{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":50160,"databundleVersionId":7921029,"sourceType":"competition"}],"dockerImageVersionId":30648,"isInternetEnabled":false,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"i did the following experiments and changes. unfortunately i did not have enough time to finish all of the ideas i had but here are the ideas i've tried in the following notebook. \n\n1. Added a Scaler.\n2. Added Log Normalization.\n3. I've tried PCA but it didn't seem to help\n4. I've found the best hyperparameters for the model using Optuna. \n5. I've changed the Voting Model to be of a Logistic Regression Nature rather than just a mean of the predictions.\n6. I've added aggrgations such as mean, max, std -- they did not improve the score so i had to erase them bc i was running out of memory\n7. I've tried using quantiasation but it's also not in the last submission bc of the memory issues (on the scoring bc of the large test data the notebook was consuming more than 30GB)","metadata":{}},{"cell_type":"code","source":"import os\nimport gc\nfrom glob import glob\nfrom pathlib import Path\nfrom datetime import datetime\n\nimport numpy as np\nimport pandas as pd\nimport polars as pl\n\nimport matplotlib.pyplot as plt\nimport seaborn as sns\n\nfrom sklearn.model_selection import StratifiedGroupKFold\nfrom sklearn.base import BaseEstimator, ClassifierMixin\n\nimport lightgbm as lgb\n\nimport warnings\nwarnings.simplefilter(action='ignore', category=FutureWarning)","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2024-03-17T19:13:54.658804Z","iopub.execute_input":"2024-03-17T19:13:54.659351Z","iopub.status.idle":"2024-03-17T19:13:54.668682Z","shell.execute_reply.started":"2024-03-17T19:13:54.659312Z","shell.execute_reply":"2024-03-17T19:13:54.667059Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Pre-Fitted Voting Model","metadata":{}},{"cell_type":"code","source":"class VotingModel(BaseEstimator, ClassifierMixin):\n    def __init__(self, estimators):\n        super().__init__()\n        self.estimators = estimators\n        \n    def fit(self, X, y=None):\n        return self\n    \n    def predict(self, X):\n        y_preds = [estimator.predict(X) for estimator in self.estimators]\n        return np.mean(y_preds, axis=0)\n    \n    def predict_proba(self, X):\n        y_preds = [estimator.predict_proba(X) for estimator in self.estimators]\n        return np.mean(y_preds, axis=0)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.670696Z","iopub.execute_input":"2024-03-17T19:13:54.671023Z","iopub.status.idle":"2024-03-17T19:13:54.683787Z","shell.execute_reply.started":"2024-03-17T19:13:54.670997Z","shell.execute_reply":"2024-03-17T19:13:54.682832Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Pipeline","metadata":{}},{"cell_type":"code","source":"class Pipeline:\n    @staticmethod\n    def set_table_dtypes(df):\n        for col in df.columns:\n            if col in [\"case_id\", \"WEEK_NUM\", \"num_group1\", \"num_group2\"]:\n                df = df.with_columns(pl.col(col).cast(pl.Int32))\n            elif col in [\"date_decision\"]:\n                df = df.with_columns(pl.col(col).cast(pl.Date))\n            elif col[-1] in (\"P\", \"A\"):\n                df = df.with_columns(pl.col(col).cast(pl.Float64))\n            elif col[-1] in (\"M\",):\n                df = df.with_columns(pl.col(col).cast(pl.String))\n            elif col[-1] in (\"D\",):\n                df = df.with_columns(pl.col(col).cast(pl.Date))            \n\n        return df\n    \n    @staticmethod\n    def handle_dates(df):\n        for col in df.columns:\n            if col[-1] in (\"D\",):\n                df = df.with_columns(pl.col(col) - pl.col(\"date_decision\"))\n                df = df.with_columns(pl.col(col).dt.total_days())\n                df = df.with_columns(pl.col(col).cast(pl.Float32))\n                \n        df = df.drop(\"date_decision\", \"MONTH\")\n\n        return df\n    \n    @staticmethod\n    def filter_cols(df):\n        for col in df.columns:\n            if col not in [\"target\", \"case_id\", \"WEEK_NUM\"]:\n                isnull = df[col].is_null().mean()\n\n                if isnull > 0.95:\n                    df = df.drop(col)\n\n        for col in df.columns:\n            if (col not in [\"target\", \"case_id\", \"WEEK_NUM\"]) & (df[col].dtype == pl.String):\n                freq = df[col].n_unique()\n\n                if (freq == 1) | (freq > 10):\n                    df = df.drop(col)\n\n        return df","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.685639Z","iopub.execute_input":"2024-03-17T19:13:54.686002Z","iopub.status.idle":"2024-03-17T19:13:54.702551Z","shell.execute_reply.started":"2024-03-17T19:13:54.685972Z","shell.execute_reply":"2024-03-17T19:13:54.701668Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Automatic Aggregation","metadata":{}},{"cell_type":"code","source":"class Aggregator:\n    @staticmethod\n    def num_expr_max(df):\n        cols = [col for col in df.columns if col[-1] in (\"P\", \"A\")]\n\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n\n        return expr_max\n    \n    def num_expr_min(df):\n        cols = [col for col in df.columns if col[-1] in (\"P\", \"A\")]\n\n        expr = [pl.min(col).alias(f\"min_{col}\") for col in cols]\n\n        return expr\n    \n    def num_expr_mean(df):\n        cols = [col for col in df.columns if col[-1] in (\"P\", \"A\")]\n\n        expr = [pl.mean(col).alias(f\"mean_{col}\") for col in cols]\n\n        return expr\n    \n    def num_expr_std(df):\n        cols = [col for col in df.columns if col[-1] in (\"P\", \"A\")]\n\n        expr = [pl.std(col).alias(f\"std_{col}\") for col in cols]\n\n        return expr\n\n    @staticmethod\n    def date_expr(df):\n        cols = [col for col in df.columns if col[-1] in (\"D\",)]\n\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n\n        return expr_max\n\n    @staticmethod\n    def str_expr(df):\n        cols = [col for col in df.columns if col[-1] in (\"M\",)]\n        \n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n\n        return expr_max\n\n    @staticmethod\n    def other_expr(df):\n        cols = [col for col in df.columns if col[-1] in (\"T\", \"L\")]\n        \n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n\n        return expr_max\n    \n    @staticmethod\n    def count_expr(df):\n        cols = [col for col in df.columns if \"num_group\" in col]\n\n        expr_max = [pl.max(col).alias(f\"max_{col}\") for col in cols]\n\n        return expr_max\n\n    @staticmethod\n    def get_exprs(df):\n        exprs = Aggregator.num_expr_max(df) + \\\n                Aggregator.num_expr_mean(df) + \\\n                Aggregator.date_expr(df) + \\\n                Aggregator.str_expr(df) + \\\n                Aggregator.other_expr(df) + \\\n                Aggregator.count_expr(df)\n\n        return exprs","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.704853Z","iopub.execute_input":"2024-03-17T19:13:54.705524Z","iopub.status.idle":"2024-03-17T19:13:54.721186Z","shell.execute_reply.started":"2024-03-17T19:13:54.705483Z","shell.execute_reply":"2024-03-17T19:13:54.720047Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Add feature of the quantisation of the columns ","metadata":{}},{"cell_type":"markdown","source":"### File I/O","metadata":{}},{"cell_type":"code","source":"def quantize_column(col):\n    q1 = col.quantile(0.33)\n    q2 = col.quantile(0.67)\n    return col.apply(lambda x: 0 if x <= q1 else 1 if x <= q2 else 2)\n\ndef normalize_column(column):    \n    return (column - column.min()) / (column.max() - column.min())\n\ndef read_file(path, depth=None, ids = [0]):\n    df = pl.read_parquet(path)\n    df = df.pipe(Pipeline.set_table_dtypes)\n    \n    if depth in [1, 2]:\n        df = df.group_by(\"case_id\").agg(Aggregator.get_exprs(df))\n                \n    for col_name in df.columns:\n        if df[col_name].dtype in [pl.Int64, pl.Float64] and col_name != \"target\":\n            try:\n                df = df.with_columns(normalize_column(df[col_name]).alias(col_name))\n                #df = df.with_columns(quantize_column(df[col_name]).alias(col_name + '_quantized'))\n            except:\n                pass\n    \n    return df\n\ndef read_files(regex_path, depth = None, ids = [0]):\n    chunks = []\n    for path in glob(str(regex_path)):\n        df = pl.read_parquet(path)\n        df = df.pipe(Pipeline.set_table_dtypes)\n        \n        if depth in [1, 2]:\n            df = df.group_by(\"case_id\").agg(Aggregator.get_exprs(df))\n        \n        chunks.append(df)\n        \n    df = pl.concat(chunks, how=\"vertical_relaxed\")\n    df = df.unique(subset=[\"case_id\"])\n    \n    for col_name in df.columns:\n        if df[col_name].dtype in [pl.Int64, pl.Float64] and col_name != \"target\":\n            try:\n                df = df.with_columns(normalize_column(df[col_name]).alias(col_name))\n                #df = df.with_columns(quantize_column(df[col_name]).alias(col_name + '_quantized'))\n            except:\n                pass\n    \n    return df","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.722858Z","iopub.execute_input":"2024-03-17T19:13:54.723161Z","iopub.status.idle":"2024-03-17T19:13:54.741256Z","shell.execute_reply.started":"2024-03-17T19:13:54.723136Z","shell.execute_reply":"2024-03-17T19:13:54.740485Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Feature Engineering","metadata":{}},{"cell_type":"code","source":"def feature_eng(df_base, depth_0, depth_1, depth_2):\n    df_base = (\n        df_base\n        .with_columns(\n            month_decision = pl.col(\"date_decision\").dt.month(),\n            weekday_decision = pl.col(\"date_decision\").dt.weekday(),\n        )\n    )\n    \n    for i, df in enumerate(depth_0 + depth_1 + depth_2):\n        df_base = df_base.join(df, how=\"left\", on=\"case_id\", suffix=f\"_{i}\")\n        \n    df_base = df_base.pipe(Pipeline.handle_dates)\n    \n    return df_base","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.742862Z","iopub.execute_input":"2024-03-17T19:13:54.743559Z","iopub.status.idle":"2024-03-17T19:13:54.760121Z","shell.execute_reply.started":"2024-03-17T19:13:54.743527Z","shell.execute_reply":"2024-03-17T19:13:54.758614Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def to_pandas(df_data, cat_cols=None):\n    df_data = df_data.to_pandas()\n    \n    if cat_cols is None:\n        cat_cols = list(df_data.select_dtypes(\"object\").columns)\n    \n    df_data[cat_cols] = df_data[cat_cols].astype(\"category\")\n    \n    return df_data, cat_cols","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.763673Z","iopub.execute_input":"2024-03-17T19:13:54.764012Z","iopub.status.idle":"2024-03-17T19:13:54.775024Z","shell.execute_reply.started":"2024-03-17T19:13:54.763985Z","shell.execute_reply":"2024-03-17T19:13:54.774123Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Configuration","metadata":{}},{"cell_type":"code","source":"ROOT            = Path(\"/kaggle/input/home-credit-credit-risk-model-stability\")\nTRAIN_DIR       = ROOT / \"parquet_files\" / \"train\"\nTEST_DIR        = ROOT / \"parquet_files\" / \"test\"","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.776629Z","iopub.execute_input":"2024-03-17T19:13:54.777166Z","iopub.status.idle":"2024-03-17T19:13:54.788550Z","shell.execute_reply.started":"2024-03-17T19:13:54.777137Z","shell.execute_reply":"2024-03-17T19:13:54.787518Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Train Files Read & Feature Engineering","metadata":{}},{"cell_type":"code","source":"import random\n\nunique_ids = random.sample(range(1526659), 300000)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:54.790997Z","iopub.execute_input":"2024-03-17T19:13:54.791538Z","iopub.status.idle":"2024-03-17T19:13:55.160884Z","shell.execute_reply.started":"2024-03-17T19:13:54.791508Z","shell.execute_reply":"2024-03-17T19:13:55.159670Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"data_store = {\n    \"df_base\": read_file(TRAIN_DIR / \"train_base.parquet\", unique_ids),\n    \"depth_0\": [\n        read_file(TRAIN_DIR / \"train_static_cb_0.parquet\", unique_ids),\n        read_files(TRAIN_DIR / \"train_static_0_*.parquet\", unique_ids),\n    ],\n    \"depth_1\": [\n        read_files(TRAIN_DIR / \"train_applprev_1_*.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_tax_registry_a_1.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_tax_registry_b_1.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_tax_registry_c_1.parquet\", 1, unique_ids),\n        read_files(TRAIN_DIR / \"train_credit_bureau_a_1_*.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_credit_bureau_b_1.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_other_1.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_person_1.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_deposit_1.parquet\", 1, unique_ids),\n        read_file(TRAIN_DIR / \"train_debitcard_1.parquet\", 1, unique_ids),\n    ],\n    \"depth_2\": [\n        read_file(TRAIN_DIR / \"train_credit_bureau_b_2.parquet\", 2, unique_ids),\n        read_files(TRAIN_DIR / \"train_credit_bureau_a_2_*.parquet\", 2, unique_ids),\n    ]\n}","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:13:55.163046Z","iopub.execute_input":"2024-03-17T19:13:55.163565Z","iopub.status.idle":"2024-03-17T19:16:50.299358Z","shell.execute_reply.started":"2024-03-17T19:13:55.163520Z","shell.execute_reply":"2024-03-17T19:16:50.298186Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train = feature_eng(**data_store)\n\nprint(\"train data shape:\\t\", df_train.shape)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:16:50.301171Z","iopub.execute_input":"2024-03-17T19:16:50.301590Z","iopub.status.idle":"2024-03-17T19:17:07.592578Z","shell.execute_reply.started":"2024-03-17T19:16:50.301558Z","shell.execute_reply":"2024-03-17T19:17:07.591410Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train = df_train.sample(n=len(df_train) // 2.5, with_replacement=False)\n\nprint(df_train.shape)\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:07.594091Z","iopub.execute_input":"2024-03-17T19:17:07.594382Z","iopub.status.idle":"2024-03-17T19:17:12.316057Z","shell.execute_reply.started":"2024-03-17T19:17:07.594356Z","shell.execute_reply":"2024-03-17T19:17:12.314779Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Test Files Read & Feature Engineering","metadata":{}},{"cell_type":"code","source":"data_store = {\n    \"df_base\": read_file(TEST_DIR / \"test_base.parquet\"),\n    \"depth_0\": [\n        read_file(TEST_DIR / \"test_static_cb_0.parquet\"),\n        read_files(TEST_DIR / \"test_static_0_*.parquet\"),\n    ],\n    \"depth_1\": [\n        read_files(TEST_DIR / \"test_applprev_1_*.parquet\", 1),\n        read_file(TEST_DIR / \"test_tax_registry_a_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_tax_registry_b_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_tax_registry_c_1.parquet\", 1),\n        read_files(TEST_DIR / \"test_credit_bureau_a_1_*.parquet\", 1),\n        read_file(TEST_DIR / \"test_credit_bureau_b_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_other_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_person_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_deposit_1.parquet\", 1),\n        read_file(TEST_DIR / \"test_debitcard_1.parquet\", 1),\n    ],\n    \"depth_2\": [\n        read_file(TEST_DIR / \"test_credit_bureau_b_2.parquet\", 2),\n        read_files(TEST_DIR / \"test_credit_bureau_a_2_*.parquet\", 2),\n    ]\n}","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:12.317777Z","iopub.execute_input":"2024-03-17T19:17:12.318110Z","iopub.status.idle":"2024-03-17T19:17:12.955334Z","shell.execute_reply.started":"2024-03-17T19:17:12.318082Z","shell.execute_reply":"2024-03-17T19:17:12.954175Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test = feature_eng(**data_store)\n\nprint(\"test data shape:\\t\", df_test.shape)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:12.959790Z","iopub.execute_input":"2024-03-17T19:17:12.960179Z","iopub.status.idle":"2024-03-17T19:17:13.022766Z","shell.execute_reply.started":"2024-03-17T19:17:12.960149Z","shell.execute_reply":"2024-03-17T19:17:13.021942Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Feature Elimination","metadata":{}},{"cell_type":"code","source":"df_train.shape","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:13.023907Z","iopub.execute_input":"2024-03-17T19:17:13.024855Z","iopub.status.idle":"2024-03-17T19:17:13.037002Z","shell.execute_reply.started":"2024-03-17T19:17:13.024824Z","shell.execute_reply":"2024-03-17T19:17:13.035270Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"common_columns = list(set(df_train.columns) & set(df_test.columns))\n\ncommon_columns_train = common_columns.copy()\ncommon_columns_train.append(\"target\")\n\ndf_train = df_train.select(common_columns_train)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:13.038944Z","iopub.execute_input":"2024-03-17T19:17:13.039418Z","iopub.status.idle":"2024-03-17T19:17:13.052223Z","shell.execute_reply.started":"2024-03-17T19:17:13.039377Z","shell.execute_reply":"2024-03-17T19:17:13.051082Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test = df_test.select(common_columns)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:13.053979Z","iopub.execute_input":"2024-03-17T19:17:13.055031Z","iopub.status.idle":"2024-03-17T19:17:13.068008Z","shell.execute_reply.started":"2024-03-17T19:17:13.054983Z","shell.execute_reply":"2024-03-17T19:17:13.066798Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"del data_store\n\ngc.collect()","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:13.069830Z","iopub.execute_input":"2024-03-17T19:17:13.070602Z","iopub.status.idle":"2024-03-17T19:17:13.944666Z","shell.execute_reply.started":"2024-03-17T19:17:13.070555Z","shell.execute_reply":"2024-03-17T19:17:13.943471Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Pandas Conversion","metadata":{}},{"cell_type":"markdown","source":"1526659","metadata":{}},{"cell_type":"code","source":"df_train, cat_cols = to_pandas(df_train)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:13.946412Z","iopub.execute_input":"2024-03-17T19:17:13.946816Z","iopub.status.idle":"2024-03-17T19:17:22.180481Z","shell.execute_reply.started":"2024-03-17T19:17:13.946781Z","shell.execute_reply":"2024-03-17T19:17:22.178751Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test, cat_cols = to_pandas(df_test, cat_cols)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:22.182222Z","iopub.execute_input":"2024-03-17T19:17:22.182632Z","iopub.status.idle":"2024-03-17T19:17:22.262848Z","shell.execute_reply.started":"2024-03-17T19:17:22.182597Z","shell.execute_reply":"2024-03-17T19:17:22.261890Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import psutil\n\n# Get the total memory in bytes\ntotal_memory = psutil.virtual_memory().total\n# Get the available memory in bytes\navailable_memory = psutil.virtual_memory().available\n# Calculate the used memory in bytes\nused_memory = total_memory - available_memory\n# Convert the memory usage to MB\nused_memory_mb = used_memory / (1024 * 1024)\n\nprint(f\"Used memory: {used_memory_mb:.2f} MB\")\nprint(total_memory / (1024 * 1024))\nprint(available_memory / (1024 * 1024))\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:22.264049Z","iopub.execute_input":"2024-03-17T19:17:22.264972Z","iopub.status.idle":"2024-03-17T19:17:22.272268Z","shell.execute_reply.started":"2024-03-17T19:17:22.264940Z","shell.execute_reply":"2024-03-17T19:17:22.271259Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Garbage Collection","metadata":{}},{"cell_type":"markdown","source":"### Training","metadata":{}},{"cell_type":"code","source":"categorical_columns_train = set(df_train.select_dtypes(include=['category', 'object']).columns)\n\nprint(len(categorical_columns_train))","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:31.027658Z","iopub.execute_input":"2024-03-17T19:17:31.028037Z","iopub.status.idle":"2024-03-17T19:17:31.059783Z","shell.execute_reply.started":"2024-03-17T19:17:31.028003Z","shell.execute_reply":"2024-03-17T19:17:31.058649Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Find categorical columns in each dataset\ncategorical_columns_train = set(df_train.select_dtypes(include=['category', 'object']).columns)\ncategorical_columns_test = set(df_test.select_dtypes(include=['category', 'object']).columns)\n\n# Find columns that are categorical in one dataset but not in the other\nmismatched_categorical_columns = (categorical_columns_train - categorical_columns_test) | (categorical_columns_test - categorical_columns_train)\n\n# Drop these columns from both datasets\ndf_train.drop(columns=mismatched_categorical_columns, inplace=True)\ndf_test.drop(columns=mismatched_categorical_columns, inplace=True)\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:31.061149Z","iopub.execute_input":"2024-03-17T19:17:31.061527Z","iopub.status.idle":"2024-03-17T19:17:32.907049Z","shell.execute_reply.started":"2024-03-17T19:17:31.061495Z","shell.execute_reply":"2024-03-17T19:17:32.905828Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(df_test.shape)\nprint(df_train.shape)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:32.908745Z","iopub.execute_input":"2024-03-17T19:17:32.909149Z","iopub.status.idle":"2024-03-17T19:17:32.915920Z","shell.execute_reply.started":"2024-03-17T19:17:32.909111Z","shell.execute_reply":"2024-03-17T19:17:32.914731Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y = df_train[\"target\"]\nweeks = df_train[\"WEEK_NUM\"]\ndf_train.drop(columns=[\"target\", \"case_id\", \"WEEK_NUM\"], inplace = True)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:32.917474Z","iopub.execute_input":"2024-03-17T19:17:32.917877Z","iopub.status.idle":"2024-03-17T19:17:34.457566Z","shell.execute_reply.started":"2024-03-17T19:17:32.917837Z","shell.execute_reply":"2024-03-17T19:17:34.456262Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":" I've tried also to leave only stationary features but it didn't improve the auc score","metadata":{}},{"cell_type":"code","source":"#from sklearn.feature_selection import VarianceThreshold\n\n#selector = VarianceThreshold(threshold=0.01)  # Adjust the threshold as needed\n#reduced_data_array = selector.fit_transform(df_train[numeric_columns])\n\n#mask = selector.get_support()\n\n#columns_left = df_train[numeric_columns].columns[mask]\n\n#reduced_df = pd.DataFrame(reduced_data_array, columns=columns_left, index=df_train.index)\n\n#categorical_columns = df_train.select_dtypes(include=['category', 'object']).columns\n#categorical_df = df_train[categorical_columns]\n\n#final_df = pd.concat([reduced_df, categorical_df], axis=1)\n\n#print(\"Final DataFrame with reduced numeric columns and categorical columns:\")\n#print(final_df)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:34.458858Z","iopub.execute_input":"2024-03-17T19:17:34.460726Z","iopub.status.idle":"2024-03-17T19:17:34.472785Z","shell.execute_reply.started":"2024-03-17T19:17:34.460690Z","shell.execute_reply":"2024-03-17T19:17:34.471766Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"numeric_columns = df_train.select_dtypes(include=[np.number]).columns\ndf_train = df_train[numeric_columns].apply(np.log1p)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:34.477813Z","iopub.execute_input":"2024-03-17T19:17:34.478621Z","iopub.status.idle":"2024-03-17T19:17:39.873460Z","shell.execute_reply.started":"2024-03-17T19:17:34.478587Z","shell.execute_reply":"2024-03-17T19:17:39.872460Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.model_selection import train_test_split\n\nX_train, X_val, y_train, y_val, weeks_train, weeks_val = train_test_split(df_train, y, weeks, test_size=0.2, random_state=42)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:39.875216Z","iopub.execute_input":"2024-03-17T19:17:39.875936Z","iopub.status.idle":"2024-03-17T19:17:43.354896Z","shell.execute_reply.started":"2024-03-17T19:17:39.875901Z","shell.execute_reply":"2024-03-17T19:17:43.353513Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X_val, X_test, y_val, y_test, _, _ = train_test_split(X_val, y_val, weeks_val, test_size=0.5, random_state=42)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:43.357207Z","iopub.execute_input":"2024-03-17T19:17:43.358174Z","iopub.status.idle":"2024-03-17T19:17:44.123145Z","shell.execute_reply.started":"2024-03-17T19:17:43.358126Z","shell.execute_reply":"2024-03-17T19:17:44.122240Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(X_train.shape, y_train.shape)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:44.124544Z","iopub.execute_input":"2024-03-17T19:17:44.125605Z","iopub.status.idle":"2024-03-17T19:17:44.130677Z","shell.execute_reply.started":"2024-03-17T19:17:44.125563Z","shell.execute_reply":"2024-03-17T19:17:44.129628Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"cv = StratifiedGroupKFold(n_splits=5, shuffle=False)\n\nparams = {\n    \"boosting_type\": \"gbdt\",\n    \"objective\": \"binary\",\n    \"metric\": \"auc\",\n    \"max_depth\": 3,\n    \"learning_rate\": 0.1,\n    \"n_estimators\": 1000,\n    \"colsample_bytree\": 0.8,\n    \"colsample_bynode\": 0.9,\n    \"verbose\": -1,\n    \"random_state\": 42,\n}\n\nfitted_models = []\n\nfor idx_train, idx_valid in cv.split(X_train, y_train, groups=weeks_train):\n    X_train_cv, y_train_cv = X_train.iloc[idx_train], y_train.iloc[idx_train]\n    X_valid, y_valid = X_train.iloc[idx_valid], y_train.iloc[idx_valid]\n\n    model = lgb.LGBMClassifier(**params)\n    model.fit(\n        X_train_cv, y_train_cv,\n        eval_set=[(X_valid, y_valid)],\n        callbacks=[lgb.log_evaluation(100), lgb.early_stopping(100)]\n    )\n\n    fitted_models.append(model)\n\nmodel = VotingModel(fitted_models)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:17:44.132268Z","iopub.execute_input":"2024-03-17T19:17:44.132683Z","iopub.status.idle":"2024-03-17T19:18:24.651721Z","shell.execute_reply.started":"2024-03-17T19:17:44.132645Z","shell.execute_reply":"2024-03-17T19:18:24.650069Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.linear_model import LogisticRegression\nfrom sklearn.metrics import roc_auc_score\n\n\nclass SmartVote:\n    def __init__(self, models):\n        super().__init__()\n        self.models = models\n        \n    def fit(self, X, y):\n        predictions = []\n\n        for model in self.models:\n            y_pred_proba = model.predict_proba(X)[:, 1]\n            predictions.append(y_pred_proba)\n\n        X_train = np.concatenate([np.expand_dims(y_pred, axis=1) for y_pred in predictions], axis=1)\n        X_train = pd.DataFrame(X_train, columns=[f\"model_{i}_pred_proba\" for i in range(len(fitted_models))])\n\n        self.v_model = LogisticRegression(solver='lbfgs', max_iter=1000, class_weight='balanced')\n        self.v_model.fit(X_train, y)\n\n        y_pred_proba = self.v_model.predict_proba(X_train)[:, 1]\n\n        auc_score = roc_auc_score(y, y_pred_proba)\n        print(\"AUC Score:\", auc_score)\n\n    def predict_proba(self, X):\n        predictions = []\n\n        for model in self.models:\n            y_pred_proba = model.predict_proba(X)[:, 1]\n            predictions.append(y_pred_proba)\n            \n        X_train = np.concatenate([np.expand_dims(y_pred, axis=1) for y_pred in predictions], axis=1)\n        X_train = pd.DataFrame(X_train, columns=[f\"model_{i}_pred_proba\" for i in range(len(fitted_models))])\n        \n        y_pred_proba = self.v_model.predict_proba(X_train)[:, 1]\n        \n        return y_pred_proba\n    \n    def test(self, X, y):\n        y_pred_proba = self.predict_proba(X)\n\n        auc_score = roc_auc_score(y, y_pred_proba)\n        print(\"AUC Score:\", auc_score)\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.653097Z","iopub.status.idle":"2024-03-17T19:18:24.653583Z","shell.execute_reply.started":"2024-03-17T19:18:24.653351Z","shell.execute_reply":"2024-03-17T19:18:24.653372Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np\nimport pandas as pd\nfrom sklearn.model_selection import train_test_split\n\nvmodel = SmartVote(fitted_models)\nvmodel.fit(X_val, y_val)\n\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.655012Z","iopub.status.idle":"2024-03-17T19:18:24.655385Z","shell.execute_reply.started":"2024-03-17T19:18:24.655210Z","shell.execute_reply":"2024-03-17T19:18:24.655225Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"vmodel.test(X_test, y_test)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.656397Z","iopub.status.idle":"2024-03-17T19:18:24.656812Z","shell.execute_reply.started":"2024-03-17T19:18:24.656638Z","shell.execute_reply":"2024-03-17T19:18:24.656654Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"feature_importances = np.zeros((len(fitted_models[0].feature_importances_), len(fitted_models)))\n\nfor i, model in enumerate(fitted_models):\n    feature_importances[:, i] = model.feature_importances_\n\navg_feature_importances = np.mean(feature_importances, axis=1)\n\nfeatures_df = pd.DataFrame({\n    'Feature': fitted_models[0].feature_name_,\n    'Importance': avg_feature_importances\n})\n\nfeatures_df = features_df.sort_values(by='Importance', ascending=False)\n\nN = 30\ntop_features = features_df.head(N)\n\nprint(top_features)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.658570Z","iopub.status.idle":"2024-03-17T19:18:24.659238Z","shell.execute_reply.started":"2024-03-17T19:18:24.659056Z","shell.execute_reply":"2024-03-17T19:18:24.659072Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"features_info = pd.read_csv(\"/kaggle/input/home-credit-credit-risk-model-stability/feature_definitions.csv\")","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.660676Z","iopub.status.idle":"2024-03-17T19:18:24.661022Z","shell.execute_reply.started":"2024-03-17T19:18:24.660856Z","shell.execute_reply":"2024-03-17T19:18:24.660870Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"features_info['Variable']","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.662074Z","iopub.status.idle":"2024-03-17T19:18:24.662408Z","shell.execute_reply.started":"2024-03-17T19:18:24.662245Z","shell.execute_reply":"2024-03-17T19:18:24.662258Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"merged_df = pd.merge(features_info, top_features, left_on=['Variable'], right_on=['Feature'])\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.663436Z","iopub.status.idle":"2024-03-17T19:18:24.664108Z","shell.execute_reply.started":"2024-03-17T19:18:24.663904Z","shell.execute_reply":"2024-03-17T19:18:24.663920Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"merged_df = merged_df.sort_values(by='Importance', ascending=False)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.665116Z","iopub.status.idle":"2024-03-17T19:18:24.665471Z","shell.execute_reply.started":"2024-03-17T19:18:24.665289Z","shell.execute_reply":"2024-03-17T19:18:24.665304Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"merged_df","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.666500Z","iopub.status.idle":"2024-03-17T19:18:24.666838Z","shell.execute_reply.started":"2024-03-17T19:18:24.666673Z","shell.execute_reply":"2024-03-17T19:18:24.666687Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train[\"validfrom_1069D\"]","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.667771Z","iopub.status.idle":"2024-03-17T19:18:24.668100Z","shell.execute_reply.started":"2024-03-17T19:18:24.667937Z","shell.execute_reply":"2024-03-17T19:18:24.667951Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Optuna finding of the best hp's","metadata":{}},{"cell_type":"code","source":"#import optuna\n#import lightgbm as lgb\n#from sklearn.model_selection import train_test_split\n#from sklearn.metrics import roc_auc_score\n\n#def objective(trial):\n#    X_train_1, X_test, y_train_1, y_test = train_test_split(X_train, y_train, test_size=0.3, random_state=42)\n\n#    param = {\n#        \"boosting_type\": \"gbdt\",\n#        \"objective\": \"binary\",\n#        \"metric\": \"auc\",\n#        \"verbose\": -1,\n#        \"random_state\": 42,\n#        \"max_depth\": trial.suggest_int(\"max_depth\", 3, 10),\n#        \"learning_rate\": trial.suggest_float(\"learning_rate\", 0.01, 0.1),\n#        \"colsample_bytree\": trial.suggest_float(\"colsample_bytree\", 0.5, 1.0),\n#        \"colsample_bynode\": trial.suggest_float(\"colsample_bynode\", 0.5, 1.0),\n#    }\n\n#    model = lgb.LGBMClassifier(**param)\n#    model.fit(X_train_1, y_train_1)\n\n#    y_pred = model.predict_proba(X_test)[:, 1]\n#    score = roc_auc_score(y_test, y_pred)\n\n#    return score\n\n#study = optuna.create_study(direction=\"maximize\")\n#study.optimize(objective, n_trials=100)\n\n#print(\"Best trial:\", study.best_trial.params)\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.669339Z","iopub.status.idle":"2024-03-17T19:18:24.669701Z","shell.execute_reply.started":"2024-03-17T19:18:24.669532Z","shell.execute_reply":"2024-03-17T19:18:24.669546Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"PCA","metadata":{}},{"cell_type":"code","source":"#import pandas as pd\n#from sklearn.decomposition import PCA\n#from sklearn.experimental import enable_iterative_imputer\n#from sklearn.impute import IterativeImputer\n\n#categorical_columns = set(df_train.select_dtypes(include=['category', 'object']).columns)\n#continuous_columns = set(df_train.columns) - categorical_columns\n\n#imputer = IterativeImputer(max_iter=10, random_state=0)\n#df_imputed = pd.DataFrame(imputer.fit_transform(df_train[list(continuous_columns)]), columns=continuous_columns)\n\n#pca = PCA(n_components=20)\n#reduced_data = pca.fit_transform(df_imputed)\n\n\n#df_train = pd.concat([\n#    pd.DataFrame(reduced_data, index=df_train.index),\n#    pd.DataFrame(df_train[categorical_columns], index=df_train.index)\n#], axis=1)\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.670768Z","iopub.status.idle":"2024-03-17T19:18:24.671122Z","shell.execute_reply.started":"2024-03-17T19:18:24.670946Z","shell.execute_reply":"2024-03-17T19:18:24.670962Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Prediction","metadata":{}},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"code","source":"case_id = df_test[\"case_id\"]\ndf_test = df_test[df_test.columns.intersection(df_train.columns)]\n","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.671994Z","iopub.status.idle":"2024-03-17T19:18:24.672667Z","shell.execute_reply.started":"2024-03-17T19:18:24.672460Z","shell.execute_reply":"2024-03-17T19:18:24.672481Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test = df_test[numeric_columns].apply(np.log1p)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.673589Z","iopub.status.idle":"2024-03-17T19:18:24.674262Z","shell.execute_reply.started":"2024-03-17T19:18:24.674082Z","shell.execute_reply":"2024-03-17T19:18:24.674098Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_test.shape","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.675831Z","iopub.status.idle":"2024-03-17T19:18:24.676271Z","shell.execute_reply.started":"2024-03-17T19:18:24.676098Z","shell.execute_reply":"2024-03-17T19:18:24.676113Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_train.shape","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.677226Z","iopub.status.idle":"2024-03-17T19:18:24.677603Z","shell.execute_reply.started":"2024-03-17T19:18:24.677413Z","shell.execute_reply":"2024-03-17T19:18:24.677443Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"probas = vmodel.predict_proba(df_test)\n\ny_pred = pd.Series(probas, index=df_test.index)","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.680456Z","iopub.status.idle":"2024-03-17T19:18:24.681245Z","shell.execute_reply.started":"2024-03-17T19:18:24.681043Z","shell.execute_reply":"2024-03-17T19:18:24.681062Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Submission","metadata":{}},{"cell_type":"code","source":"submission = pd.DataFrame({\n    \"case_id\": case_id,\n    \"score\": y_pred\n}).set_index('case_id')","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.682449Z","iopub.status.idle":"2024-03-17T19:18:24.682801Z","shell.execute_reply.started":"2024-03-17T19:18:24.682630Z","shell.execute_reply":"2024-03-17T19:18:24.682644Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission.head()","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.683990Z","iopub.status.idle":"2024-03-17T19:18:24.684405Z","shell.execute_reply.started":"2024-03-17T19:18:24.684193Z","shell.execute_reply":"2024-03-17T19:18:24.684208Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission.to_csv(\"submission.csv\")","metadata":{"execution":{"iopub.status.busy":"2024-03-17T19:18:24.685612Z","iopub.status.idle":"2024-03-17T19:18:24.685965Z","shell.execute_reply.started":"2024-03-17T19:18:24.685790Z","shell.execute_reply":"2024-03-17T19:18:24.685805Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}