{"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":"# HSE, CDM: HW2\n\n## What was changed?\n\n1. Date columns in the `test_base` table, specifically `date_decision`, `MONTH`, and `WEEK_NUM`, have been altered.\n2. `MONTH` and `WEEK_NUM` columns now hold only one constant value.\n3. `date_decision` has been modified and is no longer the same.\n4. The remaining data in the dataset have been transformed.\n5. Data types remain unchanged.\n6. Authirs didn’t find a metric that would satisfy both requirements and that is why they decided to keep the metric as it is now.\n\n\n## What I added?\n\nThe original notebook included only P and A features. I added columns with dates (D) to the train and test, resulting in a measurable improvement to 0.45. To enable LightGBM to work with date values, I subtracted the date_decision column to work with them as integers (difference in days).\n\nI also added various other features from `train_person_1`. For example, `sex_738L` is the gender of the applicant, `language1_981M` is the primary language of the applicant, and `empl_industry_691L` is the industry in which the applicant worked. I also summed up the number of children in the household and added total income. I empirically reviewed which features could make sense when considering credit risk.\n\nI also added features from the `train_tax_registry_b_1` table since it contains information on tax deductions and masked names of employers. Moreover, I added a set of features from `train_static_cb`, for instance, `for3years_584L` – how often the applicant was denied in the last three years.\n\nPotentially, it makes sense to explore the `test_credit_bureau_a` table further, as it contains several useful features, such as the past due amount for an active contract. Unfortunately, I ran into RAM limitations, but adding these features improved the score locally.\n\nAdditionally, I used optuna to select hyperparameters.","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19"}},{"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 \nimport matplotlib.pyplot as plt\n\ndataPath = \"/kaggle/input/home-credit-credit-risk-model-stability/\"","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:21:47.503285Z","iopub.execute_input":"2024-03-18T10:21:47.503990Z","iopub.status.idle":"2024-03-18T10:21:47.517077Z","shell.execute_reply.started":"2024-03-18T10:21:47.503942Z","shell.execute_reply":"2024-03-18T10:21:47.515551Z"},"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        if col[-1] in (\"D\"):\n            df = df.with_columns(pl.col(col).cast(pl.Date))\n        if col[-1] in (\"M\", \"L\", \"T\"):\n            df = df.with_columns(pl.col(col).cast(pl.String))\n        if col in (\"date_decision\"):\n            df = df.with_columns(pl.col(col).cast(pl.Date))\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":"2024-03-18T10:21:47.789294Z","iopub.execute_input":"2024-03-18T10:21:47.790519Z","iopub.status.idle":"2024-03-18T10:21:47.804452Z","shell.execute_reply.started":"2024-03-18T10:21:47.790469Z","shell.execute_reply":"2024-03-18T10:21:47.802490Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_basetable = pl.read_parquet(dataPath + \"parquet_files/train/train_base.parquet\").pipe(set_table_dtypes)\ntrain_static = pl.concat(\n    [\n        pl.read_parquet(dataPath + \"parquet_files/train/train_static_0_0.parquet\").pipe(set_table_dtypes),\n        pl.read_parquet(dataPath + \"parquet_files/train/train_static_0_1.parquet\").pipe(set_table_dtypes),\n    ],\n    how=\"vertical_relaxed\",\n)\ntrain_static_cb = pl.read_parquet(dataPath + \"parquet_files/train/train_static_cb_0.parquet\").pipe(set_table_dtypes)\ntrain_person_1 = pl.read_parquet(dataPath + \"parquet_files/train/train_person_1.parquet\").pipe(set_table_dtypes) \ntrain_credit_bureau_b_2 = pl.read_parquet(dataPath + \"parquet_files/train/train_credit_bureau_b_2.parquet\").pipe(set_table_dtypes) ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:21:48.822890Z","iopub.execute_input":"2024-03-18T10:21:48.823349Z","iopub.status.idle":"2024-03-18T10:22:16.174050Z","shell.execute_reply.started":"2024-03-18T10:21:48.823311Z","shell.execute_reply":"2024-03-18T10:22:16.172446Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_tax_registry_b_1 = pl.read_parquet(dataPath+ \"parquet_files/train/train_tax_registry_b_1.parquet\").pipe(set_table_dtypes)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:22:16.176897Z","iopub.execute_input":"2024-03-18T10:22:16.177390Z","iopub.status.idle":"2024-03-18T10:22:16.790894Z","shell.execute_reply.started":"2024-03-18T10:22:16.177345Z","shell.execute_reply":"2024-03-18T10:22:16.789573Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_basetable = pl.read_parquet(dataPath + \"parquet_files/test/test_base.parquet\").pipe(set_table_dtypes)\ntest_static = pl.concat(\n    [\n        pl.read_parquet(dataPath + \"parquet_files/test/test_static_0_0.parquet\").pipe(set_table_dtypes),\n        pl.read_parquet(dataPath + \"parquet_files/test/test_static_0_1.parquet\").pipe(set_table_dtypes),\n        pl.read_parquet(dataPath + \"parquet_files/test/test_static_0_2.parquet\").pipe(set_table_dtypes),\n    ],\n    how=\"vertical_relaxed\",\n)\ntest_static_cb = pl.read_parquet(dataPath + \"parquet_files/test/test_static_cb_0.parquet\").pipe(set_table_dtypes)\ntest_person_1 = pl.read_parquet(dataPath + \"parquet_files/test/test_person_1.parquet\").pipe(set_table_dtypes) \ntest_credit_bureau_b_2 = pl.read_parquet(dataPath + \"parquet_files/test/test_credit_bureau_b_2.parquet\").pipe(set_table_dtypes) ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:22:16.792508Z","iopub.execute_input":"2024-03-18T10:22:16.793040Z","iopub.status.idle":"2024-03-18T10:22:16.886044Z","shell.execute_reply.started":"2024-03-18T10:22:16.792997Z","shell.execute_reply":"2024-03-18T10:22:16.884381Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_tax_registry_b_1 = pl.read_parquet(dataPath+ \"parquet_files/test/test_tax_registry_b_1.parquet\").pipe(set_table_dtypes)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:22:16.891942Z","iopub.execute_input":"2024-03-18T10:22:16.892351Z","iopub.status.idle":"2024-03-18T10:22:16.900164Z","shell.execute_reply.started":"2024-03-18T10:22:16.892314Z","shell.execute_reply":"2024-03-18T10:22:16.898938Z"},"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    pl.col(\"childnum_185L\").sum().alias(\"total_childnum\"),\n    pl.col(\"mainoccupationinc_384A\").sum().alias(\"total_income\"),\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\", \n                                                \"num_group1\", \n                                                \"housetype_905L\",\n                                                \"empl_employedfrom_271D\",\n                                                \"empl_industry_691L\",\n                                                \"familystate_447L\",\n                                                \"sex_738L\",\n                                                \"role_1084L\",\n                                                \"language1_981M\",\n                                                \"isreference_387L\",\n                                                \"incometype_1044T\",\n                                                \"empladdr_district_926M\",\n                                               ]).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    \n    \n    if col[-1] in (\"A\", \"M\", \"P\", \"D\"):\n        selected_static_cols.append(col)\n    if col in [\"date_decision\"]:\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\", \"P\", \"D\"):\n        selected_static_cb_cols.append(col)\n    if col in [\"date_decision\"]:\n        selected_static_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\", \"for3years_584L\", \"days360_512L\", \"numberofqueries_373L\", \"riskassesment_302T\", \"riskassesment_940T\"]+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":"2024-03-18T10:22:16.901701Z","iopub.execute_input":"2024-03-18T10:22:16.902337Z","iopub.status.idle":"2024-03-18T10:22:20.726832Z","shell.execute_reply.started":"2024-03-18T10:22:16.902302Z","shell.execute_reply":"2024-03-18T10:22:20.725834Z"},"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    pl.col(\"childnum_185L\").sum().alias(\"total_childnum\"),\n    pl.col(\"mainoccupationinc_384A\").sum().alias(\"total_income\"),\n)\n\ntest_person_1_feats_2 = test_person_1.select([\"case_id\", \n                                              \"num_group1\",\n                                              \"housetype_905L\",\n                                              \"empl_employedfrom_271D\",\n                                              \"empl_industry_691L\",\n                                              \"familystate_447L\",\n                                              \"sex_738L\",\n                                              \"role_1084L\",\n                                              \"language1_981M\",\n                                              \"isreference_387L\",\n                                              \"incometype_1044T\",\n                                              \"safeguarantyflag_411L\",\n                                              \"empladdr_district_926M\",]).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\", \"for3years_584L\", \"days360_512L\", \"numberofqueries_373L\", \"riskassesment_302T\", \"riskassesment_940T\"]+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":"2024-03-18T10:22:20.728321Z","iopub.execute_input":"2024-03-18T10:22:20.728917Z","iopub.status.idle":"2024-03-18T10:22:20.749376Z","shell.execute_reply.started":"2024-03-18T10:22:20.728882Z","shell.execute_reply":"2024-03-18T10:22:20.748338Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# threshold = 0.9\n# Calculate the proportion of nulls for each column and filter\n# columns_to_keep = [col for col in data.columns if data[col].is_null().sum() / data.height < threshold]","metadata":{"execution":{"iopub.status.busy":"2024-03-17T21:18:07.311024Z","iopub.execute_input":"2024-03-17T21:18:07.311903Z","iopub.status.idle":"2024-03-17T21:18:07.321360Z","shell.execute_reply.started":"2024-03-17T21:18:07.311865Z","shell.execute_reply":"2024-03-17T21:18:07.320465Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# data = data.select(columns_to_keep)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T21:18:07.322752Z","iopub.execute_input":"2024-03-17T21:18:07.324027Z","iopub.status.idle":"2024-03-17T21:18:07.340100Z","shell.execute_reply.started":"2024-03-17T21:18:07.323982Z","shell.execute_reply":"2024-03-17T21:18:07.338847Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# columns_to_keep.remove('target')","metadata":{"execution":{"iopub.status.busy":"2024-03-17T21:18:07.341469Z","iopub.execute_input":"2024-03-17T21:18:07.342325Z","iopub.status.idle":"2024-03-17T21:18:07.355675Z","shell.execute_reply.started":"2024-03-17T21:18:07.342281Z","shell.execute_reply":"2024-03-17T21:18:07.353924Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# data_submission = data_submission.select(columns_to_keep)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T21:18:07.360625Z","iopub.execute_input":"2024-03-17T21:18:07.361083Z","iopub.status.idle":"2024-03-17T21:18:07.369348Z","shell.execute_reply.started":"2024-03-17T21:18:07.361047Z","shell.execute_reply":"2024-03-17T21:18:07.368133Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def dates_to_int(df):\n    for col in df.columns:\n        if col[-1] in (\"D\",):\n            # number_of_nulls = df.select(pl.col(col).is_null().sum())\n            df = df.with_columns(pl.col(col) - pl.col(\"date_decision\"))\n            df = df.with_columns(pl.col(col).dt.total_days())\n    df = df.drop(\"date_decision\")\n    return df","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:22:20.750879Z","iopub.execute_input":"2024-03-18T10:22:20.751483Z","iopub.status.idle":"2024-03-18T10:22:20.760879Z","shell.execute_reply.started":"2024-03-18T10:22:20.751442Z","shell.execute_reply":"2024-03-18T10:22:20.759487Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data = data.pipe(dates_to_int)\ndata = data.drop(\"dateofbirth_342D\", \"birthdate_574D\")","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:22:20.762394Z","iopub.execute_input":"2024-03-18T10:22:20.762821Z","iopub.status.idle":"2024-03-18T10:22:23.078099Z","shell.execute_reply.started":"2024-03-18T10:22:20.762787Z","shell.execute_reply":"2024-03-18T10:22:23.076821Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data_submission = data_submission.pipe(dates_to_int)\ndata_submission = data_submission.drop(\"dateofbirth_342D\", \"birthdate_574D\")","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:22:23.079749Z","iopub.execute_input":"2024-03-18T10:22:23.080138Z","iopub.status.idle":"2024-03-18T10:22:23.098132Z","shell.execute_reply.started":"2024-03-18T10:22:23.080102Z","shell.execute_reply":"2024-03-18T10:22:23.096966Z"},"trusted":true},"execution_count":null,"outputs":[]},{"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)","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:22:23.101198Z","iopub.execute_input":"2024-03-18T10:22:23.101641Z","iopub.status.idle":"2024-03-18T10:22:40.260646Z","shell.execute_reply.started":"2024-03-18T10:22:23.101603Z","shell.execute_reply":"2024-03-18T10:22:40.259418Z"},"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":"2024-03-18T10:22:40.262163Z","iopub.execute_input":"2024-03-18T10:22:40.262558Z","iopub.status.idle":"2024-03-18T10:22:40.269826Z","shell.execute_reply.started":"2024-03-18T10:22:40.262523Z","shell.execute_reply":"2024-03-18T10:22:40.268821Z"},"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":"%%time\nimport optuna\n\ndef objective(trial):\n    # Suggest values for the hyperparameters\n    param = {\n        \"objective\": \"binary\",\n        \"metric\": \"auc\",\n        \"boosting_type\": \"gbdt\",\n        \"max_depth\": trial.suggest_int(\"max_depth\", 1, 7),\n        \"num_leaves\": trial.suggest_int(\"num_leaves\", 20, 40),\n        \"learning_rate\": trial.suggest_float(\"learning_rate\", 0.01, 0.1),\n        \"feature_fraction\": trial.suggest_float(\"feature_fraction\", 0.8, 1.0),\n        \"bagging_fraction\": trial.suggest_float(\"bagging_fraction\", 0.7, 1.0),\n        \"bagging_freq\": trial.suggest_int(\"bagging_freq\", 1, 7),\n        \"n_estimators\": 1000,\n        \"verbose\": -1\n    }\n    \n    # Create datasets\n    lgb_train = lgb.Dataset(X_train, label=y_train)\n    lgb_valid = lgb.Dataset(X_valid, label=y_valid, reference=lgb_train)\n    \n    # Model training\n    gbm = lgb.train(\n        param,\n        lgb_train,\n        valid_sets=[lgb_valid],\n        verbose_eval=False,\n        early_stopping_rounds=10\n    )\n    \n    # Prediction\n    preds = gbm.predict(X_valid, num_iteration=gbm.best_iteration)\n    auc = roc_auc_score(y_valid, preds)\n    \n    return auc","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:00:36.525139Z","iopub.execute_input":"2024-03-18T10:00:36.525620Z","iopub.status.idle":"2024-03-18T10:00:38.055713Z","shell.execute_reply.started":"2024-03-18T10:00:36.525585Z","shell.execute_reply":"2024-03-18T10:00:38.054226Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# %%time\n# study = optuna.create_study(direction=\"maximize\")\n# study.optimize(objective, n_trials=30)\n\n# # print(\"Best trial:\")\n# trial = study.best_trial\n\n# print(f\"AUC: {trial.value}\")\n# print(\"Best hyperparameters:\")\n# for key, value in trial.params.items():\n#     print(f\"{key}: {value}\")","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:31:18.494537Z","iopub.execute_input":"2024-03-18T10:31:18.495117Z","iopub.status.idle":"2024-03-18T11:54:11.995247Z","shell.execute_reply.started":"2024-03-18T10:31:18.495068Z","shell.execute_reply":"2024-03-18T11:54:11.993989Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"params = {'max_depth': 5,\n          'num_leaves': 27,\n          'learning_rate': 0.04110493425829409,\n          'feature_fraction': 0.9897240634011022,\n          'bagging_fraction': 0.861387542594332,\n          'bagging_freq': 1}","metadata":{"execution":{"iopub.status.busy":"2024-03-18T10:00:48.197685Z","iopub.execute_input":"2024-03-18T10:00:48.198204Z","iopub.status.idle":"2024-03-18T10:00:48.206745Z","shell.execute_reply.started":"2024-03-18T10:00:48.198161Z","shell.execute_reply":"2024-03-18T10:00:48.204458Z"},"trusted":true},"execution_count":null,"outputs":[]},{"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\": 26,\n    \"learning_rate\": 0.03427736556910803,\n    \"feature_fraction\": 0.896099343234861,\n    \"bagging_fraction\": 0.9323664168499988,\n    \"bagging_freq\": 2,\n    \"n_estimators\": 1500,\n    \"verbose\": -1,\n}\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":"2024-03-18T11:57:23.907160Z","iopub.execute_input":"2024-03-18T11:57:23.907777Z","iopub.status.idle":"2024-03-18T12:01:57.209972Z","shell.execute_reply.started":"2024-03-18T11:57:23.907732Z","shell.execute_reply":"2024-03-18T12:01:57.208283Z"},"trusted":true},"execution_count":null,"outputs":[]},{"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\n# print(f'The AUC score on the train set is: {roc_auc_score(base_train[\"target\"], base_train[\"score\"])}') \n# print(f'The AUC score on the valid set is: {roc_auc_score(base_valid[\"target\"], base_valid[\"score\"])}') \n# print(f'The AUC score on the test set is: {roc_auc_score(base_test[\"target\"], base_test[\"score\"])}')  ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T12:01:57.212796Z","iopub.execute_input":"2024-03-18T12:01:57.213386Z","iopub.status.idle":"2024-03-18T12:03:06.151814Z","shell.execute_reply.started":"2024-03-18T12:01:57.213338Z","shell.execute_reply":"2024-03-18T12:03:06.150101Z"},"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\n# stability_score_train = gini_stability(base_train)\n# stability_score_valid = gini_stability(base_valid)\n# stability_score_test = gini_stability(base_test)\n\n# print(f'The stability score on the train set is: {stability_score_train}') \n# print(f'The stability score on the valid set is: {stability_score_valid}') \n# print(f'The stability score on the test set is: {stability_score_test}')  ","metadata":{"execution":{"iopub.status.busy":"2024-03-18T12:03:06.156528Z","iopub.execute_input":"2024-03-18T12:03:06.156979Z","iopub.status.idle":"2024-03-18T12:03:09.629109Z","shell.execute_reply.started":"2024-03-18T12:03:06.156942Z","shell.execute_reply":"2024-03-18T12:03:09.627795Z"},"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 = 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":"2024-03-18T12:03:43.728526Z","iopub.execute_input":"2024-03-18T12:03:43.729058Z","iopub.status.idle":"2024-03-18T12:03:43.946146Z","shell.execute_reply.started":"2024-03-18T12:03:43.729017Z","shell.execute_reply":"2024-03-18T12:03:43.944773Z"},"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":"2024-03-18T12:03:44.160824Z","iopub.execute_input":"2024-03-18T12:03:44.161340Z","iopub.status.idle":"2024-03-18T12:03:44.172985Z","shell.execute_reply.started":"2024-03-18T12:03:44.161298Z","shell.execute_reply":"2024-03-18T12:03:44.171615Z"},"trusted":true},"execution_count":null,"outputs":[]}]}