{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.12","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":30635,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# Example Notebook\n\nWelcome to the example notebook for the Home Credit Kaggle competition. The goal of this competition is to determine how likely a customer is going to default on an issued loan. The main difference between the [first](https://www.kaggle.com/c/home-credit-default-risk) and this competition is that now your submission will be scored with a custom metric that will take into account how well the model performs in future. A decline in performance will be penalized. The goal is to create a model that is stable and performs well in the future.\n\nIn this notebook you will see how to:\n* Load the data\n* Join tables with Polars - a DataFrame library implemented in Rust language, designed to be blazingy fast and memory efficient.  \n* Create simple aggregation features\n* Train a LightGBM model\n* Create a submission table\n\n## Load the data","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19"}},{"cell_type":"markdown","source":"My name: Lutan Maksim","metadata":{}},{"cell_type":"markdown","source":"Original notebook: https://www.kaggle.com/code/jetakow/home-credit-2024-starter-notebook","metadata":{}},{"cell_type":"markdown","source":"Added new hyperparams based on what usually happened at humans life and what can be concluded from uor data. Also I have chosen some hyperparams. The next step is to check new models. Also I want to check every column on it's quality on the result. Also I want to make EDA better. Because I think there should be something better.","metadata":{}},{"cell_type":"code","source":"import polars as pl\nimport numpy as np\nimport pandas as pd\nimport lightgbm as lgb\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import roc_auc_score \n\ndataPath = \"/kaggle/input/home-credit-credit-risk-model-stability/\"","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:42:42.328097Z","iopub.execute_input":"2024-03-18T20:42:42.329246Z","iopub.status.idle":"2024-03-18T20:42:42.335478Z","shell.execute_reply.started":"2024-03-18T20:42:42.329190Z","shell.execute_reply":"2024-03-18T20:42:42.334013Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Adding my personal imports","metadata":{}},{"cell_type":"code","source":"import warnings\nwarnings.filterwarnings(\"ignore\")\nimport matplotlib.pyplot as plt\n\nimport polars as pl\n\nimport lightgbm as lgb\nfrom sklearn.metrics import (\n    accuracy_score,\n    roc_curve,\n    roc_auc_score\n)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:42:42.348159Z","iopub.execute_input":"2024-03-18T20:42:42.348858Z","iopub.status.idle":"2024-03-18T20:42:42.358466Z","shell.execute_reply.started":"2024-03-18T20:42:42.348817Z","shell.execute_reply":"2024-03-18T20:42:42.357335Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def set_table_dtypes(df: pl.DataFrame) -> pl.DataFrame:\n    # implement here all desired dtypes for tables\n    # the following is just an example\n    for col in df.columns:\n        # last letter of column name will help you determine the type\n        if col[-1] in (\"P\", \"A\"):\n            df = df.with_columns(pl.col(col).cast(pl.Float64).alias(col))\n\n    return df\n\ndef convert_strings(df: pd.DataFrame) -> pd.DataFrame:\n    for col in df.columns:  \n        if df[col].dtype.name in ['object', 'string']:\n            df[col] = df[col].astype(\"string\").astype('category')\n            current_categories = df[col].cat.categories\n            new_categories = current_categories.to_list() + [\"Unknown\"]\n            new_dtype = pd.CategoricalDtype(categories=new_categories, ordered=True)\n            df[col] = df[col].astype(new_dtype)\n    return df","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:42:42.373550Z","iopub.execute_input":"2024-03-18T20:42:42.373957Z","iopub.status.idle":"2024-03-18T20:42:42.382033Z","shell.execute_reply.started":"2024-03-18T20:42:42.373922Z","shell.execute_reply":"2024-03-18T20:42:42.381126Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_basetable = pl.read_csv(dataPath + \"csv_files/train/train_base.csv\")\ntrain_static = pl.concat(\n    [\n        pl.read_csv(dataPath + \"csv_files/train/train_static_0_0.csv\").pipe(set_table_dtypes),\n        pl.read_csv(dataPath + \"csv_files/train/train_static_0_1.csv\").pipe(set_table_dtypes),\n    ],\n    how=\"vertical_relaxed\",\n)\ntrain_static_cb = pl.read_csv(dataPath + \"csv_files/train/train_static_cb_0.csv\").pipe(set_table_dtypes)\ntrain_person_1 = pl.read_csv(dataPath + \"csv_files/train/train_person_1.csv\").pipe(set_table_dtypes) \ntrain_credit_bureau_b_2 = pl.read_csv(dataPath + \"csv_files/train/train_credit_bureau_b_2.csv\").pipe(set_table_dtypes) ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:42:42.404729Z","iopub.execute_input":"2024-03-18T20:42:42.405455Z","iopub.status.idle":"2024-03-18T20:42:59.907513Z","shell.execute_reply.started":"2024-03-18T20:42:42.405403Z","shell.execute_reply":"2024-03-18T20:42:59.906496Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_basetable = pl.read_csv(dataPath + \"csv_files/test/test_base.csv\")\ntest_static = pl.concat(\n    [\n        pl.read_csv(dataPath + \"csv_files/test/test_static_0_0.csv\").pipe(set_table_dtypes),\n        pl.read_csv(dataPath + \"csv_files/test/test_static_0_1.csv\").pipe(set_table_dtypes),\n        pl.read_csv(dataPath + \"csv_files/test/test_static_0_2.csv\").pipe(set_table_dtypes),\n    ],\n    how=\"vertical_relaxed\",\n)\ntest_static_cb = pl.read_csv(dataPath + \"csv_files/test/test_static_cb_0.csv\").pipe(set_table_dtypes)\ntest_person_1 = pl.read_csv(dataPath + \"csv_files/test/test_person_1.csv\").pipe(set_table_dtypes) \ntest_credit_bureau_b_2 = pl.read_csv(dataPath + \"csv_files/test/test_credit_bureau_b_2.csv\").pipe(set_table_dtypes) ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:42:59.909969Z","iopub.execute_input":"2024-03-18T20:42:59.910440Z","iopub.status.idle":"2024-03-18T20:42:59.959249Z","shell.execute_reply.started":"2024-03-18T20:42:59.910395Z","shell.execute_reply":"2024-03-18T20:42:59.958138Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Feature engineering\n\nIn this part, we can see a simple example of joining tables via `case_id`. Here the loading and joining is done with polars library. Polars library is blazingly fast and has much smaller memory footprint than pandas. ","metadata":{}},{"cell_type":"code","source":"# We need to use aggregation functions in tables with depth > 1, so tables that contain num_group1 column or \n# also num_group2 column.\ntrain_person_1_feats_1 = train_person_1.group_by(\"case_id\").agg(\n    pl.col(\"mainoccupationinc_384A\").max().alias(\"mainoccupationinc_384A_max\"),\n    (pl.col(\"incometype_1044T\") == \"SELFEMPLOYED\").max().alias(\"mainoccupationinc_384A_any_selfemployed\")\n)\n\n# Here num_group1=0 has special meaning, it is the person who applied for the loan.\ntrain_person_1_feats_2 = train_person_1.select([\"case_id\", \"num_group1\", \"housetype_905L\"]).filter(\n    pl.col(\"num_group1\") == 0\n).drop(\"num_group1\").rename({\"housetype_905L\": \"person_housetype\"})\n\n# Here we have num_goup1 and num_group2, so we need to aggregate again.\ntrain_credit_bureau_b_2_feats = train_credit_bureau_b_2.group_by(\"case_id\").agg(\n    pl.col(\"pmts_pmtsoverdue_635A\").max().alias(\"pmts_pmtsoverdue_635A_max\"),\n    (pl.col(\"pmts_dpdvalue_108P\") > 31).max().alias(\"pmts_dpdvalue_108P_over31\")\n)\n\n# We will process in this examples only A-type and M-type columns, so we need to select them.\nselected_static_cols = []\nfor col in train_static.columns:\n    if col[-1] in (\"A\", \"M\"):\n        selected_static_cols.append(col)\nprint(selected_static_cols)\n\nselected_static_cb_cols = []\nfor col in train_static_cb.columns:\n    if col[-1] in (\"A\", \"M\"):\n        selected_static_cb_cols.append(col)\nprint(selected_static_cb_cols)\n\n# Join all tables together.\ndata = train_basetable.join(\n    train_static.select([\"case_id\"]+selected_static_cols), how=\"left\", on=\"case_id\"\n).join(\n    train_static_cb.select([\"case_id\"]+selected_static_cb_cols), how=\"left\", on=\"case_id\"\n).join(\n    train_person_1_feats_1, how=\"left\", on=\"case_id\"\n).join(\n    train_person_1_feats_2, how=\"left\", on=\"case_id\"\n).join(\n    train_credit_bureau_b_2_feats, how=\"left\", on=\"case_id\"\n)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:42:59.960532Z","iopub.execute_input":"2024-03-18T20:42:59.961350Z","iopub.status.idle":"2024-03-18T20:43:02.261986Z","shell.execute_reply.started":"2024-03-18T20:42:59.961315Z","shell.execute_reply":"2024-03-18T20:43:02.261000Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_person_1_feats_1 = test_person_1.group_by(\"case_id\").agg(\n    pl.col(\"mainoccupationinc_384A\").max().alias(\"mainoccupationinc_384A_max\"),\n    (pl.col(\"incometype_1044T\") == \"SELFEMPLOYED\").max().alias(\"mainoccupationinc_384A_any_selfemployed\")\n)\n\ntest_person_1_feats_2 = test_person_1.select([\"case_id\", \"num_group1\", \"housetype_905L\"]).filter(\n    pl.col(\"num_group1\") == 0\n).drop(\"num_group1\").rename({\"housetype_905L\": \"person_housetype\"})\n\ntest_credit_bureau_b_2_feats = test_credit_bureau_b_2.group_by(\"case_id\").agg(\n    pl.col(\"pmts_pmtsoverdue_635A\").max().alias(\"pmts_pmtsoverdue_635A_max\"),\n    (pl.col(\"pmts_dpdvalue_108P\") > 31).max().alias(\"pmts_dpdvalue_108P_over31\")\n)\n\ndata_submission = test_basetable.join(\n    test_static.select([\"case_id\"]+selected_static_cols), how=\"left\", on=\"case_id\"\n).join(\n    test_static_cb.select([\"case_id\"]+selected_static_cb_cols), how=\"left\", on=\"case_id\"\n).join(\n    test_person_1_feats_1, how=\"left\", on=\"case_id\"\n).join(\n    test_person_1_feats_2, how=\"left\", on=\"case_id\"\n).join(\n    test_credit_bureau_b_2_feats, how=\"left\", on=\"case_id\"\n)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:43:02.264424Z","iopub.execute_input":"2024-03-18T20:43:02.264767Z","iopub.status.idle":"2024-03-18T20:43:02.282100Z","shell.execute_reply.started":"2024-03-18T20:43:02.264736Z","shell.execute_reply":"2024-03-18T20:43:02.281246Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# new params","metadata":{}},{"cell_type":"markdown","source":"Here we are adding NEW columns for new data and better quality of the code","metadata":{}},{"cell_type":"code","source":"def add_custom_features(df):\n    try:\n        df['cur_to_max_ratio'] = df['currdebt_22A'] / df['maxdebt4_972A']\n    except:\n        df['cur_to_max_ratio'] = 0\n    try:\n        df['est_delta'] = df['sumoutstandtotalest_4493215A'] - df['currdebt_22A']\n    except:\n        df['est_delta'] = 0\n    try:\n        df['got_married'] = (df['was_singlE'] == True) & (df['is_marrieD'] == True)\n    except:\n        df['got_married'] = 0\n    try:\n        df['children_increment'] = df['num_children_noW'] - df['num_of_children_beforE']\n    except:\n        df['children_increment'] = 0\n    try:\n        df['got_divorced'] = (df['was_marrieD'] == True) & (df['is_singlE'] == True)        \n    except:\n        df['got_divorced'] = 0\n\n\n    return df","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:43:02.283365Z","iopub.execute_input":"2024-03-18T20:43:02.283877Z","iopub.status.idle":"2024-03-18T20:43:02.291834Z","shell.execute_reply.started":"2024-03-18T20:43:02.283845Z","shell.execute_reply":"2024-03-18T20:43:02.290732Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"H","metadata":{}},{"cell_type":"code","source":"case_ids = data[\"case_id\"].unique().shuffle(seed=1)\ncase_ids_train, case_ids_test = train_test_split(case_ids, train_size=0.6, random_state=1)\ncase_ids_valid, case_ids_test = train_test_split(case_ids_test, train_size=0.5, random_state=1)\n\ncols_pred = []\nfor col in data.columns:\n    if col[-1].isupper() and col[:-1].islower():\n        cols_pred.append(col)\n\nprint(cols_pred)\n\ndef from_polars_to_pandas(case_ids: pl.DataFrame) -> pl.DataFrame:\n    return (\n        data.filter(pl.col(\"case_id\").is_in(case_ids))[[\"case_id\", \"WEEK_NUM\", \"target\"]].to_pandas(),\n        data.filter(pl.col(\"case_id\").is_in(case_ids))[cols_pred].to_pandas(),\n        data.filter(pl.col(\"case_id\").is_in(case_ids))[\"target\"].to_pandas()\n    )\n\nbase_train, X_train, y_train = from_polars_to_pandas(case_ids_train)\nbase_valid, X_valid, y_valid = from_polars_to_pandas(case_ids_valid)\nbase_test, X_test, y_test = from_polars_to_pandas(case_ids_test)\n\nfor df in [X_train, X_valid, X_test]:\n    df = convert_strings(df)\n    df = add_custom_features(df)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:43:02.293377Z","iopub.execute_input":"2024-03-18T20:43:02.293791Z","iopub.status.idle":"2024-03-18T20:43:10.217642Z","shell.execute_reply.started":"2024-03-18T20:43:02.293752Z","shell.execute_reply":"2024-03-18T20:43:10.216618Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(f\"Train: {X_train.shape}\")\nprint(f\"Valid: {X_valid.shape}\")\nprint(f\"Test: {X_test.shape}\")","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:43:10.219199Z","iopub.execute_input":"2024-03-18T20:43:10.219554Z","iopub.status.idle":"2024-03-18T20:43:10.229328Z","shell.execute_reply.started":"2024-03-18T20:43:10.219514Z","shell.execute_reply":"2024-03-18T20:43:10.228106Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Training LightGBM\n\nMinimal example of LightGBM training is shown below.","metadata":{}},{"cell_type":"code","source":"lgb_train = lgb.Dataset(X_train, label=y_train)\nlgb_valid = lgb.Dataset(X_valid, label=y_valid, reference=lgb_train)\n\nparams = {\n    \"boosting_type\": \"gbdt\",\n    \"objective\": \"binary\",\n    \"metric\": \"auc\",\n    \"max_depth\": 5,\n    \"num_leaves\": 31,\n    \"learning_rate\": 0.05,\n    \"feature_fraction\": 0.9,\n    \"bagging_fraction\": 0.8,\n    \"bagging_freq\": 5,\n    \"n_estimators\": 2000,\n    \"verbose\": -1,\n}\ngbm = lgb.train(\n    params,\n    lgb_train,\n    valid_sets=lgb_valid,\n    callbacks=[lgb.log_evaluation(50), lgb.early_stopping(10)]\n)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:43:10.231284Z","iopub.execute_input":"2024-03-18T20:43:10.231733Z","iopub.status.idle":"2024-03-18T20:44:17.543502Z","shell.execute_reply.started":"2024-03-18T20:43:10.231697Z","shell.execute_reply":"2024-03-18T20:44:17.542305Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"First of all, let's convert our dataset, and create validation part","metadata":{}},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"code","source":"path_data = \"/kaggle/input/home-credit-credit-risk-model-stability/\"","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:17.544979Z","iopub.execute_input":"2024-03-18T20:44:17.545347Z","iopub.status.idle":"2024-03-18T20:44:17.550262Z","shell.execute_reply.started":"2024-03-18T20:44:17.545315Z","shell.execute_reply":"2024-03-18T20:44:17.548831Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def set_pl_dtypes(df):\n    for col in df.columns:\n        if col[-1] in (\"P\", \"A\"):\n            df = df.with_columns(pl.col(col).cast(pl.Float64))\n        if col[-1] in (\"M\"):\n            df = df.with_columns(pl.col(col).cast(pl.Categorical))\n        if col[-1] in (\"D\"):\n            df = df.with_columns(pl.col(col).cast(pl.Date))\n    return df","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:17.554331Z","iopub.execute_input":"2024-03-18T20:44:17.554659Z","iopub.status.idle":"2024-03-18T20:44:17.563807Z","shell.execute_reply.started":"2024-03-18T20:44:17.554631Z","shell.execute_reply":"2024-03-18T20:44:17.562818Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train_base = pl.read_csv(path_data + \"csv_files/train/train_base.csv\")\ndf_train_static = pl.concat([pl.read_csv(path_data + \"csv_files/train/train_static_0_0.csv\").pipe(set_pl_dtypes),\n                             pl.read_csv(path_data + \"csv_files/train/train_static_0_1.csv\").pipe(set_pl_dtypes)],\n                            how=\"vertical_relaxed\")\n\ndf_train_static_cb = pl.read_csv(path_data + \"csv_files/train/train_static_cb_0.csv\").pipe(set_pl_dtypes)\n\ndf_train_person_1 = pl.read_csv(path_data + \"csv_files/train/train_person_1.csv\").pipe(set_pl_dtypes)\n\ndf_train_credit_bureau_b_2 = pl.read_csv(path_data + \"csv_files/train/train_credit_bureau_b_2.csv\").pipe(set_pl_dtypes)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:17.565203Z","iopub.execute_input":"2024-03-18T20:44:17.565566Z","iopub.status.idle":"2024-03-18T20:44:35.956439Z","shell.execute_reply.started":"2024-03-18T20:44:17.565536Z","shell.execute_reply":"2024-03-18T20:44:35.955262Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test_base = pl.read_csv(path_data + \"csv_files/test/test_base.csv\")\n\ndf_test_static = pl.concat(\n    [pl.read_csv(path_data + \"csv_files/test/test_static_0_0.csv\").pipe(set_pl_dtypes),\n     pl.read_csv(path_data + \"csv_files/test/test_static_0_1.csv\").pipe(set_pl_dtypes),\n     pl.read_csv(path_data + \"csv_files/test/test_static_0_2.csv\").pipe(set_pl_dtypes)],\n    how=\"vertical_relaxed\")\ndf_test_static_cb = pl.read_csv(path_data + \"csv_files/test/test_static_cb_0.csv\").pipe(set_pl_dtypes)\n\ndf_test_person_1 = pl.read_csv(path_data + \"csv_files/test/test_person_1.csv\").pipe(set_pl_dtypes)\n\ndf_test_credit_bureau_b_2 = pl.read_csv(path_data + \"csv_files/test/test_credit_bureau_b_2.csv\").pipe(set_pl_dtypes)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:35.958512Z","iopub.execute_input":"2024-03-18T20:44:35.958855Z","iopub.status.idle":"2024-03-18T20:44:36.027274Z","shell.execute_reply.started":"2024-03-18T20:44:35.958824Z","shell.execute_reply":"2024-03-18T20:44:36.025724Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cols_static_selec = []\nfor col in df_train_static.columns:\n    if col[-1] in (\"A\", \"M\"):\n        cols_static_selec.append(col)\n\ncols_static_cb_selec = []\nfor col in df_train_static_cb.columns:\n    if col[-1] in (\"A\", \"M\"):\n        cols_static_cb_selec.append(col)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:36.029257Z","iopub.execute_input":"2024-03-18T20:44:36.029662Z","iopub.status.idle":"2024-03-18T20:44:36.037028Z","shell.execute_reply.started":"2024-03-18T20:44:36.029623Z","shell.execute_reply":"2024-03-18T20:44:36.035526Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train_person_1_group = df_train_person_1.group_by(\"case_id\")\ndf_train_person_1_1 = df_train_person_1_group.agg(\n    pl.col(\"mainoccupationinc_384A\").max().alias(\"mainoccupationinc_max_A\"),\n    (pl.col(\"incometype_1044T\") == \"SELFEMPLOYED\").any().alias(\"anyselfemployed_T\")\n)\ndf_train_person_1_2 = df_train_person_1.select(\n    [\"case_id\", \"num_group1\", \"housetype_905L\"]\n).filter(pl.col(\"num_group1\") == 0).drop(\"num_group1\").rename(\n    {\"housetype_905L\": \"housetype_applicant_L\"}\n)\ndf_train_credit_bureau_b_2 = df_train_credit_bureau_b_2.group_by(\"case_id\").agg(\n    pl.col(\"pmts_pmtsoverdue_635A\").max().alias(\"pmts_pmtsoverdue_max_A\"),\n    (pl.col(\"pmts_dpdvalue_108P\") > 31).any().alias(\"pmts_dpdvalue_anyover31_P\")\n)\ndf_train = df_train_base\\\n.join(df_train_static.select([\"case_id\"] + cols_static_selec),\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_train_static_cb.select([\"case_id\"] + cols_static_cb_selec),\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_train_person_1_1,\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_train_person_1_2,\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_train_credit_bureau_b_2,\n      how=\"left\",\n      on=\"case_id\")","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:36.038623Z","iopub.execute_input":"2024-03-18T20:44:36.039141Z","iopub.status.idle":"2024-03-18T20:44:39.805793Z","shell.execute_reply.started":"2024-03-18T20:44:36.039095Z","shell.execute_reply":"2024-03-18T20:44:39.804679Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test_person_1_group = df_test_person_1.group_by(\"case_id\")\ndf_test_person_1_1 = df_test_person_1_group.agg(\n    pl.col(\"mainoccupationinc_384A\").max().alias(\"mainoccupationinc_max_A\"),\n    (pl.col(\"incometype_1044T\") == \"SELFEMPLOYED\").any().alias(\"anyselfemployed_T\")\n)\n\ndf_test_person_1_2 = df_test_person_1.select(\n    [\"case_id\", \"num_group1\", \"housetype_905L\"]\n).filter(pl.col(\"num_group1\") == 0).drop(\"num_group1\").rename(\n    {\"housetype_905L\": \"housetype_applicant_L\"}\n)\n\n\ndf_test_credit_bureau_b_2 = df_test_credit_bureau_b_2.group_by(\"case_id\").agg(\n    pl.col(\"pmts_pmtsoverdue_635A\").max().alias(\"pmts_pmtsoverdue_max_A\"),\n    (pl.col(\"pmts_dpdvalue_108P\") > 31).any().alias(\"pmts_dpdvalue_anyover31_P\")\n)\n\ndf_submission = df_test_base\\\n.join(df_test_static.select([\"case_id\"] + cols_static_selec),\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_test_static_cb.select([\"case_id\"] + cols_static_cb_selec),\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_test_person_1_1,\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_test_person_1_2,\n      how=\"left\",\n      on=\"case_id\")\\\n.join(df_test_credit_bureau_b_2,\n      how=\"left\",\n      on=\"case_id\")","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:39.807043Z","iopub.execute_input":"2024-03-18T20:44:39.807521Z","iopub.status.idle":"2024-03-18T20:44:39.827905Z","shell.execute_reply.started":"2024-03-18T20:44:39.807467Z","shell.execute_reply":"2024-03-18T20:44:39.826463Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_valid_test_split = {\n    \"valid_frac\": 0.2,\n    \"test_frac\": 0.2\n}","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:39.829602Z","iopub.execute_input":"2024-03-18T20:44:39.829967Z","iopub.status.idle":"2024-03-18T20:44:39.837853Z","shell.execute_reply.started":"2024-03-18T20:44:39.829936Z","shell.execute_reply":"2024-03-18T20:44:39.836598Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"case_id_train_old = df_train[\"case_id\"].unique().shuffle(seed=42)\nN_train_old = len(case_id_train_old)\nN_train = int((1 - train_valid_test_split[\"valid_frac\"] -\n               train_valid_test_split[\"test_frac\"]) * N_train_old)\nN_valid = int(train_valid_test_split[\"valid_frac\"] * N_train_old)\ncase_id_train = case_id_train_old.head(N_train)\ncase_id_valid = case_id_train_old.tail(-N_train).head(N_valid)\ncase_id_test = case_id_train_old.tail(-N_train).tail(-N_valid)\ncase_id_submission = df_submission[\"case_id\"].unique()","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:39.839488Z","iopub.execute_input":"2024-03-18T20:44:39.839859Z","iopub.status.idle":"2024-03-18T20:44:39.945262Z","shell.execute_reply.started":"2024-03-18T20:44:39.839827Z","shell.execute_reply":"2024-03-18T20:44:39.944083Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def from_polars_to_pandas(df, case_id, cols_x):\n    df_case_id = df.filter(pl.col(\"case_id\").is_in(case_id))\n    \n    dt = {\n        \"base\": df_case_id[[\"case_id\", \"WEEK_NUM\"]].to_pandas(),\n        \"x\": df_case_id[cols_x].to_pandas()\n    }\n    \n    if \"target\" in df_case_id.columns:\n        dt[\"base\"].insert(loc=dt[\"base\"].shape[1], column=\"y\", value=df_case_id[\"target\"])\n        dt[\"y\"] = df_case_id[\"target\"].to_pandas()\n\n    return dt\n\ndef convert_cols_obj_to_cols_cat(*dfs):\n    cols_object = list(set().union(*(df.select_dtypes(include=[\"object\"]).columns for df in dfs)))\n    for col in cols_object:\n        for df in dfs:\n            df[col] = df[col].astype(\"category\")\n            new_dtype = pd.CategoricalDtype(categories=df[col].cat.categories.to_list() +\n                                            [\"Unknown\"],\n                                            ordered=True)\n            df[col] = df[col].astype(new_dtype)\n    return dfs\n\ndef make_cat_excl_unknown(df, df_ref):\n    for col in df_ref.select_dtypes(include=[\"category\"]).columns:\n        cat_ref = df_ref[col].cat.categories.to_list()\n        cat = df[col].cat.categories.to_list()\n        cat_common = list(set(cat).intersection(cat_ref))\n        cat_exc = list(set(cat).difference(cat_common))\n        new_dtype = pd.CategoricalDtype(categories=cat_common,\n                                        ordered=True)\n        df[col] = df[col].replace(to_replace=cat_exc, value=\"Unknown\")\n        df[col] = df[col].astype(new_dtype)\n    return df","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:39.946928Z","iopub.execute_input":"2024-03-18T20:44:39.947300Z","iopub.status.idle":"2024-03-18T20:44:39.959993Z","shell.execute_reply.started":"2024-03-18T20:44:39.947267Z","shell.execute_reply":"2024-03-18T20:44:39.958517Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"case_id_train_old = df_train[\"case_id\"].unique().shuffle(seed=42)\n\nN_train_old = len(case_id_train_old)\n\nN_train = int((1 - train_valid_test_split[\"valid_frac\"] -\n               train_valid_test_split[\"test_frac\"]) * N_train_old)\n\nN_valid = int(train_valid_test_split[\"valid_frac\"] * N_train_old)\n\ncase_id_train = case_id_train_old.head(N_train)\n\ncase_id_valid = case_id_train_old.tail(-N_train).head(N_valid)\n\ncase_id_test = case_id_train_old.tail(-N_train).tail(-N_valid)\n\ncase_id_submission = df_submission[\"case_id\"].unique()","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:39.961649Z","iopub.execute_input":"2024-03-18T20:44:39.962554Z","iopub.status.idle":"2024-03-18T20:44:40.069555Z","shell.execute_reply.started":"2024-03-18T20:44:39.962515Z","shell.execute_reply":"2024-03-18T20:44:40.068114Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cols_x = []\nfor col in df_train.columns:\n    if col[-1].isupper() and col[:-1].islower():\n        cols_x.append(col)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:40.071016Z","iopub.execute_input":"2024-03-18T20:44:40.071924Z","iopub.status.idle":"2024-03-18T20:44:40.077578Z","shell.execute_reply.started":"2024-03-18T20:44:40.071885Z","shell.execute_reply":"2024-03-18T20:44:40.076262Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dt_train = from_polars_to_pandas(df_train, case_id_train, cols_x)\ndt_valid = from_polars_to_pandas(df_train, case_id_valid, cols_x)\ndt_test = from_polars_to_pandas(df_train, case_id_test, cols_x)\ndt_submission = from_polars_to_pandas(df_submission, case_id_submission, cols_x)\n\n(dt_train[\"x\"], dt_valid[\"x\"], dt_test[\"x\"], dt_submission[\"x\"]) = convert_cols_obj_to_cols_cat(\n    dt_train[\"x\"], dt_valid[\"x\"], dt_test[\"x\"], dt_submission[\"x\"]\n)\ndt_valid[\"x\"] = make_cat_excl_unknown(df=dt_valid[\"x\"], df_ref=dt_train[\"x\"])\ndt_test[\"x\"] = make_cat_excl_unknown(df=dt_test[\"x\"], df_ref=dt_train[\"x\"])\ndt_submission[\"x\"] = make_cat_excl_unknown(df=dt_submission[\"x\"], df_ref=dt_train[\"x\"])\n\ndisplay(pd.DataFrame(data=\n                     {\"Feature dataset\": [\"train\", \"valid\", \"test\", \"submission\"],\n                      \"N_rows\": [dt[\"x\"].shape[0]for\n                                          dt in (dt_train, dt_valid, dt_test, dt_submission)],\n                      \"N_cols\": [dt[\"x\"].shape[1]for\n                                          dt in (dt_train, dt_valid, dt_test, dt_submission)]\n                     }))","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:40.079233Z","iopub.execute_input":"2024-03-18T20:44:40.079687Z","iopub.status.idle":"2024-03-18T20:44:41.793169Z","shell.execute_reply.started":"2024-03-18T20:44:40.079644Z","shell.execute_reply":"2024-03-18T20:44:41.792198Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"We have finished with our splitting? now let's train our model","metadata":{}},{"cell_type":"markdown","source":"Let's select hyperparams","metadata":{}},{"cell_type":"code","source":"print(df_train.columns)\n# X_train = df_train.drop(\"y\")\n# y_train = df_train[\"y\"]","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:41.794677Z","iopub.execute_input":"2024-03-18T20:44:41.795330Z","iopub.status.idle":"2024-03-18T20:44:41.801297Z","shell.execute_reply.started":"2024-03-18T20:44:41.795291Z","shell.execute_reply":"2024-03-18T20:44:41.799911Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# new hyperparams","metadata":{}},{"cell_type":"code","source":"from hyperopt import hp, fmin, tpe, Trials, STATUS_OK\nfrom sklearn.model_selection import cross_val_score\n\n\nlgb_cls_params = {\n    'learning_rate':    hp.uniform('learning_rate', 0.1, 1),\n    'max_depth':        hp.choice('max_depth', np.arange(2, 100, 1, dtype=int)),\n    'min_child_weight': hp.choice('min_child_weight', np.arange(1, 50, 1, dtype=int)),\n    'colsample_bytree': hp.uniform('colsample_bytree', 0.4, 1),\n    'subsample':        hp.uniform('subsample', 0.6, 1),\n    'num_leaves':       hp.choice('num_leaves', np.arange(1, 200, 1, dtype=int)),\n    'min_split_gain':   hp.uniform('min_split_gain', 0, 1),\n    'reg_alpha':        hp.uniform('reg_alpha', 0, 1),\n    'reg_lambda':       hp.uniform('reg_lambda', 0, 1),\n    'n_estimators':     5,\n    'objective':        'binary',  # or 'multiclass' for multiclass classification\n    'metric':           'binary_logloss',  # or 'multi_logloss' for multiclass classification\n    'boosting_type':    'gbdt'  # Gradient Boosting Decision Tree\n}\n\n\ndef f(params):\n    lgbm = lgb.LGBMClassifier(n_jobs=-1,early_stopping_rounds=None,**params)\n    score = cross_val_score(lgbm, X_train, y_train, cv=2,n_jobs=-1).mean()\n    return score\n\ntrials = Trials()\nresult = fmin(\n    fn=f,                           # objective function\n    space=lgb_cls_params,           # parameter space\n    algo=tpe.suggest,               # surrogate algorithm\n    max_evals=50,                   # no. of evaluations\n    trials=trials\n)\nprint(result)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:44:41.802890Z","iopub.execute_input":"2024-03-18T20:44:41.803287Z","iopub.status.idle":"2024-03-18T20:49:56.328808Z","shell.execute_reply.started":"2024-03-18T20:44:41.803251Z","shell.execute_reply":"2024-03-18T20:49:56.327994Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"best_params = result\nbest_params['metric'] = [\"auc\", \"cross_entropy\"]","metadata":{"execution":{"iopub.status.busy":"2024-03-18T20:49:56.330245Z","iopub.execute_input":"2024-03-18T20:49:56.331416Z","iopub.status.idle":"2024-03-18T20:49:56.337010Z","shell.execute_reply.started":"2024-03-18T20:49:56.331368Z","shell.execute_reply":"2024-03-18T20:49:56.335754Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"\n\nparams = {\n    \"boosting_type\": \"gbdt\",\n    \"objective\": \"binary\",\n    \"metric\": \"auc\",\n    \"max_depth\": 6,\n    \"num_leaves\": 152,\n    \"learning_rate\": 0.05,\n    \"feature_fraction\": 0.9,\n    \"bagging_fraction\": 0.8,\n    \"bagging_freq\": 5,\n    \"n_estimators\": 2000,\n    \"verbose\": -1,\n}\n\n\nmodel_eval_result = {}\n\nmodel = lgb.train(\n    params=params,\n    train_set=lgb_train,\n    valid_sets=[lgb_valid, lgb_train],\n    valid_names=[\"valid\", \"train\"],\n    callbacks=[\n        lgb.log_evaluation(period=50),\n        lgb.early_stopping(\n            first_metric_only=True,\n            stopping_rounds=10,\n            verbose=True),\n        lgb.record_evaluation(eval_result=model_eval_result)\n    ]\n)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T21:13:26.208006Z","iopub.execute_input":"2024-03-18T21:13:26.208520Z","iopub.status.idle":"2024-03-18T21:14:22.697011Z","shell.execute_reply.started":"2024-03-18T21:13:26.208477Z","shell.execute_reply":"2024-03-18T21:14:22.695131Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"We checked and updated our hyperparams","metadata":{}},{"cell_type":"markdown","source":"Evaluation with AUC and then comparison with the stability metric is shown below.","metadata":{}},{"cell_type":"code","source":"for base, X in [(base_train, X_train), (base_valid, X_valid), (base_test, X_test)]:\n    y_pred = gbm.predict(X, num_iteration=gbm.best_iteration)\n    base[\"score\"] = y_pred\n\nprint(f'The AUC score on the train set is: {roc_auc_score(base_train[\"target\"], base_train[\"score\"])}') \nprint(f'The AUC score on the valid set is: {roc_auc_score(base_valid[\"target\"], base_valid[\"score\"])}') \nprint(f'The AUC score on the test set is: {roc_auc_score(base_test[\"target\"], base_test[\"score\"])}')  ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T21:15:02.503670Z","iopub.execute_input":"2024-03-18T21:15:02.504695Z","iopub.status.idle":"2024-03-18T21:15:29.353908Z","shell.execute_reply.started":"2024-03-18T21:15:02.504652Z","shell.execute_reply":"2024-03-18T21:15:29.352610Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def gini_stability(base, w_fallingrate=88.0, w_resstd=-0.5):\n    gini_in_time = base.loc[:, [\"WEEK_NUM\", \"target\", \"score\"]]\\\n        .sort_values(\"WEEK_NUM\")\\\n        .groupby(\"WEEK_NUM\")[[\"target\", \"score\"]]\\\n        .apply(lambda x: 2*roc_auc_score(x[\"target\"], x[\"score\"])-1).tolist()\n    \n    x = np.arange(len(gini_in_time))\n    y = gini_in_time\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    avg_gini = np.mean(gini_in_time)\n    return avg_gini + w_fallingrate * min(0, a) + w_resstd * res_std\n\nstability_score_train = gini_stability(base_train)\nstability_score_valid = gini_stability(base_valid)\nstability_score_test = gini_stability(base_test)\n\nprint(f'The stability score on the train set is: {stability_score_train}') \nprint(f'The stability score on the valid set is: {stability_score_valid}') \nprint(f'The stability score on the test set is: {stability_score_test}')  ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T21:15:29.355979Z","iopub.execute_input":"2024-03-18T21:15:29.357065Z","iopub.status.idle":"2024-03-18T21:15:30.723613Z","shell.execute_reply.started":"2024-03-18T21:15:29.357023Z","shell.execute_reply":"2024-03-18T21:15:30.722459Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Submission\n\nScoring the submission dataset is below, we need to take care of new categories. Then we save the score as a last step. ","metadata":{}},{"cell_type":"code","source":"X_submission = data_submission[cols_pred].to_pandas()\nX_submission = add_custom_features(convert_strings(X_submission))\ncategorical_cols = X_train.select_dtypes(include=['category']).columns\n\nfor col in categorical_cols:\n    train_categories = set(X_train[col].cat.categories)\n    submission_categories = set(X_submission[col].cat.categories)\n    new_categories = submission_categories - train_categories\n    X_submission.loc[X_submission[col].isin(new_categories), col] = \"Unknown\"\n    new_dtype = pd.CategoricalDtype(categories=train_categories, ordered=True)\n    X_train[col] = X_train[col].astype(new_dtype)\n    X_submission[col] = X_submission[col].astype(new_dtype)\n\ny_submission_pred = gbm.predict(X_submission, num_iteration=gbm.best_iteration)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T21:15:30.725089Z","iopub.execute_input":"2024-03-18T21:15:30.725761Z","iopub.status.idle":"2024-03-18T21:15:30.827390Z","shell.execute_reply.started":"2024-03-18T21:15:30.725713Z","shell.execute_reply":"2024-03-18T21:15:30.826257Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission = pd.DataFrame({\n    \"case_id\": data_submission[\"case_id\"].to_numpy(),\n    \"score\": y_submission_pred\n}).set_index('case_id')\nsubmission.to_csv(\"./submission.csv\")","metadata":{"execution":{"iopub.status.busy":"2024-03-18T21:15:30.829903Z","iopub.execute_input":"2024-03-18T21:15:30.830394Z","iopub.status.idle":"2024-03-18T21:15:30.838546Z","shell.execute_reply.started":"2024-03-18T21:15:30.830348Z","shell.execute_reply":"2024-03-18T21:15:30.837400Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Let's add something complete","metadata":{}},{"cell_type":"markdown","source":"TL-DR: we have added some features in our solution and updated hyperparams. That's why our quality grew up. THANKS FOR ATTENTION!","metadata":{}}]}