{"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":7602123,"sourceType":"competition"}],"dockerImageVersionId":30646,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# DuckDB for Feature engineering - Home Credit 2024","metadata":{}},{"cell_type":"markdown","source":"DuckDB (https://duckdb.org/)\nWhy DuckDB: https://duckdb.org/why_duckdb\n\nDuckDB is an open-source tool designed for efficient querying data and processing. It's suitable for analytical query workloads, especially in this competition.\n\n**SQL Syntax is more intuitive and easy to write complex queries.**","metadata":{"execution":{"iopub.status.busy":"2024-02-17T11:39:34.409183Z","iopub.execute_input":"2024-02-17T11:39:34.409588Z","iopub.status.idle":"2024-02-17T11:39:34.457523Z","shell.execute_reply.started":"2024-02-17T11:39:34.409553Z","shell.execute_reply":"2024-02-17T11:39:34.455871Z"}}},{"cell_type":"markdown","source":"I rewrote this notebook based on the Starter Notebook: https://www.kaggle.com/code/jetakow/home-credit-2024-starter-notebook","metadata":{}},{"cell_type":"code","source":"!pip install -q duckdb","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:09.604022Z","iopub.execute_input":"2024-02-17T12:09:09.604732Z","iopub.status.idle":"2024-02-17T12:09:24.699500Z","shell.execute_reply.started":"2024-02-17T12:09:09.604695Z","shell.execute_reply":"2024-02-17T12:09:24.698175Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import os\nimport gc\nimport 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\ndata_path = \"/kaggle/input/home-credit-credit-risk-model-stability/\"","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:24.702048Z","iopub.execute_input":"2024-02-17T12:09:24.702463Z","iopub.status.idle":"2024-02-17T12:09:26.863014Z","shell.execute_reply.started":"2024-02-17T12:09:24.702427Z","shell.execute_reply":"2024-02-17T12:09:26.861712Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import duckdb\nduckdb.__version__","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:26.864367Z","iopub.execute_input":"2024-02-17T12:09:26.864724Z","iopub.status.idle":"2024-02-17T12:09:27.107412Z","shell.execute_reply.started":"2024-02-17T12:09:26.864694Z","shell.execute_reply":"2024-02-17T12:09:27.106021Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# create in memory connection\nconn = duckdb.connect()","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:27.109729Z","iopub.execute_input":"2024-02-17T12:09:27.110089Z","iopub.status.idle":"2024-02-17T12:09:27.122797Z","shell.execute_reply.started":"2024-02-17T12:09:27.110059Z","shell.execute_reply":"2024-02-17T12:09:27.121539Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## 1. Loading Data","metadata":{}},{"cell_type":"markdown","source":"### 1.1 Train","metadata":{}},{"cell_type":"code","source":"path_train_basetable = os.path.join(data_path, \"parquet_files/train/train_base.parquet\")\npath_train_static = lambda idx: os.path.join(\n    data_path, f\"parquet_files/train/train_static_0_{idx}.parquet\")\npath_train_static_cb = os.path.join(\n    data_path, f\"parquet_files/train/train_static_cb_0.parquet\")\npath_train_person_1 = os.path.join(\n    data_path, f\"parquet_files/train/train_person_1.parquet\")\npath_train_credit_bureau_b_2 = os.path.join(\n    data_path, f\"parquet_files/train/train_credit_bureau_b_2.parquet\")\n\nconn.sql(f\"\"\"\n\n-- Create train_basetable\nCREATE OR REPLACE TABLE train_basetable AS (\n   SELECT * FROM '{path_train_basetable}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE train_static AS (\n    SELECT *\n    FROM '{path_train_static(0)}'\n    UNION ALL\n    SELECT *\n    FROM '{path_train_static(1)}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE train_static_cb AS (\n    SELECT *\n    FROM '{path_train_static_cb}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE train_person_1 AS (\n    SELECT *\n    FROM '{path_train_person_1}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE train_credit_bureau_b_2 AS (\n    SELECT *\n    FROM '{path_train_credit_bureau_b_2}'\n);\n\"\"\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:27.124498Z","iopub.execute_input":"2024-02-17T12:09:27.125255Z","iopub.status.idle":"2024-02-17T12:09:48.177185Z","shell.execute_reply.started":"2024-02-17T12:09:27.125218Z","shell.execute_reply":"2024-02-17T12:09:48.176181Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"We can easily convert to pandas dataframe by `conn.sql` then `.df()`","metadata":{}},{"cell_type":"code","source":"conn.sql(\"SELECT * FROM train_basetable LIMIT 1\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.178673Z","iopub.execute_input":"2024-02-17T12:09:48.179741Z","iopub.status.idle":"2024-02-17T12:09:48.195536Z","shell.execute_reply.started":"2024-02-17T12:09:48.179697Z","shell.execute_reply":"2024-02-17T12:09:48.194356Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"example_convert_df = conn.sql(\"SELECT * FROM train_basetable LIMIT 1\").df()\nprint(type(example_convert_df))\nexample_convert_df","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.196863Z","iopub.execute_input":"2024-02-17T12:09:48.197340Z","iopub.status.idle":"2024-02-17T12:09:48.215181Z","shell.execute_reply.started":"2024-02-17T12:09:48.197307Z","shell.execute_reply":"2024-02-17T12:09:48.213809Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Explore data :D\n# conn.sql(\"SELECT * FROM train_static LIMIT 1\")\n# conn.sql(f\"SELECT * FROM train_static_cb LIMIT 1\")\n# conn.sql(f\"SELECT * FROM train_person_1 LIMIT 1\")\n# conn.sql(f\"SELECT * FROM train_credit_bureau_b_2 LIMIT 1\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.217116Z","iopub.execute_input":"2024-02-17T12:09:48.217734Z","iopub.status.idle":"2024-02-17T12:09:48.223405Z","shell.execute_reply.started":"2024-02-17T12:09:48.217689Z","shell.execute_reply":"2024-02-17T12:09:48.222257Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### 1.2 Test","metadata":{}},{"cell_type":"code","source":"path_test_basetable = os.path.join(data_path, \"parquet_files/test/test_base.parquet\")\npath_test_static = lambda idx: os.path.join(\n    data_path, f\"parquet_files/test/test_static_0_{idx}.parquet\")\npath_test_static_cb = os.path.join(\n    data_path, f\"parquet_files/test/test_static_cb_0.parquet\")\npath_test_person_1 = os.path.join(\n    data_path, f\"parquet_files/test/test_person_1.parquet\")\npath_test_credit_bureau_b_2 = os.path.join(\n    data_path, f\"parquet_files/test/test_credit_bureau_b_2.parquet\")\n\nconn.sql(f\"\"\"\n\n-- Create test_basetable\nCREATE OR REPLACE TABLE test_basetable AS (\n   SELECT * FROM '{path_test_basetable}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE test_static AS (\n    SELECT *\n    FROM '{path_test_static(0)}'\n    UNION ALL\n    SELECT *\n    FROM '{path_test_static(1)}'\n    UNION ALL\n    SELECT *\n    FROM '{path_test_static(2)}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE test_static_cb AS (\n    SELECT *\n    FROM '{path_test_static_cb}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE test_person_1 AS (\n    SELECT *\n    FROM '{path_test_person_1}'\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE test_credit_bureau_b_2 AS (\n    SELECT *\n    FROM '{path_test_credit_bureau_b_2}'\n);\n\"\"\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.225039Z","iopub.execute_input":"2024-02-17T12:09:48.225789Z","iopub.status.idle":"2024-02-17T12:09:48.378520Z","shell.execute_reply.started":"2024-02-17T12:09:48.225750Z","shell.execute_reply":"2024-02-17T12:09:48.377313Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## 2. Feature Engineering","metadata":{}},{"cell_type":"markdown","source":"### 2.1 Train:","metadata":{}},{"cell_type":"code","source":"conn.sql(\"\"\"\n\nCREATE OR REPLACE TABLE train_person_1_feats_1 AS (\n    SELECT \n        case_id,\n        MAX(mainoccupationinc_384A) mainoccupationinc_384A_max,\n        MAX(\n            CASE WHEN incometype_1044T='SELFEMPLOYED' THEN 1 ELSE 0 END\n        ) mainoccupationinc_384A_any_selfemployed \n    FROM train_person_1\n    GROUP BY case_id\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE train_person_1_feats_2 AS (\n    SELECT \n        case_id, housetype_905L person_housetype\n    FROM train_person_1\n    WHERE num_group1 = 0\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE train_credit_bureau_b_2_feats AS (\n    SELECT\n        case_id,\n        MAX(pmts_pmtsoverdue_635A) AS pmts_pmtsoverdue_635A_max,\n        MAX(\n            CASE WHEN pmts_dpdvalue_108P > 31 THEN 1 ELSE 0 END\n        ) pmts_dpdvalue_108P_over31\n    FROM train_credit_bureau_b_2\n    GROUP BY case_id\n)\n\n\"\"\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.382931Z","iopub.execute_input":"2024-02-17T12:09:48.383339Z","iopub.status.idle":"2024-02-17T12:09:48.775577Z","shell.execute_reply.started":"2024-02-17T12:09:48.383306Z","shell.execute_reply":"2024-02-17T12:09:48.774441Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"conn.sql(\"\"\"\nSELECT *\nFROM train_credit_bureau_b_2_feats\nLIMIT 10\n\"\"\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.777101Z","iopub.execute_input":"2024-02-17T12:09:48.777525Z","iopub.status.idle":"2024-02-17T12:09:48.793140Z","shell.execute_reply.started":"2024-02-17T12:09:48.777491Z","shell.execute_reply":"2024-02-17T12:09:48.791622Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"selected_static_cols = []\nfor col in conn.sql(\"DESCRIBE train_static\").df().column_name.tolist():\n    if col[-1] in (\"A\", \"M\"):\n        selected_static_cols.append(col)\nselected_static_cols = ', '.join(selected_static_cols)\nprint(selected_static_cols)","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.795181Z","iopub.execute_input":"2024-02-17T12:09:48.795687Z","iopub.status.idle":"2024-02-17T12:09:48.808141Z","shell.execute_reply.started":"2024-02-17T12:09:48.795650Z","shell.execute_reply":"2024-02-17T12:09:48.806370Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"selected_static_cb_cols = []\nfor col in conn.sql(\"DESCRIBE train_static_cb\").df().column_name.tolist():\n    if col[-1] in (\"A\", \"M\"):\n        selected_static_cb_cols.append(col)\nselected_static_cb_cols = ', '.join(selected_static_cb_cols)\nprint(selected_static_cb_cols)","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.809172Z","iopub.execute_input":"2024-02-17T12:09:48.809521Z","iopub.status.idle":"2024-02-17T12:09:48.819642Z","shell.execute_reply.started":"2024-02-17T12:09:48.809493Z","shell.execute_reply":"2024-02-17T12:09:48.818316Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### Join Features","metadata":{}},{"cell_type":"code","source":"conn.sql(f\"\"\"\n\nCREATE OR REPLACE TABLE data AS (\n\nWITH selected_train_static AS (\n    SELECT case_id, {selected_static_cols}\n    FROM train_static\n),\nselected_train_static_cb AS (\n    SELECT case_id, {selected_static_cb_cols}\n    FROM train_static_cb\n)\n\nSELECT *\nFROM train_basetable t1\n    LEFT JOIN selected_train_static t2 ON t1.case_id=t2.case_id\n    LEFT JOIN selected_train_static_cb t3 ON t1.case_id=t3.case_id\n    LEFT JOIN train_person_1_feats_1 t4 ON t1.case_id=t4.case_id\n    LEFT JOIN train_person_1_feats_2 t5 ON t1.case_id=t5.case_id\n    LEFT JOIN train_credit_bureau_b_2_feats t6 ON t1.case_id=t6.case_id\n\n)\n\"\"\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:48.821744Z","iopub.execute_input":"2024-02-17T12:09:48.822122Z","iopub.status.idle":"2024-02-17T12:09:52.375422Z","shell.execute_reply.started":"2024-02-17T12:09:48.822091Z","shell.execute_reply":"2024-02-17T12:09:52.373955Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### 2.2 Test:","metadata":{}},{"cell_type":"code","source":"conn.sql(\"\"\"\n\nCREATE OR REPLACE TABLE test_person_1_feats_1 AS (\n    SELECT \n        case_id,\n        MAX(mainoccupationinc_384A) mainoccupationinc_384A_max,\n        MAX(\n            CASE WHEN incometype_1044T='SELFEMPLOYED' THEN 1 ELSE 0 END\n        ) mainoccupationinc_384A_any_selfemployed \n    FROM test_person_1\n    GROUP BY case_id\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE test_person_1_feats_2 AS (\n    SELECT \n        case_id, housetype_905L person_housetype\n    FROM test_person_1\n    WHERE num_group1 = 0\n);\n\n-----------------------------------------------\n\nCREATE OR REPLACE TABLE test_credit_bureau_b_2_feats AS (\n    SELECT\n        case_id,\n        MAX(pmts_pmtsoverdue_635A) AS pmts_pmtsoverdue_635A_max,\n        MAX(\n            CASE WHEN pmts_dpdvalue_108P > 31 THEN 1 ELSE 0 END\n        ) pmts_dpdvalue_108P_over31\n    FROM test_credit_bureau_b_2\n    GROUP BY case_id\n)\n\n\"\"\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:52.377046Z","iopub.execute_input":"2024-02-17T12:09:52.377416Z","iopub.status.idle":"2024-02-17T12:09:52.388614Z","shell.execute_reply.started":"2024-02-17T12:09:52.377386Z","shell.execute_reply":"2024-02-17T12:09:52.387341Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"#### Join Features","metadata":{}},{"cell_type":"code","source":"conn.sql(f\"\"\"\n\nCREATE OR REPLACE TABLE data_submission AS (\n\nWITH selected_test_static AS (\n    SELECT case_id, {selected_static_cols}\n    FROM test_static\n),\nselected_test_static_cb AS (\n    SELECT case_id, {selected_static_cb_cols}\n    FROM test_static_cb\n)\n\nSELECT *\nFROM test_basetable t1\n    LEFT JOIN selected_test_static t2 ON t1.case_id=t2.case_id\n    LEFT JOIN selected_test_static_cb t3 ON t1.case_id=t3.case_id\n    LEFT JOIN test_person_1_feats_1 t4 ON t1.case_id=t4.case_id\n    LEFT JOIN test_person_1_feats_2 t5 ON t1.case_id=t5.case_id\n    LEFT JOIN test_credit_bureau_b_2_feats t6 ON t1.case_id=t6.case_id\n\n)\n\"\"\")","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:52.389746Z","iopub.execute_input":"2024-02-17T12:09:52.390091Z","iopub.status.idle":"2024-02-17T12:09:52.419405Z","shell.execute_reply.started":"2024-02-17T12:09:52.390061Z","shell.execute_reply":"2024-02-17T12:09:52.418265Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## 3.Train model & Create Submission","metadata":{}},{"cell_type":"code","source":"np.random.seed(0)\ncase_ids = conn.sql(\"SELECT case_id FROM data\").df().case_id.unique()\nnp.random.shuffle(case_ids)\n\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)","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:52.420847Z","iopub.execute_input":"2024-02-17T12:09:52.421188Z","iopub.status.idle":"2024-02-17T12:09:52.694867Z","shell.execute_reply.started":"2024-02-17T12:09:52.421160Z","shell.execute_reply":"2024-02-17T12:09:52.693816Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cols_pred = []\nfor col in conn.sql(\"DESCRIBE data\").df().column_name.tolist():\n    if col[-1].isupper() and col[:-1].islower():\n        cols_pred.append(col)","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:52.698816Z","iopub.execute_input":"2024-02-17T12:09:52.699200Z","iopub.status.idle":"2024-02-17T12:09:52.709363Z","shell.execute_reply.started":"2024-02-17T12:09:52.699168Z","shell.execute_reply":"2024-02-17T12:09:52.706795Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(cols_pred)","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:52.711539Z","iopub.execute_input":"2024-02-17T12:09:52.711908Z","iopub.status.idle":"2024-02-17T12:09:52.720195Z","shell.execute_reply.started":"2024-02-17T12:09:52.711876Z","shell.execute_reply":"2024-02-17T12:09:52.718072Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def from_duckdb_to_pandas(case_ids) -> pl.DataFrame:\n    dataframe = conn.sql(f\"\"\"\n    SELECT \"case_id\", \"WEEK_NUM\", \"target\", {', '.join(cols_pred)}\n    FROM data\n    WHERE case_id IN {str(tuple(case_ids.tolist()))}\n    \"\"\").df()\n    base = dataframe[[\"case_id\", \"WEEK_NUM\", \"target\"]].copy()\n    x = dataframe[cols_pred].copy()\n    y = dataframe[[\"target\"]].copy()\n    del dataframe\n    return base, x, y","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:52.721696Z","iopub.execute_input":"2024-02-17T12:09:52.722018Z","iopub.status.idle":"2024-02-17T12:09:52.734155Z","shell.execute_reply.started":"2024-02-17T12:09:52.721991Z","shell.execute_reply":"2024-02-17T12:09:52.732697Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"base_train, X_train, y_train = from_duckdb_to_pandas(case_ids_train)\nbase_valid, X_valid, y_valid = from_duckdb_to_pandas(case_ids_valid)\nbase_test, X_test, y_test = from_duckdb_to_pandas(case_ids_test)","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:09:52.735947Z","iopub.execute_input":"2024-02-17T12:09:52.736824Z","iopub.status.idle":"2024-02-17T12:10:16.288470Z","shell.execute_reply.started":"2024-02-17T12:09:52.736779Z","shell.execute_reply":"2024-02-17T12:10:16.287201Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"gc.collect()","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:10:16.290082Z","iopub.execute_input":"2024-02-17T12:10:16.290831Z","iopub.status.idle":"2024-02-17T12:10:16.422073Z","shell.execute_reply.started":"2024-02-17T12:10:16.290787Z","shell.execute_reply":"2024-02-17T12:10:16.421189Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def 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":"2024-02-17T12:10:16.423521Z","iopub.execute_input":"2024-02-17T12:10:16.423857Z","iopub.status.idle":"2024-02-17T12:10:16.437320Z","shell.execute_reply.started":"2024-02-17T12:10:16.423828Z","shell.execute_reply":"2024-02-17T12:10:16.435935Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"for df in [X_train, X_valid, X_test]:\n    df = convert_strings(df)","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:10:16.438963Z","iopub.execute_input":"2024-02-17T12:10:16.439662Z","iopub.status.idle":"2024-02-17T12:10:23.274551Z","shell.execute_reply.started":"2024-02-17T12:10:16.439625Z","shell.execute_reply":"2024-02-17T12:10:23.273204Z"},"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":"2024-02-17T12:10:23.276163Z","iopub.execute_input":"2024-02-17T12:10:23.276592Z","iopub.status.idle":"2024-02-17T12:10:23.282841Z","shell.execute_reply.started":"2024-02-17T12:10:23.276560Z","shell.execute_reply":"2024-02-17T12:10:23.281481Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Training LightGBM\n\nJust an example ;D","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\": 3,\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\": 1000,\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":"2024-02-17T12:10:23.284700Z","iopub.execute_input":"2024-02-17T12:10:23.285213Z","iopub.status.idle":"2024-02-17T12:11:57.826918Z","shell.execute_reply.started":"2024-02-17T12:10:23.285166Z","shell.execute_reply":"2024-02-17T12:11:57.825607Z"},"trusted":true},"execution_count":null,"outputs":[]},{"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":"2024-02-17T12:11:57.828286Z","iopub.execute_input":"2024-02-17T12:11:57.828644Z","iopub.status.idle":"2024-02-17T12:12:21.855525Z","shell.execute_reply.started":"2024-02-17T12:11:57.828615Z","shell.execute_reply":"2024-02-17T12:12:21.854070Z"},"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":"2024-02-17T12:12:21.861913Z","iopub.execute_input":"2024-02-17T12:12:21.862347Z","iopub.status.idle":"2024-02-17T12:12:23.031253Z","shell.execute_reply.started":"2024-02-17T12:12:21.862313Z","shell.execute_reply":"2024-02-17T12:12:23.029927Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Submission\n","metadata":{}},{"cell_type":"code","source":"data_submission = conn.sql(f\"\"\"\n    SELECT \"case_id\", {', '.join(cols_pred)}\n    FROM data_submission\n\"\"\").df()","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:12:23.033096Z","iopub.execute_input":"2024-02-17T12:12:23.033644Z","iopub.status.idle":"2024-02-17T12:12:23.050388Z","shell.execute_reply.started":"2024-02-17T12:12:23.033598Z","shell.execute_reply":"2024-02-17T12:12:23.049221Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X_submission = data_submission[cols_pred].copy()\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":"2024-02-17T12:12:23.051785Z","iopub.execute_input":"2024-02-17T12:12:23.052142Z","iopub.status.idle":"2024-02-17T12:12:23.164214Z","shell.execute_reply.started":"2024-02-17T12:12:23.052113Z","shell.execute_reply":"2024-02-17T12:12:23.162954Z"},"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\")\nsubmission.head()","metadata":{"execution":{"iopub.status.busy":"2024-02-17T12:12:23.166095Z","iopub.execute_input":"2024-02-17T12:12:23.166895Z","iopub.status.idle":"2024-02-17T12:12:23.182723Z","shell.execute_reply.started":"2024-02-17T12:12:23.166860Z","shell.execute_reply":"2024-02-17T12:12:23.181016Z"},"trusted":true},"execution_count":null,"outputs":[]}]}