).alias('overdue_payment_mean2'),\n])\n\ndata = data.fill_null(0.0)\ndata = data.fill_nan(0.0)\n\ndisplay(data.head(3))\n\ntarget_columns = list(filter(lambda x: x not in ['case_id', 'num_group1', 'num_group2'], data.columns))\n\ndata = data.group_by('case_id', maintain_order=True).agg(\n    [pl.col(col).max().alias(f'{col}_max') for col in target_columns]\n    + [pl.col(col).mean().alias(f'{col}_mean') for col in target_columns]\n    + [pl.col(col).sum().alias(f'{col}_sum') for col in target_columns]\n    + [pl.col('num_group1').count().alias('contract_count')]\n)\n\ndisplay(data.head(3))","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:09.644189Z","iopub.execute_input":"2024-04-28T11:51:09.645251Z","iopub.status.idle":"2024-04-28T11:51:10.167079Z","shell.execute_reply.started":"2024-04-28T11:51:09.645207Z","shell.execute_reply":"2024-04-28T11:51:10.166246Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# target_corr(data).head(10)","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:10.168133Z","iopub.execute_input":"2024-04-28T11:51:10.168632Z","iopub.status.idle":"2024-04-28T11:51:10.173527Z","shell.execute_reply.started":"2024-04-28T11:51:10.168603Z","shell.execute_reply":"2024-04-28T11:51:10.172369Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"select_cols = reduce_columns_by_correlation(data)\n\ndata.select(select_cols).write_parquet('train_credit_bureau_b_2.parquet')","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:10.175226Z","iopub.execute_input":"2024-04-28T11:51:10.175665Z","iopub.status.idle":"2024-04-28T11:51:10.328977Z","shell.execute_reply.started":"2024-04-28T11:51:10.175626Z","shell.execute_reply":"2024-04-28T11:51:10.327365Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def make_credit_bureau_b_2(is_train):\n    if is_train:\n        df = pl.read_parquet(TRAIN_DIR / \"train_credit_bureau_b_2.parquet\")\n    else:\n        df = pl.read_parquet(TEST_DIR / \"test_credit_bureau_b_2.parquet\")\n    df = df.with_columns(pl.col('pmts_date_1107D').cast(pl.Date))\n    \n    columns = ['case_id', 'payment_count_max', 'due_payment_sum_max', 'due_payment_max_max', 'overdue_payment_sum_max', 'overdue_payment_max_max', \\\n               'due_payment_mean1_max', 'due_payment_mean2_max', 'overdue_payment_mean1_max', 'due_payment_mean1_mean', 'overdue_payment_mean1_mean', \\\n               'payment_count_sum', 'due_payment_mean1_sum', 'overdue_payment_mean1_sum', 'contract_count']\n\n    data = df.group_by('case_id', 'num_group1', maintain_order=True).agg(\n        payment_count = pl.col('num_group2').count(),\n        payment_date_duration = pl.col('pmts_date_1107D').max() - pl.col('pmts_date_1107D').min(),\n        due_payment_sum = pl.col('pmts_dpdvalue_108P').sum(),\n        due_payment_max = pl.col('pmts_dpdvalue_108P').max(),\n        # due_payment_last = pl.col('pmts_dpdvalue_108P').last(),\n        overdue_payment_sum = pl.col('pmts_pmtsoverdue_635A').sum(),\n        overdue_payment_max = pl.col('pmts_pmtsoverdue_635A').max(),\n        # overdue_payment_last = pl.col('pmts_pmtsoverdue_635A').last(),\n    )\n\n    data = data.fill_null(0.0)\n    data = data.fill_nan(0.0)\n\n    data = data.with_columns([\n        np.log(pl.col('due_payment_sum') + 1),\n        np.log(pl.col('overdue_payment_sum') + 1),\n        (pl.col('due_payment_sum') / pl.col('payment_date_duration')).alias('due_payment_mean1'),\n        (pl.col('due_payment_sum') / pl.col('payment_count')).alias('due_payment_mean2'),\n        (pl.col('overdue_payment_sum') / pl.col('payment_date_duration')).alias('overdue_payment_mean1'),\n        # (pl.col('overdue_payment_sum') / pl.col('payment_count')).alias('overdue_payment_mean2'),\n    ])\n\n    data = data.fill_null(0.0)\n    data = data.fill_nan(0.0)\n\n    display(data.head(3))\n\n    target_columns = list(filter(lambda x: x not in ['case_id', 'num_group1', 'num_group2'], data.columns))\n\n    max_cols = []\n    mean_cols = []\n    sum_cols = []\n    for col in columns:\n        if col.endswith('_max'):\n            max_cols.append(col[:-4])\n        elif col.endswith('_mean'):\n            mean_cols.append(col[:-5])\n        elif col.endswith('_sum'):\n            sum_cols.append(col[:-4])\n\n    data = data.group_by('case_id', maintain_order=True).agg(\n        [pl.col(col).max().alias(f'{col}_max') for col in max_cols]\n        + [pl.col(col).mean().alias(f'{col}_mean') for col in mean_cols]\n        + [pl.col(col).sum().alias(f'{col}_sum') for col in sum_cols]\n        + [pl.col('num_group1').count().alias('contract_count')]\n    )\n\n    return data","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:10.330976Z","iopub.execute_input":"2024-04-28T11:51:10.332112Z","iopub.status.idle":"2024-04-28T11:51:10.352556Z","shell.execute_reply.started":"2024-04-28T11:51:10.332064Z","shell.execute_reply":"2024-04-28T11:51:10.351537Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"make_credit_bureau_b_2(False)","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:10.354037Z","iopub.execute_input":"2024-04-28T11:51:10.354406Z","iopub.status.idle":"2024-04-28T11:51:10.384248Z","shell.execute_reply.started":"2024-04-28T11:51:10.354375Z","shell.execute_reply":"2024-04-28T11:51:10.383132Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### train_applprev_2.parquet","metadata":{}},{"cell_type":"code","source":"df = dataframe_information(\"train_applprev_2.parquet\")","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:10.385898Z","iopub.execute_input":"2024-04-28T11:51:10.386466Z","iopub.status.idle":"2024-04-28T11:51:12.579580Z","shell.execute_reply.started":"2024-04-28T11:51:10.386396Z","shell.execute_reply":"2024-04-28T11:51:12.578390Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# print(df['cacccardblochreas_147M'].unique())\n# print(df['conts_type_509L'].unique())\n# print(df['credacc_cards_status_52L'].unique())","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:12.581026Z","iopub.execute_input":"2024-04-28T11:51:12.581454Z","iopub.status.idle":"2024-04-28T11:51:12.586983Z","shell.execute_reply.started":"2024-04-28T11:51:12.581394Z","shell.execute_reply":"2024-04-28T11:51:12.585708Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"contract_mapping = {\n                    \"PRIMARY_MOBILE\": \"MOBILE\", \"EMPLOYMENT_PHONE\": \"MOBILE\", \"PHONE\": \"MOBILE\", \"PRIMARY_EMAIL\": \"INTERNET\", \\\n                    \"HOME_PHONE\": \"MOBILE\", \"SECONDARY_MOBILE\": \"MOBILE\", \"ALTERNATIVE_PHONE\": \"MOBILE\", \\\n                    \"WHATSAPP\": \"APP\", \"SKYPE\": \"INTERNET\",\n                   }\n\ncard_mapping = {\n    \"CANCELLED\": \"INACTIVE\", \"INACTIVE\": \"INACTIVE\", \"RENEWED\": \"ACTIVE\", \"BLOCKED\": \"INACTIVE\", \\\n    \"UNCONFIRMED\": \"INACTIVE\", \"ACITVE\": \"ACTIVE\"\n}\n\ndata = df.with_columns(\n    pl.col('cacccardblochreas_147M').cast(pl.Categorical),\n    df['conts_type_509L'].replace(contract_mapping, default=None).cast(pl.Categorical),\n    df['credacc_cards_status_52L'].replace(card_mapping, default=None).cast(pl.Categorical)\n)\n\ndata = data.with_columns(\n     data.to_dummies(['cacccardblochreas_147M', 'conts_type_509L', 'credacc_cards_status_52L'])\n)\n\ndata = data.drop(['cacccardblochreas_147M_null', \\\n                  'conts_type_509L', 'conts_type_509L_null', \\\n                  'credacc_cards_status_52L', 'credacc_cards_status_52L_null'])\n\ncols = list(filter(lambda x: x not in ['case_id', 'num_group1', 'num_group2', 'cacccardblochreas_147M'], data.columns))\ndata = data.group_by('case_id', maintain_order=True).agg(\n    [pl.col(x).sum() for x in cols]\n    + [pl.col('cacccardblochreas_147M').count().alias('card_blocking_reason_count')]\n)\n\ndata","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:12.588765Z","iopub.execute_input":"2024-04-28T11:51:12.589671Z","iopub.status.idle":"2024-04-28T11:51:18.750285Z","shell.execute_reply.started":"2024-04-28T11:51:12.589636Z","shell.execute_reply":"2024-04-28T11:51:18.749127Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"select_cols = reduce_columns_by_correlation(data)\n\ndata.select(select_cols).write_parquet('train_applprev_2.parquet')","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:18.752019Z","iopub.execute_input":"2024-04-28T11:51:18.752799Z","iopub.status.idle":"2024-04-28T11:51:19.325131Z","shell.execute_reply.started":"2024-04-28T11:51:18.752746Z","shell.execute_reply":"2024-04-28T11:51:19.324162Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def make_applprev_2(is_train):\n    if is_train:\n        df = pl.read_parquet(TRAIN_DIR / \"train_applprev_2.parquet\")\n    else:\n        df = pl.read_parquet(TEST_DIR / \"test_applprev_2.parquet\")\n\n    contract_mapping = {\n                    \"PRIMARY_MOBILE\": \"MOBILE\", \"EMPLOYMENT_PHONE\": \"MOBILE\", \"PHONE\": \"MOBILE\", \"PRIMARY_EMAIL\": \"INTERNET\", \\\n                    \"HOME_PHONE\": \"MOBILE\", \"SECONDARY_MOBILE\": \"MOBILE\", \"ALTERNATIVE_PHONE\": \"MOBILE\", \\\n                    \"WHATSAPP\": \"APP\", \"SKYPE\": \"INTERNET\",\n                   }\n\n    card_mapping = {\n        \"CANCELLED\": \"INACTIVE\", \"INACTIVE\": \"INACTIVE\", \"RENEWED\": \"ACTIVE\", \"BLOCKED\": \"INACTIVE\", \\\n        \"UNCONFIRMED\": \"INACTIVE\", \"ACITVE\": \"ACTIVE\"\n    }\n\n    data = df.with_columns(\n        pl.col('cacccardblochreas_147M').cast(pl.Categorical),\n        df['conts_type_509L'].replace(contract_mapping, default=None).cast(pl.Categorical),\n        # df['credacc_cards_status_52L'].replace(card_mapping, default=None).cast(pl.Categorical) # fail when every value is None\n    )\n\n    data = data.to_dummies(['cacccardblochreas_147M', 'conts_type_509L', 'credacc_cards_status_52L'])\n    data = data.drop(['cacccardblochreas_147M_null', 'conts_type_509L_null', 'conts_type_509L_MOBILE', 'credacc_cards_status_52L_null'])\n\n    cols = list(filter(lambda x: x not in ['case_id', 'num_group1', 'num_group2'], data.columns))\n    data = data.group_by('case_id', maintain_order=True).agg(\n        [pl.col(x).sum() for x in cols]\n    )\n\n    return data","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:19.326689Z","iopub.execute_input":"2024-04-28T11:51:19.327333Z","iopub.status.idle":"2024-04-28T11:51:19.337029Z","shell.execute_reply.started":"2024-04-28T11:51:19.327302Z","shell.execute_reply":"2024-04-28T11:51:19.336122Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"make_applprev_2(False)","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:19.338391Z","iopub.execute_input":"2024-04-28T11:51:19.338915Z","iopub.status.idle":"2024-04-28T11:51:19.362922Z","shell.execute_reply.started":"2024-04-28T11:51:19.338886Z","shell.execute_reply":"2024-04-28T11:51:19.361972Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### train_person_2","metadata":{}},{"cell_type":"code","source":"# https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/476907\ndf = dataframe_information(\"train_person_2.parquet\")","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:19.364268Z","iopub.execute_input":"2024-04-28T11:51:19.365349Z","iopub.status.idle":"2024-04-28T11:51:19.967680Z","shell.execute_reply.started":"2024-04-28T11:51:19.365317Z","shell.execute_reply":"2024-04-28T11:51:19.966514Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def positive_category_any(col):\n    pos_categories = df[['case_id', col]] \\\n        .join(df_base[['case_id', 'target']], on='case_id') \\\n        .filter(pl.col('target')==1) \\\n        .select(pl.col(col)).unique()\n    \n    print(col)\n    if len(pos_categories) < 10:\n        print(f'{len(pos_categories)}, {pos_categories}')\n    else:\n        print(len(pos_categories))\n        \n    print(f'{len(pos_categories) / len(df[col].unique()) * 100:.2f}%\\n')\n    \n    return pos_categories\n\ntarget_address_district = positive_category_any('addres_district_368M')\ntarget_addres_role = positive_category_any('addres_role_871L')\ntarget_addres_zip = positive_category_any('addres_zip_823M')\ntarget_conts_role = positive_category_any('conts_role_79M')\ntarget_empls_economicalst = positive_category_any('empls_economicalst_849M')\ntarget_empls_employer_name = positive_category_any('empls_employer_name_740M')\ntarget_relatedpersons_role = positive_category_any('relatedpersons_role_762T')","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:19.969469Z","iopub.execute_input":"2024-04-28T11:51:19.969935Z","iopub.status.idle":"2024-04-28T11:51:21.641172Z","shell.execute_reply.started":"2024-04-28T11:51:19.969896Z","shell.execute_reply":"2024-04-28T11:51:21.640194Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data = df.with_columns(\n    pl.col('addres_district_368M').is_in(target_address_district),\n    pl.col('addres_role_871L').is_in(target_addres_role),\n    pl.col('empls_economicalst_849M').is_in(target_empls_economicalst),\n    pl.col('empls_employedfrom_796D').cast(pl.Date),\n)\n\ndata = data.filter(pl.col('num_group1')==0) \\\n        .filter(pl.col('addres_role_871L')) \\\n        .drop('num_group1', 'addres_zip_823M', 'empls_employer_name_740M')\n\ndata = data.group_by('case_id', maintain_order=True).agg(\n    pl.col('addres_district_368M').count().alias('num_group2'),\n    pl.col('addres_district_368M').any().name.prefix('any_'),\n    pl.col('addres_district_368M').all().name.prefix('all_'),\n    pl.col('empls_economicalst_849M').any().name.prefix('any_'),\n    pl.col('empls_economicalst_849M').all().name.prefix('all_'),\n    pl.col('empls_employedfrom_796D').first().name.prefix('first_'),\n    pl.col('empls_employedfrom_796D').last().name.prefix('last_'),\n)  \n\ndata","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:21.642744Z","iopub.execute_input":"2024-04-28T11:51:21.643241Z","iopub.status.idle":"2024-04-28T11:51:21.862201Z","shell.execute_reply.started":"2024-04-28T11:51:21.643201Z","shell.execute_reply":"2024-04-28T11:51:21.861222Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"select_cols = reduce_columns_by_correlation(data.drop('first_empls_employedfrom_796D', 'last_empls_employedfrom_796D'))\nselect_cols += ['first_empls_employedfrom_796D', 'last_empls_employedfrom_796D']\n\ndata.select(select_cols).write_parquet('train_person_2.parquet')","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:21.863531Z","iopub.execute_input":"2024-04-28T11:51:21.864637Z","iopub.status.idle":"2024-04-28T11:51:21.882020Z","shell.execute_reply.started":"2024-04-28T11:51:21.864599Z","shell.execute_reply":"2024-04-28T11:51:21.880917Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def make_train_2(is_train):\n    if is_train:\n        df = pl.read_parquet(TRAIN_DIR / \"train_person_2.parquet\")\n    else:\n        df = pl.read_parquet(TEST_DIR / \"test_person_2.parquet\")\n\n    data = df.with_columns(\n        pl.col('addres_district_368M').is_in(target_address_district),\n        pl.col('addres_role_871L').is_in(target_addres_role),\n        pl.col('empls_economicalst_849M').is_in(target_empls_economicalst),\n        pl.col('empls_employedfrom_796D').cast(pl.Date),\n    )\n\n    data = data.filter(pl.col('num_group1')==0) \\\n            .filter(pl.col('addres_role_871L')) \\\n            .drop('num_group1', 'addres_zip_823M', 'empls_employer_name_740M')\n\n    data = data.group_by('case_id', maintain_order=True).agg(\n        pl.col('addres_district_368M').count().alias('num_group2'),\n        pl.col('addres_district_368M').any().name.prefix('any_'),\n        pl.col('empls_economicalst_849M').any().name.prefix('any_'),\n        pl.col('empls_employedfrom_796D').first().name.prefix('first_'),\n        pl.col('empls_employedfrom_796D').last().name.prefix('last_'),\n    )  \n\n    return data","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:21.883919Z","iopub.execute_input":"2024-04-28T11:51:21.884791Z","iopub.status.idle":"2024-04-28T11:51:21.895861Z","shell.execute_reply.started":"2024-04-28T11:51:21.884758Z","shell.execute_reply":"2024-04-28T11:51:21.894393Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"target_address_district.write_parquet('target_address_district.parquet')","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:49.269857Z","iopub.execute_input":"2024-04-28T11:51:49.270309Z","iopub.status.idle":"2024-04-28T11:51:49.279328Z","shell.execute_reply.started":"2024-04-28T11:51:49.270275Z","shell.execute_reply":"2024-04-28T11:51:49.277820Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"make_train_2(False)","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:21.897544Z","iopub.execute_input":"2024-04-28T11:51:21.897956Z","iopub.status.idle":"2024-04-28T11:51:21.915863Z","shell.execute_reply.started":"2024-04-28T11:51:21.897910Z","shell.execute_reply":"2024-04-28T11:51:21.914222Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### train_credit_bureau_a_2_*.parquet","metadata":{}},{"cell_type":"code","source":"# # https://www.kaggle.com/competitions/home-credit-credit-risk-model-stability/discussion/478360\n# df = dataframe_information('train_credit_bureau_a_2_0.parquet')","metadata":{"execution":{"iopub.status.busy":"2024-04-28T11:51:21.917594Z","iopub.execute_input":"2024-04-28T11:51:21.918645Z","iopub.status.idle":"2024-04-28T11:51:21.924231Z","shell.execute_reply.started":"2024-04-28T11:51:21.918586Z","shell.execute_reply":"2024-04-28T11:51:21.922882Z"},"trusted":true},"execution_count":null,"outputs":[]}]}