{ "cells": [ { "cell_type": "markdown", "id": "6d28f4ac", "metadata": {}, "source": [ "# Прототипная нейросеть для задачи регрессии\n", "\n", "В работе автоматически определяется зависимая переменная, выполняется предобработка данных, обучается прототипная нейросеть для регрессии и анализируются основные прототипы.\n", "\n", "Под **прототипной нейросетью** далее понимается RBF-сеть:\n", "- сначала выделяются прототипы объектов методом `KMeans`,\n", "- затем по расстояниям до прототипов строятся радиально-базисные активации,\n", "- по этим активациям обучается линейный регрессионный выходной слой.\n", "\n", "В конце ноутбука есть отдельная ячейка, которая печатает все ключевые результаты для итогового вывода." ] }, { "cell_type": "code", "execution_count": 1, "id": "c927c6f3", "metadata": {}, "outputs": [], "source": [ "import os\n", "import re\n", "import glob\n", "import math\n", "import random\n", "import warnings\n", "\n", "import numpy as np\n", "import pandas as pd\n", "import matplotlib.pyplot as plt\n", "\n", "from sklearn.base import BaseEstimator, RegressorMixin\n", "from sklearn.compose import ColumnTransformer\n", "from sklearn.impute import SimpleImputer\n", "from sklearn.cluster import KMeans\n", "from sklearn.pipeline import Pipeline\n", "from sklearn.linear_model import Ridge\n", "from sklearn.metrics import r2_score, mean_absolute_error, mean_squared_error\n", "from sklearn.model_selection import train_test_split\n", "from sklearn.preprocessing import OneHotEncoder, StandardScaler\n", "from sklearn.metrics import pairwise_distances\n", "\n", "warnings.filterwarnings(\"ignore\")\n", "\n", "SEED = 42\n", "random.seed(SEED)\n", "np.random.seed(SEED)\n", "\n", "pd.set_option(\"display.max_columns\", 200)\n", "pd.set_option(\"display.width\", 200)" ] }, { "cell_type": "code", "execution_count": 2, "id": "ebaa2660", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Используемый файл: ./housing_price_dataset.csv\n", "Форма датасета: (50000, 6)\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
SquareFeetBedroomsBathroomsNeighborhoodYearBuiltPrice
0212641Rural1969215355.283618
1245932Rural1980195014.221626
2186021Suburb1970306891.012076
3229421Urban1996206786.787153
4213052Suburb2001272436.239065
\n", "
" ], "text/plain": [ " SquareFeet Bedrooms Bathrooms Neighborhood YearBuilt Price\n", "0 2126 4 1 Rural 1969 215355.283618\n", "1 2459 3 2 Rural 1980 195014.221626\n", "2 1860 2 1 Suburb 1970 306891.012076\n", "3 2294 2 1 Urban 1996 206786.787153\n", "4 2130 5 2 Suburb 2001 272436.239065" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "\n", "Типы данных:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
dtype
SquareFeetint64
Bedroomsint64
Bathroomsint64
Neighborhoodobject
YearBuiltint64
Pricefloat64
\n", "
" ], "text/plain": [ " dtype\n", "SquareFeet int64\n", "Bedrooms int64\n", "Bathrooms int64\n", "Neighborhood object\n", "YearBuilt int64\n", "Price float64" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "def find_csv_file():\n", " candidates = [\n", " \"housing_price_dataset.csv\",\n", " \"Housing price dataset.csv\",\n", " \"Housing Price Dataset.csv\",\n", " \"Housing Price dataset.csv\",\n", " \"housing.csv\",\n", " \"data.csv\",\n", " ]\n", " search_dirs = [\".\", \"/mnt/data\", os.getcwd()]\n", "\n", " for d in search_dirs:\n", " for name in candidates:\n", " path = os.path.join(d, name)\n", " if os.path.exists(path):\n", " return path\n", "\n", " patterns = [\n", " \"*.csv\",\n", " \"/mnt/data/*.csv\",\n", " ]\n", " files = []\n", " for p in patterns:\n", " files.extend(glob.glob(p))\n", "\n", " # Ищем наиболее подходящий файл по имени\n", " priority_tokens = [\"housing\", \"price\", \"dataset\"]\n", " scored = []\n", " for f in files:\n", " base = os.path.basename(f).lower()\n", " score = sum(tok in base for tok in priority_tokens)\n", " scored.append((score, f))\n", " scored.sort(reverse=True)\n", " if scored and scored[0][0] > 0:\n", " return scored[0][1]\n", " if files:\n", " return files[0]\n", " raise FileNotFoundError(\"CSV-файл не найден. Поместите housing_price_dataset.csv рядом с ноутбуком.\")\n", "\n", "csv_path = find_csv_file()\n", "print(\"Используемый файл:\", csv_path)\n", "\n", "df_raw = pd.read_csv(csv_path)\n", "print(\"Форма датасета:\", df_raw.shape)\n", "display(df_raw.head())\n", "print(\"\\nТипы данных:\")\n", "display(df_raw.dtypes.rename(\"dtype\").to_frame())" ] }, { "cell_type": "code", "execution_count": 3, "id": "2ef5eeb1", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
original_namenormalized_name
0SquareFeetsquarefeet
1Bedroomsbedrooms
2Bathroomsbathrooms
3Neighborhoodneighborhood
4YearBuiltyearbuilt
5Priceprice
\n", "
" ], "text/plain": [ " original_name normalized_name\n", "0 SquareFeet squarefeet\n", "1 Bedrooms bedrooms\n", "2 Bathrooms bathrooms\n", "3 Neighborhood neighborhood\n", "4 YearBuilt yearbuilt\n", "5 Price price" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Выбранная зависимая переменная: price\n", "Найденные идентификаторы / служебные поля: []\n", "Размер после удаления пропусков по целевой переменной: (50000, 5)\n", "\n", "Числовые признаки: ['squarefeet', 'bedrooms', 'bathrooms', 'yearbuilt']\n", "Категориальные признаки: ['neighborhood']\n" ] } ], "source": [ "def normalize_col_name(col):\n", " col = str(col).strip().lower()\n", " col = col.replace(\"/\", \"_\").replace(\"-\", \"_\")\n", " col = re.sub(r\"[^a-z0-9_]+\", \"_\", col)\n", " col = re.sub(r\"_+\", \"_\", col).strip(\"_\")\n", " return col\n", "\n", "original_columns = list(df_raw.columns)\n", "normalized_columns = [normalize_col_name(c) for c in original_columns]\n", "\n", "column_map_df = pd.DataFrame({\n", " \"original_name\": original_columns,\n", " \"normalized_name\": normalized_columns\n", "})\n", "display(column_map_df)\n", "\n", "df = df_raw.copy()\n", "df.columns = normalized_columns\n", "\n", "possible_targets = [\n", " \"price\", \"target\", \"y\", \"saleprice\", \"amount\", \"cost\"\n", "]\n", "\n", "target_col = None\n", "for c in possible_targets:\n", " if c in df.columns:\n", " target_col = c\n", " break\n", "\n", "if target_col is None:\n", " numeric_candidates = df.select_dtypes(include=[np.number]).columns.tolist()\n", " if not numeric_candidates:\n", " raise ValueError(\"Не удалось автоматически определить числовую зависимую переменную.\")\n", " target_col = numeric_candidates[-1]\n", "\n", "print(\"Выбранная зависимая переменная:\", target_col)\n", "\n", "# Удаляем явные идентификаторы\n", "id_like = [c for c in df.columns if c != target_col and (\"id\" in c or c in {\"name\"})]\n", "print(\"Найденные идентификаторы / служебные поля:\", id_like)\n", "\n", "X = df.drop(columns=[target_col] + id_like, errors=\"ignore\").copy()\n", "y = pd.to_numeric(df[target_col], errors=\"coerce\").copy()\n", "\n", "valid_mask = y.notna()\n", "X = X.loc[valid_mask].reset_index(drop=True)\n", "y = y.loc[valid_mask].reset_index(drop=True)\n", "\n", "print(\"Размер после удаления пропусков по целевой переменной:\", X.shape)\n", "\n", "numeric_features = X.select_dtypes(include=[np.number]).columns.tolist()\n", "categorical_features = [c for c in X.columns if c not in numeric_features]\n", "\n", "print(\"\\nЧисловые признаки:\", numeric_features)\n", "print(\"Категориальные признаки:\", categorical_features)" ] }, { "cell_type": "code", "execution_count": 4, "id": "158aea3b", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Размеры выборок:\n", "train: (30000, 5) (30000,)\n", "validation: (10000, 5) (10000,)\n", "test: (10000, 5) (10000,)\n", "Число исходных признаков: 5\n", "Число признаков после кодирования: 7\n", "Первые обработанные признаки: ['num__squarefeet', 'num__bedrooms', 'num__bathrooms', 'num__yearbuilt', 'cat__neighborhood_Rural', 'cat__neighborhood_Suburb', 'cat__neighborhood_Urban']\n" ] } ], "source": [ "X_train_full, X_test, y_train_full, y_test = train_test_split(\n", " X, y, test_size=0.20, random_state=SEED\n", ")\n", "\n", "X_train, X_val, y_train, y_val = train_test_split(\n", " X_train_full, y_train_full, test_size=0.25, random_state=SEED\n", ")\n", "\n", "print(\"Размеры выборок:\")\n", "print(\"train:\", X_train.shape, y_train.shape)\n", "print(\"validation:\", X_val.shape, y_val.shape)\n", "print(\"test:\", X_test.shape, y_test.shape)\n", "\n", "num_pipe = Pipeline([\n", " (\"imputer\", SimpleImputer(strategy=\"median\")),\n", " (\"scaler\", StandardScaler()),\n", "])\n", "\n", "try:\n", " ohe = OneHotEncoder(handle_unknown=\"ignore\", sparse_output=False)\n", "except TypeError:\n", " ohe = OneHotEncoder(handle_unknown=\"ignore\", sparse=False)\n", "\n", "cat_pipe = Pipeline([\n", " (\"imputer\", SimpleImputer(strategy=\"most_frequent\")),\n", " (\"onehot\", ohe),\n", "])\n", "\n", "preprocessor = ColumnTransformer([\n", " (\"num\", num_pipe, numeric_features),\n", " (\"cat\", cat_pipe, categorical_features),\n", "], remainder=\"drop\")\n", "\n", "X_train_proc = preprocessor.fit_transform(X_train)\n", "X_val_proc = preprocessor.transform(X_val)\n", "X_test_proc = preprocessor.transform(X_test)\n", "\n", "feature_names_processed = preprocessor.get_feature_names_out()\n", "\n", "print(\"Число исходных признаков:\", X.shape[1])\n", "print(\"Число признаков после кодирования:\", X_train_proc.shape[1])\n", "print(\"Первые обработанные признаки:\", list(feature_names_processed[:20]))" ] }, { "cell_type": "code", "execution_count": 5, "id": "a4f4c3c3", "metadata": {}, "outputs": [], "source": [ "class PrototypeRBFRegressor(BaseEstimator, RegressorMixin):\n", " def __init__(self, n_prototypes=8, alpha=1.0, random_state=42):\n", " self.n_prototypes = n_prototypes\n", " self.alpha = alpha\n", " self.random_state = random_state\n", "\n", " def _rbf_features(self, X):\n", " d2 = pairwise_distances(X, self.prototypes_, metric=\"euclidean\") ** 2\n", " gamma = 1.0 / max(2.0 * (self.sigma_ ** 2), 1e-12)\n", " return np.exp(-gamma * d2)\n", "\n", " def fit(self, X, y):\n", " X = np.asarray(X, dtype=float)\n", " y = np.asarray(y, dtype=float)\n", "\n", " n_clusters = int(min(max(2, self.n_prototypes), len(X)))\n", " self.kmeans_ = KMeans(n_clusters=n_clusters, random_state=self.random_state, n_init=10)\n", " self.kmeans_.fit(X)\n", " self.prototypes_ = self.kmeans_.cluster_centers_\n", "\n", " assigned = self.kmeans_.predict(X)\n", " dist_to_center = np.linalg.norm(X - self.prototypes_[assigned], axis=1)\n", " self.sigma_ = float(np.median(dist_to_center))\n", " if not np.isfinite(self.sigma_) or self.sigma_ <= 1e-8:\n", " all_d = pairwise_distances(self.prototypes_)\n", " self.sigma_ = float(np.median(all_d[all_d > 0])) if np.any(all_d > 0) else 1.0\n", "\n", " Phi = self._rbf_features(X)\n", " self.ridge_ = Ridge(alpha=self.alpha, random_state=self.random_state)\n", " self.ridge_.fit(Phi, y)\n", " return self\n", "\n", " def predict(self, X):\n", " X = np.asarray(X, dtype=float)\n", " Phi = self._rbf_features(X)\n", " return self.ridge_.predict(Phi)\n", "\n", " @property\n", " def prototype_weights_(self):\n", " return np.asarray(self.ridge_.coef_, dtype=float).ravel()\n", "\n", "def regression_metrics(y_true, y_pred):\n", " return {\n", " \"r2\": r2_score(y_true, y_pred),\n", " \"rmse\": float(np.sqrt(mean_squared_error(y_true, y_pred))),\n", " \"mae\": mean_absolute_error(y_true, y_pred),\n", " }" ] }, { "cell_type": "code", "execution_count": 6, "id": "d0be083a", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Лучшие конфигурации по validation RMSE:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
n_prototypesalphatrain_r2train_rmsetrain_maeval_r2val_rmseval_maesigma
081.00000.51017053409.39694342622.2460030.51279253050.56771242496.2730111.410672
180.10000.51017053409.38296142622.4097680.51278853050.78533742496.6367161.410672
280.01000.51017053409.38282242622.4264380.51278853050.80839742496.6741211.410672
380.00100.51017053409.38282042622.4281050.51278853050.81071642496.6778611.410672
480.00010.51017053409.38282042622.4282720.51278853050.81094842496.6782361.410672
5101.00000.50877453485.43996542675.4523220.51182253103.35501242523.5839381.364013
6100.10000.50877553485.42488742675.6365450.51181853103.55350842523.8244041.364013
7100.01000.50877553485.42473642675.6557970.51181853103.57474342523.8485071.364013
8100.00100.50877553485.42473442675.6577230.51181853103.57688042523.8509181.364013
9100.00010.50877553485.42473442675.6579150.51181853103.57709442523.8511591.364013
\n", "
" ], "text/plain": [ " n_prototypes alpha train_r2 train_rmse train_mae val_r2 val_rmse val_mae sigma\n", "0 8 1.0000 0.510170 53409.396943 42622.246003 0.512792 53050.567712 42496.273011 1.410672\n", "1 8 0.1000 0.510170 53409.382961 42622.409768 0.512788 53050.785337 42496.636716 1.410672\n", "2 8 0.0100 0.510170 53409.382822 42622.426438 0.512788 53050.808397 42496.674121 1.410672\n", "3 8 0.0010 0.510170 53409.382820 42622.428105 0.512788 53050.810716 42496.677861 1.410672\n", "4 8 0.0001 0.510170 53409.382820 42622.428272 0.512788 53050.810948 42496.678236 1.410672\n", "5 10 1.0000 0.508774 53485.439965 42675.452322 0.511822 53103.355012 42523.583938 1.364013\n", "6 10 0.1000 0.508775 53485.424887 42675.636545 0.511818 53103.553508 42523.824404 1.364013\n", "7 10 0.0100 0.508775 53485.424736 42675.655797 0.511818 53103.574743 42523.848507 1.364013\n", "8 10 0.0010 0.508775 53485.424734 42675.657723 0.511818 53103.576880 42523.850918 1.364013\n", "9 10 0.0001 0.508775 53485.424734 42675.657915 0.511818 53103.577094 42523.851159 1.364013" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Лучшая конфигурация:\n", "n_prototypes = 8\n", "alpha = 1.0\n", "sigma = 1.410672\n" ] } ], "source": [ "prototype_grid = [4, 6, 8, 10, 12]\n", "alpha_grid = [1e-4, 1e-3, 1e-2, 1e-1, 1.0]\n", "\n", "search_rows = []\n", "best_model = None\n", "best_key = None\n", "best_val_rmse = np.inf\n", "\n", "for n_prototypes in prototype_grid:\n", " for alpha in alpha_grid:\n", " model = PrototypeRBFRegressor(\n", " n_prototypes=n_prototypes,\n", " alpha=alpha,\n", " random_state=SEED\n", " )\n", " model.fit(X_train_proc, y_train)\n", "\n", " pred_train = model.predict(X_train_proc)\n", " pred_val = model.predict(X_val_proc)\n", "\n", " train_m = regression_metrics(y_train, pred_train)\n", " val_m = regression_metrics(y_val, pred_val)\n", "\n", " row = {\n", " \"n_prototypes\": n_prototypes,\n", " \"alpha\": alpha,\n", " \"train_r2\": train_m[\"r2\"],\n", " \"train_rmse\": train_m[\"rmse\"],\n", " \"train_mae\": train_m[\"mae\"],\n", " \"val_r2\": val_m[\"r2\"],\n", " \"val_rmse\": val_m[\"rmse\"],\n", " \"val_mae\": val_m[\"mae\"],\n", " \"sigma\": model.sigma_,\n", " }\n", " search_rows.append(row)\n", "\n", " if val_m[\"rmse\"] < best_val_rmse:\n", " best_val_rmse = val_m[\"rmse\"]\n", " best_model = model\n", " best_key = (n_prototypes, alpha)\n", "\n", "search_df = pd.DataFrame(search_rows).sort_values([\"val_rmse\", \"val_mae\", \"val_r2\"], ascending=[True, True, False]).reset_index(drop=True)\n", "\n", "print(\"Лучшие конфигурации по validation RMSE:\")\n", "display(search_df.head(10))\n", "\n", "best_n_prototypes, best_alpha = best_key\n", "print(\"Лучшая конфигурация:\")\n", "print(\"n_prototypes =\", best_n_prototypes)\n", "print(\"alpha =\", best_alpha)\n", "print(\"sigma =\", round(best_model.sigma_, 6))" ] }, { "cell_type": "code", "execution_count": 7, "id": "77eb02e8", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
splitr2rmsemae
0train0.51017053409.39694342622.246003
1validation0.51279253050.56771242496.273011
2test0.52106152431.73511541934.625581
\n", "
" ], "text/plain": [ " split r2 rmse mae\n", "0 train 0.510170 53409.396943 42622.246003\n", "1 validation 0.512792 53050.567712 42496.273011\n", "2 test 0.521061 52431.735115 41934.625581" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Первые 10 фактических и предсказанных значений на тесте:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
y_truey_pred
0170835.035713208189.809824
1126913.469998187521.135570
2246611.883092269935.842537
3244250.462969270102.333733
4271127.650112265637.877925
5189289.951199189107.502703
6343510.576500276873.696031
7203461.591080185461.893451
8285903.247682278803.162808
9224868.958375224449.420520
\n", "
" ], "text/plain": [ " y_true y_pred\n", "0 170835.035713 208189.809824\n", "1 126913.469998 187521.135570\n", "2 246611.883092 269935.842537\n", "3 244250.462969 270102.333733\n", "4 271127.650112 265637.877925\n", "5 189289.951199 189107.502703\n", "6 343510.576500 276873.696031\n", "7 203461.591080 185461.893451\n", "8 285903.247682 278803.162808\n", "9 224868.958375 224449.420520" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "pred_train = best_model.predict(X_train_proc)\n", "pred_val = best_model.predict(X_val_proc)\n", "pred_test = best_model.predict(X_test_proc)\n", "\n", "metrics_df = pd.DataFrame([\n", " {\"split\": \"train\", **regression_metrics(y_train, pred_train)},\n", " {\"split\": \"validation\", **regression_metrics(y_val, pred_val)},\n", " {\"split\": \"test\", **regression_metrics(y_test, pred_test)},\n", "])\n", "\n", "display(metrics_df)\n", "\n", "plt.figure(figsize=(7, 4))\n", "plt.plot(search_df[\"n_prototypes\"].astype(str) + \" | \" + search_df[\"alpha\"].astype(str), search_df[\"val_rmse\"], marker=\"o\")\n", "plt.xticks(rotation=90)\n", "plt.title(\"Validation RMSE для перебора конфигураций прототипной сети\")\n", "plt.ylabel(\"RMSE\")\n", "plt.xlabel(\"(число прототипов | alpha)\")\n", "plt.tight_layout()\n", "plt.show()\n", "\n", "comparison_df = pd.DataFrame({\n", " \"y_true\": y_test.reset_index(drop=True),\n", " \"y_pred\": pd.Series(pred_test).reset_index(drop=True)\n", "})\n", "print(\"Первые 10 фактических и предсказанных значений на тесте:\")\n", "display(comparison_df.head(10))" ] }, { "cell_type": "code", "execution_count": 8, "id": "743c25c3", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Основные прототипы (по impact_score = support_share × |вес выхода|):\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
prototype_idsupport_nsupport_share_trainprototype_output_weightimpact_scoreassigned_mean_target_trainnum__squarefeetnum__bedroomsnum__bathroomsnum__yearbuiltcat__neighborhood
0738440.12813395506.68714812237.590180272746.4398622424.0822494.4167102.6290991968.388860Urban
1037610.125367-96853.93459912142.254934175264.4224651541.3573332.5794372.6381511970.338735Suburb
2137340.12446796690.02303512034.684867272552.6606432460.8134874.4187851.3738292001.336098Suburb
3436170.120567-96452.14742011628.913907175220.7634011580.5098152.5886091.3596902002.136577Rural
4638630.128767-88722.80544011424.539914181692.4351291539.1755114.3831221.3453271970.988092Rural
5236960.12320086261.62622610627.432351267077.4016582469.3337842.5945362.6402492000.822288Suburb
6337740.12580076093.8678289572.608573261917.3456212444.7826662.5981981.3654921968.891333Urban
7537110.123700-71996.6241078905.982402191311.2102461594.2212034.4073372.6633402002.108713Suburb
\n", "
" ], "text/plain": [ " prototype_id support_n support_share_train prototype_output_weight impact_score assigned_mean_target_train num__squarefeet num__bedrooms num__bathrooms num__yearbuilt cat__neighborhood\n", "0 7 3844 0.128133 95506.687148 12237.590180 272746.439862 2424.082249 4.416710 2.629099 1968.388860 Urban\n", "1 0 3761 0.125367 -96853.934599 12142.254934 175264.422465 1541.357333 2.579437 2.638151 1970.338735 Suburb\n", "2 1 3734 0.124467 96690.023035 12034.684867 272552.660643 2460.813487 4.418785 1.373829 2001.336098 Suburb\n", "3 4 3617 0.120567 -96452.147420 11628.913907 175220.763401 1580.509815 2.588609 1.359690 2002.136577 Rural\n", "4 6 3863 0.128767 -88722.805440 11424.539914 181692.435129 1539.175511 4.383122 1.345327 1970.988092 Rural\n", "5 2 3696 0.123200 86261.626226 10627.432351 267077.401658 2469.333784 2.594536 2.640249 2000.822288 Suburb\n", "6 3 3774 0.125800 76093.867828 9572.608573 261917.345621 2444.782666 2.598198 1.365492 1968.891333 Urban\n", "7 5 3711 0.123700 -71996.624107 8905.982402 191311.210246 1594.221203 4.407337 2.663340 2002.108713 Suburb" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Топ-признаки внутри каждого прототипа:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
prototype_idrankfeaturecontribution_scoreinterpreted_value
001bedrooms80619.0419422.579437
102squarefeet78686.5089831541.357333
203bathrooms75406.2206922.638151
304yearbuilt70286.4702651970.338735
405neighborhood=Suburb467.942083Suburb
511bedrooms79061.9729034.418785
612squarefeet76262.2913512460.813487
713yearbuilt74551.6414712001.336098
814bathrooms74381.3275761.373829
915neighborhood=Suburb471.666369Suburb
1021bedrooms70633.7870112.594536
1122squarefeet69317.0007652469.333784
1223bathrooms67381.0679342.640249
1324yearbuilt64370.8267282000.822288
1425neighborhood=Suburb1251.300681Suburb
1531bedrooms62058.1473312.598198
1632yearbuilt60539.1067601968.891333
1733bathrooms59313.9056931.365492
1834squarefeet57893.2330032444.782666
1935neighborhood=Urban1074.079835Urban
2041bedrooms79490.9339952.588609
2142yearbuilt78096.2772682002.136577
2243bathrooms75867.8668591.35969
2344squarefeet71783.9127041580.509815
2445neighborhood=Rural683.868925Rural
2551bathrooms58273.5720122.66334
2652yearbuilt58198.0333522002.108713
2753bedrooms58131.1778634.407337
2854squarefeet51863.9626611594.221203
2955neighborhood=Suburb743.438785Suburb
3061squarefeet72417.6787931539.175511
3162bathrooms71348.1451381.345327
3263bedrooms69708.7795204.383122
3364yearbuilt61603.8667741970.988092
3465neighborhood=Rural1344.291752Rural
3571yearbuilt78300.8610041968.38886
3672bedrooms77916.5883834.41671
3773bathrooms73298.9869022.629099
3874squarefeet69219.9402122424.082249
3975neighborhood=Urban1426.920949Urban
\n", "
" ], "text/plain": [ " prototype_id rank feature contribution_score interpreted_value\n", "0 0 1 bedrooms 80619.041942 2.579437\n", "1 0 2 squarefeet 78686.508983 1541.357333\n", "2 0 3 bathrooms 75406.220692 2.638151\n", "3 0 4 yearbuilt 70286.470265 1970.338735\n", "4 0 5 neighborhood=Suburb 467.942083 Suburb\n", "5 1 1 bedrooms 79061.972903 4.418785\n", "6 1 2 squarefeet 76262.291351 2460.813487\n", "7 1 3 yearbuilt 74551.641471 2001.336098\n", "8 1 4 bathrooms 74381.327576 1.373829\n", "9 1 5 neighborhood=Suburb 471.666369 Suburb\n", "10 2 1 bedrooms 70633.787011 2.594536\n", "11 2 2 squarefeet 69317.000765 2469.333784\n", "12 2 3 bathrooms 67381.067934 2.640249\n", "13 2 4 yearbuilt 64370.826728 2000.822288\n", "14 2 5 neighborhood=Suburb 1251.300681 Suburb\n", "15 3 1 bedrooms 62058.147331 2.598198\n", "16 3 2 yearbuilt 60539.106760 1968.891333\n", "17 3 3 bathrooms 59313.905693 1.365492\n", "18 3 4 squarefeet 57893.233003 2444.782666\n", "19 3 5 neighborhood=Urban 1074.079835 Urban\n", "20 4 1 bedrooms 79490.933995 2.588609\n", "21 4 2 yearbuilt 78096.277268 2002.136577\n", "22 4 3 bathrooms 75867.866859 1.35969\n", "23 4 4 squarefeet 71783.912704 1580.509815\n", "24 4 5 neighborhood=Rural 683.868925 Rural\n", "25 5 1 bathrooms 58273.572012 2.66334\n", "26 5 2 yearbuilt 58198.033352 2002.108713\n", "27 5 3 bedrooms 58131.177863 4.407337\n", "28 5 4 squarefeet 51863.962661 1594.221203\n", "29 5 5 neighborhood=Suburb 743.438785 Suburb\n", "30 6 1 squarefeet 72417.678793 1539.175511\n", "31 6 2 bathrooms 71348.145138 1.345327\n", "32 6 3 bedrooms 69708.779520 4.383122\n", "33 6 4 yearbuilt 61603.866774 1970.988092\n", "34 6 5 neighborhood=Rural 1344.291752 Rural\n", "35 7 1 yearbuilt 78300.861004 1968.38886\n", "36 7 2 bedrooms 77916.588383 4.41671\n", "37 7 3 bathrooms 73298.986902 2.629099\n", "38 7 4 squarefeet 69219.940212 2424.082249\n", "39 7 5 neighborhood=Urban 1426.920949 Urban" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# Интерпретация прототипов\n", "\n", "num_pipe_fitted = preprocessor.named_transformers_[\"num\"]\n", "scaler = num_pipe_fitted.named_steps[\"scaler\"]\n", "\n", "if categorical_features:\n", " ohe_fitted = preprocessor.named_transformers_[\"cat\"].named_steps[\"onehot\"]\n", " cat_categories = ohe_fitted.categories_\n", "else:\n", " ohe_fitted = None\n", " cat_categories = []\n", "\n", "n_num = len(numeric_features)\n", "n_cat_proc = len(feature_names_processed) - n_num\n", "\n", "train_num_means = X_train[numeric_features].astype(float).mean() if numeric_features else pd.Series(dtype=float)\n", "train_num_stds = X_train[numeric_features].astype(float).std().replace(0, 1.0) if numeric_features else pd.Series(dtype=float)\n", "train_cat_freq = {c: X_train[c].astype(str).fillna(\"Missing\").value_counts(normalize=True).to_dict() for c in categorical_features}\n", "\n", "prototypes = best_model.prototypes_\n", "proto_weights = best_model.prototype_weights_\n", "train_assignments = best_model.kmeans_.predict(X_train_proc)\n", "\n", "prototype_rows = []\n", "prototype_feature_rows = []\n", "\n", "impact_scores = []\n", "\n", "for k in range(len(prototypes)):\n", " center = prototypes[k]\n", " beta = float(proto_weights[k])\n", "\n", " support_n = int((train_assignments == k).sum())\n", " support_share = support_n / len(train_assignments)\n", " assigned_mask = (train_assignments == k)\n", " assigned_mean_target = float(np.mean(y_train[assigned_mask])) if assigned_mask.any() else np.nan\n", "\n", " numeric_values = {}\n", " if n_num > 0:\n", " num_center_scaled = center[:n_num].reshape(1, -1)\n", " num_center_orig = scaler.inverse_transform(num_center_scaled).ravel()\n", " numeric_values = dict(zip(numeric_features, num_center_orig))\n", " else:\n", " num_center_orig = np.array([])\n", "\n", " dominant_cats = {}\n", " cat_offset = n_num\n", " cat_scores_local = []\n", " for cat_idx, cat_col in enumerate(categorical_features):\n", " cats = list(cat_categories[cat_idx])\n", " block = center[cat_offset: cat_offset + len(cats)]\n", " if len(block) == 0:\n", " cat_offset += len(cats)\n", " continue\n", " top_j = int(np.argmax(block))\n", " chosen_cat = str(cats[top_j])\n", " dominant_cats[cat_col] = chosen_cat\n", "\n", " global_freq = train_cat_freq.get(cat_col, {}).get(chosen_cat, 0.0)\n", " contribution_score = abs(float(block[top_j]) - global_freq) * abs(beta)\n", " cat_scores_local.append((f\"{cat_col}={chosen_cat}\", contribution_score, chosen_cat))\n", " cat_offset += len(cats)\n", "\n", " # Важность числовых признаков внутри прототипа\n", " num_scores_local = []\n", " for j, col in enumerate(numeric_features):\n", " z = abs((num_center_orig[j] - train_num_means[col]) / train_num_stds[col])\n", " contribution_score = float(z) * abs(beta)\n", " num_scores_local.append((col, contribution_score, num_center_orig[j]))\n", "\n", " all_scores_local = num_scores_local + cat_scores_local\n", " all_scores_local = sorted(all_scores_local, key=lambda x: x[1], reverse=True)\n", "\n", " impact = support_share * abs(beta)\n", " impact_scores.append((k, impact))\n", "\n", " row = {\n", " \"prototype_id\": k,\n", " \"support_n\": support_n,\n", " \"support_share_train\": support_share,\n", " \"prototype_output_weight\": beta,\n", " \"impact_score\": impact,\n", " \"assigned_mean_target_train\": assigned_mean_target,\n", " }\n", "\n", " # Добавляем несколько интерпретируемых признаков\n", " for col in numeric_features[:4]:\n", " row[f\"num__{col}\"] = numeric_values.get(col, np.nan)\n", " for col in categorical_features[:3]:\n", " row[f\"cat__{col}\"] = dominant_cats.get(col, None)\n", "\n", " prototype_rows.append(row)\n", "\n", " for rank, (fname, score, value) in enumerate(all_scores_local[:5], start=1):\n", " prototype_feature_rows.append({\n", " \"prototype_id\": k,\n", " \"rank\": rank,\n", " \"feature\": fname,\n", " \"contribution_score\": score,\n", " \"interpreted_value\": value,\n", " })\n", "\n", "prototype_summary_df = pd.DataFrame(prototype_rows).sort_values([\"impact_score\", \"support_share_train\"], ascending=[False, False]).reset_index(drop=True)\n", "prototype_top_features_df = pd.DataFrame(prototype_feature_rows).sort_values([\"prototype_id\", \"rank\"]).reset_index(drop=True)\n", "\n", "print(\"Основные прототипы (по impact_score = support_share × |вес выхода|):\")\n", "display(prototype_summary_df)\n", "\n", "print(\"Топ-признаки внутри каждого прототипа:\")\n", "display(prototype_top_features_df)\n", "\n", "plt.figure(figsize=(8, 4))\n", "plt.bar(prototype_summary_df[\"prototype_id\"].astype(str), prototype_summary_df[\"impact_score\"])\n", "plt.title(\"Сила прототипов\")\n", "plt.xlabel(\"prototype_id\")\n", "plt.ylabel(\"impact_score\")\n", "plt.tight_layout()\n", "plt.show()" ] }, { "cell_type": "code", "execution_count": 9, "id": "1917cb95", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "=== КЛЮЧЕВАЯ ИНФОРМАЦИЯ ДЛЯ ВЫВОДА ===\n", "Зависимая переменная: price\n", "\n", "Число исходных признаков: 5\n", "Число признаков после кодирования: 7\n", "Лучшая конфигурация прототипной сети:\n", "n_prototypes = 8\n", "alpha = 1.0\n", "sigma = 1.4106722300074301\n", "\n", "Метрики модели:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
splitr2rmsemae
0train0.51017053409.39694342622.246003
1validation0.51279253050.56771242496.273011
2test0.52106152431.73511541934.625581
\n", "
" ], "text/plain": [ " split r2 rmse mae\n", "0 train 0.510170 53409.396943 42622.246003\n", "1 validation 0.512792 53050.567712 42496.273011\n", "2 test 0.521061 52431.735115 41934.625581" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Топ-10 конфигураций по validation RMSE:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
n_prototypesalphatrain_r2train_rmsetrain_maeval_r2val_rmseval_maesigma
081.00000.51017053409.39694342622.2460030.51279253050.56771242496.2730111.410672
180.10000.51017053409.38296142622.4097680.51278853050.78533742496.6367161.410672
280.01000.51017053409.38282242622.4264380.51278853050.80839742496.6741211.410672
380.00100.51017053409.38282042622.4281050.51278853050.81071642496.6778611.410672
480.00010.51017053409.38282042622.4282720.51278853050.81094842496.6782361.410672
5101.00000.50877453485.43996542675.4523220.51182253103.35501242523.5839381.364013
6100.10000.50877553485.42488742675.6365450.51181853103.55350842523.8244041.364013
7100.01000.50877553485.42473642675.6557970.51181853103.57474342523.8485071.364013
8100.00100.50877553485.42473442675.6577230.51181853103.57688042523.8509181.364013
9100.00010.50877553485.42473442675.6579150.51181853103.57709442523.8511591.364013
\n", "
" ], "text/plain": [ " n_prototypes alpha train_r2 train_rmse train_mae val_r2 val_rmse val_mae sigma\n", "0 8 1.0000 0.510170 53409.396943 42622.246003 0.512792 53050.567712 42496.273011 1.410672\n", "1 8 0.1000 0.510170 53409.382961 42622.409768 0.512788 53050.785337 42496.636716 1.410672\n", "2 8 0.0100 0.510170 53409.382822 42622.426438 0.512788 53050.808397 42496.674121 1.410672\n", "3 8 0.0010 0.510170 53409.382820 42622.428105 0.512788 53050.810716 42496.677861 1.410672\n", "4 8 0.0001 0.510170 53409.382820 42622.428272 0.512788 53050.810948 42496.678236 1.410672\n", "5 10 1.0000 0.508774 53485.439965 42675.452322 0.511822 53103.355012 42523.583938 1.364013\n", "6 10 0.1000 0.508775 53485.424887 42675.636545 0.511818 53103.553508 42523.824404 1.364013\n", "7 10 0.0100 0.508775 53485.424736 42675.655797 0.511818 53103.574743 42523.848507 1.364013\n", "8 10 0.0010 0.508775 53485.424734 42675.657723 0.511818 53103.576880 42523.850918 1.364013\n", "9 10 0.0001 0.508775 53485.424734 42675.657915 0.511818 53103.577094 42523.851159 1.364013" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Основные прототипы:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
prototype_idsupport_nsupport_share_trainprototype_output_weightimpact_scoreassigned_mean_target_trainnum__squarefeetnum__bedroomsnum__bathroomsnum__yearbuiltcat__neighborhood
0738440.12813395506.68714812237.590180272746.4398622424.0822494.4167102.6290991968.388860Urban
1037610.125367-96853.93459912142.254934175264.4224651541.3573332.5794372.6381511970.338735Suburb
2137340.12446796690.02303512034.684867272552.6606432460.8134874.4187851.3738292001.336098Suburb
3436170.120567-96452.14742011628.913907175220.7634011580.5098152.5886091.3596902002.136577Rural
4638630.128767-88722.80544011424.539914181692.4351291539.1755114.3831221.3453271970.988092Rural
5236960.12320086261.62622610627.432351267077.4016582469.3337842.5945362.6402492000.822288Suburb
6337740.12580076093.8678289572.608573261917.3456212444.7826662.5981981.3654921968.891333Urban
7537110.123700-71996.6241078905.982402191311.2102461594.2212034.4073372.6633402002.108713Suburb
\n", "
" ], "text/plain": [ " prototype_id support_n support_share_train prototype_output_weight impact_score assigned_mean_target_train num__squarefeet num__bedrooms num__bathrooms num__yearbuilt cat__neighborhood\n", "0 7 3844 0.128133 95506.687148 12237.590180 272746.439862 2424.082249 4.416710 2.629099 1968.388860 Urban\n", "1 0 3761 0.125367 -96853.934599 12142.254934 175264.422465 1541.357333 2.579437 2.638151 1970.338735 Suburb\n", "2 1 3734 0.124467 96690.023035 12034.684867 272552.660643 2460.813487 4.418785 1.373829 2001.336098 Suburb\n", "3 4 3617 0.120567 -96452.147420 11628.913907 175220.763401 1580.509815 2.588609 1.359690 2002.136577 Rural\n", "4 6 3863 0.128767 -88722.805440 11424.539914 181692.435129 1539.175511 4.383122 1.345327 1970.988092 Rural\n", "5 2 3696 0.123200 86261.626226 10627.432351 267077.401658 2469.333784 2.594536 2.640249 2000.822288 Suburb\n", "6 3 3774 0.125800 76093.867828 9572.608573 261917.345621 2444.782666 2.598198 1.365492 1968.891333 Urban\n", "7 5 3711 0.123700 -71996.624107 8905.982402 191311.210246 1594.221203 4.407337 2.663340 2002.108713 Suburb" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Топ-5 вкладов признаков для основных прототипов:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
prototype_idrankfeaturecontribution_scoreinterpreted_value
001bedrooms80619.0419422.579437
102squarefeet78686.5089831541.357333
203bathrooms75406.2206922.638151
304yearbuilt70286.4702651970.338735
405neighborhood=Suburb467.942083Suburb
511bedrooms79061.9729034.418785
612squarefeet76262.2913512460.813487
713yearbuilt74551.6414712001.336098
814bathrooms74381.3275761.373829
915neighborhood=Suburb471.666369Suburb
2041bedrooms79490.9339952.588609
2142yearbuilt78096.2772682002.136577
2243bathrooms75867.8668591.35969
2344squarefeet71783.9127041580.509815
2445neighborhood=Rural683.868925Rural
3061squarefeet72417.6787931539.175511
3162bathrooms71348.1451381.345327
3263bedrooms69708.7795204.383122
3364yearbuilt61603.8667741970.988092
3465neighborhood=Rural1344.291752Rural
3571yearbuilt78300.8610041968.38886
3672bedrooms77916.5883834.41671
3773bathrooms73298.9869022.629099
3874squarefeet69219.9402122424.082249
3975neighborhood=Urban1426.920949Urban
\n", "
" ], "text/plain": [ " prototype_id rank feature contribution_score interpreted_value\n", "0 0 1 bedrooms 80619.041942 2.579437\n", "1 0 2 squarefeet 78686.508983 1541.357333\n", "2 0 3 bathrooms 75406.220692 2.638151\n", "3 0 4 yearbuilt 70286.470265 1970.338735\n", "4 0 5 neighborhood=Suburb 467.942083 Suburb\n", "5 1 1 bedrooms 79061.972903 4.418785\n", "6 1 2 squarefeet 76262.291351 2460.813487\n", "7 1 3 yearbuilt 74551.641471 2001.336098\n", "8 1 4 bathrooms 74381.327576 1.373829\n", "9 1 5 neighborhood=Suburb 471.666369 Suburb\n", "20 4 1 bedrooms 79490.933995 2.588609\n", "21 4 2 yearbuilt 78096.277268 2002.136577\n", "22 4 3 bathrooms 75867.866859 1.35969\n", "23 4 4 squarefeet 71783.912704 1580.509815\n", "24 4 5 neighborhood=Rural 683.868925 Rural\n", "30 6 1 squarefeet 72417.678793 1539.175511\n", "31 6 2 bathrooms 71348.145138 1.345327\n", "32 6 3 bedrooms 69708.779520 4.383122\n", "33 6 4 yearbuilt 61603.866774 1970.988092\n", "34 6 5 neighborhood=Rural 1344.291752 Rural\n", "35 7 1 yearbuilt 78300.861004 1968.38886\n", "36 7 2 bedrooms 77916.588383 4.41671\n", "37 7 3 bathrooms 73298.986902 2.629099\n", "38 7 4 squarefeet 69219.940212 2424.082249\n", "39 7 5 neighborhood=Urban 1426.920949 Urban" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Первые 10 фактических и предсказанных значений на тесте:\n" ] }, { "data": { "text/html": [ "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
y_truey_pred
0170835.035713208189.809824
1126913.469998187521.135570
2246611.883092269935.842537
3244250.462969270102.333733
4271127.650112265637.877925
5189289.951199189107.502703
6343510.576500276873.696031
7203461.591080185461.893451
8285903.247682278803.162808
9224868.958375224449.420520
\n", "
" ], "text/plain": [ " y_true y_pred\n", "0 170835.035713 208189.809824\n", "1 126913.469998 187521.135570\n", "2 246611.883092 269935.842537\n", "3 244250.462969 270102.333733\n", "4 271127.650112 265637.877925\n", "5 189289.951199 189107.502703\n", "6 343510.576500 276873.696031\n", "7 203461.591080 185461.893451\n", "8 285903.247682 278803.162808\n", "9 224868.958375 224449.420520" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "print(\"=== КЛЮЧЕВАЯ ИНФОРМАЦИЯ ДЛЯ ВЫВОДА ===\")\n", "print(\"Зависимая переменная:\", target_col)\n", "print()\n", "print(\"Число исходных признаков:\", X.shape[1])\n", "print(\"Число признаков после кодирования:\", X_train_proc.shape[1])\n", "print(\"Лучшая конфигурация прототипной сети:\")\n", "print(\"n_prototypes =\", best_n_prototypes)\n", "print(\"alpha =\", best_alpha)\n", "print(\"sigma =\", best_model.sigma_)\n", "print()\n", "\n", "print(\"Метрики модели:\")\n", "display(metrics_df)\n", "\n", "print(\"Топ-10 конфигураций по validation RMSE:\")\n", "display(search_df.head(10))\n", "\n", "print(\"Основные прототипы:\")\n", "display(prototype_summary_df.head(10))\n", "\n", "print(\"Топ-5 вкладов признаков для основных прототипов:\")\n", "top_proto_ids = prototype_summary_df[\"prototype_id\"].head(5).tolist()\n", "display(prototype_top_features_df[prototype_top_features_df[\"prototype_id\"].isin(top_proto_ids)].copy())\n", "\n", "print(\"Первые 10 фактических и предсказанных значений на тесте:\")\n", "display(comparison_df.head(10))" ] }, { "cell_type": "markdown", "id": "a6504550", "metadata": {}, "source": [ "Итог\n", "\n", "Зависимой переменной выбрана price. Лучшая прототипная сеть использует 8 прототипов (alpha = 1.0, sigma ≈ 1.41) и показывает на тесте R² = 0.521, RMSE ≈ 52432, MAE ≈ 41935.\n", "\n", "Основные прототипы соответствуют типичным сегментам объектов: более дорогие дома с большой площадью, большим числом комнат и расположением в Urban/Suburb, и менее дорогие объекты с меньшей площадью и Rural/Suburb-локацией.\n", "\n", "Наибольший вклад в прототипах дают признаки squarefeet, bedrooms, bathrooms, yearbuilt; категориальный признак neighborhood также участвует, но его вклад заметно ниже.\n", "\n", "Следовательно, прототипная сеть выделяет интерпретируемые группы жилья, а ключевыми факторами цены являются площадь, число спален/ванных, год постройки и тип района." ] } ], "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "name": "python" } }, "nbformat": 4, "nbformat_minor": 5 }