12_embed_cust_hier.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "8816b366-5be4-4748-b769-4dc6f7d1735f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Modify the file 00_setup.py to define input/output file paths on your system\n",
    "# The information in 00_setup.py will be used across notebooks\n",
    "from importlib.machinery import SourceFileLoader\n",
    "setup = SourceFileLoader(\"setup\", \"./00_setup.py\").load_module()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a945de0c-f042-46b2-9d06-ab064af6c05b",
   "metadata": {},
   "source": [
    "## Neural Network With NAICS Embedding Layer, Custom Generator Inject 10% Unseen, NAICS hierarchy codes\n",
    "Here, instead of randomizing the input training data, I use a custom data generator to inject 10% random cases for each batch.  This is to help the model learn to handle unseen codes"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9e6fba8f-d564-466c-8e50-7773b095157f",
   "metadata": {},
   "source": [
    "Encode the base NAICS and higher hierarchy levels.  This script should be compared to 04 and 06."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "32221be3-556a-441d-a30c-1f22f5dd664f",
   "metadata": {},
   "source": [
    "Custom data generators are used to retrieve data at each batch or epoch, and can be used to modify data for every batch or epoch.  In future scripts, I modify at the batch level, injecting random \"1\" values.   See also  https://www.tensorflow.org/versions/r2.1/api_docs/python/tf/keras/utils/Sequence"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e169c2e8-ba0f-488f-acf6-87d419bd4e99",
   "metadata": {},
   "source": [
    "Use the dataset with no missing values. "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d47a7ec9-7d7b-47df-a902-b2f1e54f1135",
   "metadata": {},
   "source": [
    "*This script takes about 2 hours on my MacBook Air*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "fb42f9b4-99a8-4753-aacf-8722469be4c8",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "bf51170a-4487-4d10-b63d-c88f41b529f9",
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "import importlib, pickle\n",
    "from sklearn import model_selection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "09dd973b-60a0-434b-9193-8c5fcec27dd5",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import os\n",
    "import re\n",
    "\n",
    "import keras\n",
    "from tensorflow.keras import layers, optimizers, losses, metrics, Model\n",
    "from sklearn import preprocessing, feature_extraction, model_selection\n",
    "from IPython.display import display, HTML\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "9f1e4f2e-0e45-4e28-8e9c-212e0bb51199",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sba_nn.sba_nn import sbnn_metrics, sbnn_model\n",
    "from sba_nn.sba_nn.sbnn_model import CatInjectGenerator"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3cbc00d2-bc83-42d9-83eb-65fcf306dccd",
   "metadata": {},
   "source": [
    "## Input Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "e014cd31-0f8d-42e3-82f9-b529eec8a3cc",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "sba_loans = pd.read_parquet(Path(setup.temp_path).joinpath('01_DATA_transformed_nomiss.parquet'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "8e65cb7c-792c-46ff-a530-0d5c64240d60",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(Path(setup.temp_path).joinpath('01_DATA_features.pkl'), 'rb') as fin:\n",
    "    imputer_features = pickle.load(fin)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "6f0ce28e-a80c-415b-b068-5c4d49370da8",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(Path(setup.temp_path).joinpath('01_DATA_naics_max_encodings.pkl'), 'rb') as fin:\n",
    "    naics_max_levels= pickle.load(fin)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d4116da-d689-478a-9c5d-a330f231e615",
   "metadata": {},
   "source": [
    "##### Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "f98c6e41-38d0-4d5f-b821-e3837b8c2068",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['NoEmp', 'CreateJob', 'LowDoc', 'DisbursementGross', 'new_business', 'urban_flag', 'franchise_flag', 'missingindicator_LowDoc', 'missingindicator_new_business', 'missingindicator_urban_flag']\n"
     ]
    }
   ],
   "source": [
    "# Numeric features\n",
    "features_numeric = [f for f in imputer_features if 'NAICS' not in f]\n",
    "features_numeric_len = len(features_numeric)\n",
    "print(features_numeric)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "a74fa386-4011-45ee-bdb5-d671db9ffa91",
   "metadata": {},
   "outputs": [],
   "source": [
    "# NAICS feature - just bas\n",
    "features_naics = ['NAICS', 'NAICS_4', 'NAICS_3', 'NAICS_sector']\n",
    "features_naics_max_levels  = [naics_max_levels[n] for n in features_naics]\n",
    "features_naics_emb_dim = [setup.nn_naics_embed_size_dict[n] for n in features_naics]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "867031ee-8b6e-424d-b497-8e19409ba1d9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1170, 345, 107, 21]\n",
      "[8, 4, 4, 2]\n"
     ]
    }
   ],
   "source": [
    "print(features_naics_max_levels)\n",
    "print(features_naics_emb_dim)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b979396c-c822-4940-a292-6e07f1b54f58",
   "metadata": {},
   "source": [
    "##### Datasets for train, validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "43bca011-a59d-4496-9f03-87f4e350a253",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "X = sba_loans[['dset', 'LoanNr_ChkDgt'] + features_numeric + features_naics].set_index('LoanNr_ChkDgt').sort_index()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "aa162d01-ed73-4ae3-832e-4aeea09b70f4",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "training X: (446848, 14), y:(446848,)\n"
     ]
    }
   ],
   "source": [
    "X_train = X[X['dset'] == 'train'].drop(columns='dset')\n",
    "y_train = sba_loans[sba_loans['dset'] == 'train'].set_index('LoanNr_ChkDgt').sort_index()['target']\n",
    "print(f'training X: {X_train.shape}, y:{y_train.shape}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "b32b5fd3-6f6e-4aec-8e20-0e2b71a4814c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "val X: (95604, 14), y:(95604,)\n"
     ]
    }
   ],
   "source": [
    "X_val = X[X['dset'] == 'val'].drop(columns='dset')\n",
    "y_val = sba_loans[sba_loans['dset'] == 'val'].set_index('LoanNr_ChkDgt').sort_index()['target']\n",
    "print(f'val X: {X_val.shape}, y:{y_val.shape}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "01b8ff96-159b-4283-b5fb-c4b5dbd277d5",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.20230592953308507\n"
     ]
    }
   ],
   "source": [
    "base_thresh = y_train.mean()\n",
    "print(base_thresh)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3db4ea1c-77b2-4721-b106-1e66fc502139",
   "metadata": {},
   "source": [
    "## Create, fit model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "4d48ddfa-580c-4321-9917-0bd176b67d87",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-05-18 21:24:34.138399: I metal_plugin/src/device/metal_device.cc:1154] Metal device set to: Apple M1\n",
      "2024-05-18 21:24:34.138429: I metal_plugin/src/device/metal_device.cc:296] systemMemory: 16.00 GB\n",
      "2024-05-18 21:24:34.138438: I metal_plugin/src/device/metal_device.cc:313] maxCacheSize: 5.33 GB\n",
      "2024-05-18 21:24:34.138494: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:303] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.\n",
      "2024-05-18 21:24:34.138523: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:269] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)\n"
     ]
    }
   ],
   "source": [
    "this_model = sbnn_model.create_emb_model(features_numeric_len,\n",
    "                                         features_naics_max_levels,\n",
    "                                         features_naics_emb_dim,\n",
    "                                         naics_embedding_names = features_naics,\n",
    "                                         hidden_size = setup.nn_layer_sizes,\n",
    "                                         activation='tanh',\n",
    "                                         lr=setup.nn_learning_rate,\n",
    "                                         opt_func = setup.nn_optimizer,\n",
    "                                         dropout = setup.nn_dropout)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "47036621-ef13-4957-997b-f8ef86016f7f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "__________________________________________________________________________________________________\n",
      " Layer (type)                Output Shape                 Param #   Connected to                  \n",
      "==================================================================================================\n",
      " input_naics_00 (InputLayer  [(None, 1)]                  0         []                            \n",
      " )                                                                                                \n",
      "                                                                                                  \n",
      " input_naics_01 (InputLayer  [(None, 1)]                  0         []                            \n",
      " )                                                                                                \n",
      "                                                                                                  \n",
      " input_naics_02 (InputLayer  [(None, 1)]                  0         []                            \n",
      " )                                                                                                \n",
      "                                                                                                  \n",
      " input_naics_03 (InputLayer  [(None, 1)]                  0         []                            \n",
      " )                                                                                                \n",
      "                                                                                                  \n",
      " emb_naics_00 (Embedding)    (None, 1, 8)                 9360      ['input_naics_00[0][0]']      \n",
      "                                                                                                  \n",
      " emb_naics_01 (Embedding)    (None, 1, 4)                 1380      ['input_naics_01[0][0]']      \n",
      "                                                                                                  \n",
      " emb_naics_02 (Embedding)    (None, 1, 4)                 428       ['input_naics_02[0][0]']      \n",
      "                                                                                                  \n",
      " emb_naics_03 (Embedding)    (None, 1, 2)                 42        ['input_naics_03[0][0]']      \n",
      "                                                                                                  \n",
      " NAICS (Flatten)             (None, 8)                    0         ['emb_naics_00[0][0]']        \n",
      "                                                                                                  \n",
      " NAICS_4 (Flatten)           (None, 4)                    0         ['emb_naics_01[0][0]']        \n",
      "                                                                                                  \n",
      " NAICS_3 (Flatten)           (None, 4)                    0         ['emb_naics_02[0][0]']        \n",
      "                                                                                                  \n",
      " NAICS_sector (Flatten)      (None, 2)                    0         ['emb_naics_03[0][0]']        \n",
      "                                                                                                  \n",
      " input_numeric (InputLayer)  [(None, 10)]                 0         []                            \n",
      "                                                                                                  \n",
      " input_concat (Concatenate)  (None, 28)                   0         ['NAICS[0][0]',               \n",
      "                                                                     'NAICS_4[0][0]',             \n",
      "                                                                     'NAICS_3[0][0]',             \n",
      "                                                                     'NAICS_sector[0][0]',        \n",
      "                                                                     'input_numeric[0][0]']       \n",
      "                                                                                                  \n",
      " layer_00 (Dense)            (None, 128)                  3712      ['input_concat[0][0]']        \n",
      "                                                                                                  \n",
      " dropout_00 (Dropout)        (None, 128)                  0         ['layer_00[0][0]']            \n",
      "                                                                                                  \n",
      " layer_01 (Dense)            (None, 64)                   8256      ['dropout_00[0][0]']          \n",
      "                                                                                                  \n",
      " dropout_01 (Dropout)        (None, 64)                   0         ['layer_01[0][0]']            \n",
      "                                                                                                  \n",
      " output (Dense)              (None, 1)                    65        ['dropout_01[0][0]']          \n",
      "                                                                                                  \n",
      "==================================================================================================\n",
      "Total params: 23243 (90.79 KB)\n",
      "Trainable params: 23243 (90.79 KB)\n",
      "Non-trainable params: 0 (0.00 Byte)\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "this_model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4ed7408d-99ce-413e-aef8-23743b5601cd",
   "metadata": {},
   "source": [
    "##### Use data generator to inject 1 values into training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "f1dfde86-af9b-4110-aae9-30f3d59404af",
   "metadata": {},
   "outputs": [],
   "source": [
    "generator = CatInjectGenerator(X_train, y_train,\n",
    "                               categorical_columns=features_naics,\n",
    "                               batch_size = setup.nn_batch_size,\n",
    "                               injection_rate = 0.1,\n",
    "                               injection_value = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "0a74a486-8426-4117-821b-5db3d92326f6",
   "metadata": {},
   "outputs": [],
   "source": [
    "callback = keras.callbacks.EarlyStopping(patience=3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "fd9ebbdb-daae-45c7-bbe2-751f8c245c31",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-05-18 21:24:35.994766: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "13964/13964 [==============================] - ETA: 0s - loss: 0.4630 - auc: 0.3357 - auc_1: 0.6986"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-05-18 21:31:01.788571: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "13964/13964 [==============================] - 425s 30ms/step - loss: 0.4630 - auc: 0.3357 - auc_1: 0.6986 - val_loss: 0.4538 - val_auc: 0.3721 - val_auc_1: 0.7232\n",
      "Epoch 2/20\n",
      "13964/13964 [==============================] - 379s 27ms/step - loss: 0.4549 - auc: 0.3588 - auc_1: 0.7155 - val_loss: 0.4540 - val_auc: 0.3749 - val_auc_1: 0.7238\n",
      "Epoch 3/20\n",
      "13964/13964 [==============================] - 392s 28ms/step - loss: 0.4529 - auc: 0.3643 - auc_1: 0.7200 - val_loss: 0.4530 - val_auc: 0.3763 - val_auc_1: 0.7258\n",
      "Epoch 4/20\n",
      "13964/13964 [==============================] - 392s 28ms/step - loss: 0.4516 - auc: 0.3691 - auc_1: 0.7226 - val_loss: 0.4513 - val_auc: 0.3816 - val_auc_1: 0.7283\n",
      "Epoch 5/20\n",
      "13964/13964 [==============================] - 391s 28ms/step - loss: 0.4509 - auc: 0.3717 - auc_1: 0.7242 - val_loss: 0.4496 - val_auc: 0.3845 - val_auc_1: 0.7304\n",
      "Epoch 6/20\n",
      "13964/13964 [==============================] - 402s 29ms/step - loss: 0.4501 - auc: 0.3745 - auc_1: 0.7257 - val_loss: 0.4512 - val_auc: 0.3856 - val_auc_1: 0.7309\n",
      "Epoch 7/20\n",
      "13964/13964 [==============================] - 402s 29ms/step - loss: 0.4497 - auc: 0.3754 - auc_1: 0.7263 - val_loss: 0.4523 - val_auc: 0.3850 - val_auc_1: 0.7309\n",
      "Epoch 8/20\n",
      "13964/13964 [==============================] - 430s 31ms/step - loss: 0.4493 - auc: 0.3762 - auc_1: 0.7274 - val_loss: 0.4497 - val_auc: 0.3858 - val_auc_1: 0.7304\n"
     ]
    }
   ],
   "source": [
    "this_history = this_model.fit(generator,\n",
    "                              batch_size=setup.nn_batch_size,\n",
    "                              epochs = setup.nn_epochs,\n",
    "                              validation_data=([X_val[features_numeric]] + [X_val[f] for f in features_naics], \n",
    "                                               y_val),\n",
    "                              shuffle=True,\n",
    "                              callbacks=[callback])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "460c4b43-c673-4ee1-b070-d2d596ac4561",
   "metadata": {},
   "outputs": [],
   "source": [
    "this_history_df = sbnn_model.process_history(this_history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "ad4783f2-8d0b-4850-a45a-3a4be8e63318",
   "metadata": {},
   "outputs": [],
   "source": [
    "this_history_df.to_csv(Path(setup.temp_path).joinpath('12_REPORT_fit_history.csv'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "73e89efd-75a9-4b27-81a8-13e12d5482be",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "this_history_df[['loss', 'val_loss']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "2342a31f-4891-4834-851a-b90b8b729c3c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "this_history_df[['auc', 'val_auc']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "cee80637-cabf-4974-b476-08c5fba165ff",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "this_history_df[['auc_roc', 'val_auc_roc']].plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "e184b0ca-1d31-4798-88a0-4e590b67062e",
   "metadata": {},
   "outputs": [],
   "source": [
    "this_model.save(Path(setup.temp_path).joinpath('12_DATA_model.keras'),save_format='tf')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d326c9ea-4759-4418-a64d-56c151049c25",
   "metadata": {},
   "source": [
    "## Predictions on all data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "cd6910ee-5ffe-4def-9aa8-7dba76446b08",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "   12/21735 [..............................] - ETA: 1:47   "
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-05-18 22:18:09.468774: I tensorflow/core/grappler/optimizers/custom_graph_optimizer_registry.cc:114] Plugin optimizer for device_type GPU is enabled.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "21735/21735 [==============================] - 106s 5ms/step\n"
     ]
    }
   ],
   "source": [
    "all_predictions = this_model.predict([X[features_numeric]]+ [X[f] for f in features_naics])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "98a2cce1-b7d3-4ebc-8869-80705297e145",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_predictions_df = pd.DataFrame(all_predictions, index=X.index) \\\n",
    "    .set_axis(['predict_prob'], axis=1) \\\n",
    "    .reset_index() \\\n",
    "    .merge(sba_loans[['target', 'LoanNr_ChkDgt', 'dset', 'dset_naics_holdout', 'NAICS']], \n",
    "           on='LoanNr_ChkDgt')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "3fc7da9c-6bd2-4d1b-b77f-b62679825d5d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>predict_prob</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>predict_prob</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.325987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>target</th>\n",
       "      <td>0.325987</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              predict_prob    target\n",
       "predict_prob      1.000000  0.325987\n",
       "target            0.325987  1.000000"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_predictions_df[['predict_prob', 'target']].corr(method='spearman')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e79afc4e-cb67-4064-9b19-dc52500baef2",
   "metadata": {},
   "source": [
    "##### Threshold Tune & Binary Predictions\n",
    "Using training probability predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "ccd0bc23-fbc0-40d0-8d8b-f2302405a875",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_pred_train = all_predictions_df[all_predictions_df['dset'] == 'train']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "df36d9e9-f35e-46b3-a5be-08081f998319",
   "metadata": {},
   "outputs": [],
   "source": [
    "thresh_tune_data = sbnn_metrics.get_f1_frame(all_pred_train['target'], \n",
    "                                        all_pred_train['predict_prob'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "de634156-caf6-4a52-b5d7-44b8e986aa1d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>thresh</th>\n",
       "      <th>f1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>0.22</td>\n",
       "      <td>0.453893</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>0.24</td>\n",
       "      <td>0.452835</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.20</td>\n",
       "      <td>0.452030</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    thresh        f1\n",
       "11    0.22  0.453893\n",
       "12    0.24  0.452835\n",
       "10    0.20  0.452030"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "thresh_tune_data.sort_values('f1', ascending=False, inplace=True)\n",
    "thresh_tune_data.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "c746b880-d781-4bc1-9df2-89211127c22a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.22"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_thresh = thresh_tune_data['thresh'].iloc[0]\n",
    "best_thresh"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3c0ac5e2-be0d-42f0-960d-1f81b4a5d4e2",
   "metadata": {},
   "source": [
    "##### Append binary predictions to probability predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "641da72e-1bca-4fe7-96c4-6c029b521bd7",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_predictions_df['predict_bin'] = sbnn_metrics.get_binary_predictions(all_predictions_df['predict_prob'], best_thresh)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "9f6ecdf4-ac01-4c39-8315-e8583daa337e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "predict_bin\n",
       "0    0.584037\n",
       "1    0.415963\n",
       "Name: proportion, dtype: float64"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_predictions_df['predict_bin'].value_counts(normalize=True, dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "cbfed830-7ff8-40f6-a691-6b894b26d6bc",
   "metadata": {},
   "outputs": [],
   "source": [
    "all_predictions_df.to_parquet(Path(setup.temp_path).joinpath('12_DATA_predictions.parquet'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb0ecbd2-abff-46d8-8c70-f7bccd363b69",
   "metadata": {},
   "source": [
    "## Metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "be4ceefe-c7a8-45bd-8573-2daefe99a1c2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>dset</th>\n",
       "      <th>accuracy_score</th>\n",
       "      <th>f1_score</th>\n",
       "      <th>precision_score</th>\n",
       "      <th>recall_score</th>\n",
       "      <th>average_precision_score</th>\n",
       "      <th>roc_auc_score</th>\n",
       "      <th>dset_naics_holdout</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>test</td>\n",
       "      <td>0.661668</td>\n",
       "      <td>0.450652</td>\n",
       "      <td>0.335069</td>\n",
       "      <td>0.687970</td>\n",
       "      <td>0.380998</td>\n",
       "      <td>0.731598</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>train</td>\n",
       "      <td>0.662030</td>\n",
       "      <td>0.453893</td>\n",
       "      <td>0.337164</td>\n",
       "      <td>0.694248</td>\n",
       "      <td>0.392419</td>\n",
       "      <td>0.735919</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>val</td>\n",
       "      <td>0.660255</td>\n",
       "      <td>0.451899</td>\n",
       "      <td>0.336652</td>\n",
       "      <td>0.687125</td>\n",
       "      <td>0.385774</td>\n",
       "      <td>0.730476</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>test</td>\n",
       "      <td>0.657880</td>\n",
       "      <td>0.448441</td>\n",
       "      <td>0.333442</td>\n",
       "      <td>0.684522</td>\n",
       "      <td>0.379818</td>\n",
       "      <td>0.726954</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>test</td>\n",
       "      <td>0.667965</td>\n",
       "      <td>0.454397</td>\n",
       "      <td>0.337824</td>\n",
       "      <td>0.693812</td>\n",
       "      <td>0.385357</td>\n",
       "      <td>0.739736</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    dset  accuracy_score  f1_score  precision_score  recall_score  \\\n",
       "0   test        0.661668  0.450652         0.335069      0.687970   \n",
       "1  train        0.662030  0.453893         0.337164      0.694248   \n",
       "2    val        0.660255  0.451899         0.336652      0.687125   \n",
       "0   test        0.657880  0.448441         0.333442      0.684522   \n",
       "1   test        0.667965  0.454397         0.337824      0.693812   \n",
       "\n",
       "   average_precision_score  roc_auc_score  dset_naics_holdout  \n",
       "0                 0.380998       0.731598                 NaN  \n",
       "1                 0.392419       0.735919                 NaN  \n",
       "2                 0.385774       0.730476                 NaN  \n",
       "0                 0.379818       0.726954                 0.0  \n",
       "1                 0.385357       0.739736                 1.0  "
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics_dset_df = all_predictions_df.groupby('dset') \\\n",
    "    .apply(lambda x: sbnn_metrics.dset_metrics(x.target, x.predict_bin, x.predict_prob)) \\\n",
    "    .reset_index()\n",
    "metrics_test_df = all_predictions_df[all_predictions_df['dset'] == 'test'] \\\n",
    "    .groupby(['dset', 'dset_naics_holdout']) \\\n",
    "    .apply(lambda x: sbnn_metrics.dset_metrics(x.target, x.predict_bin, x.predict_prob)) \\\n",
    "    .reset_index()\n",
    "metrics_df = pd.concat([metrics_dset_df, metrics_test_df])\n",
    "metrics_df.to_csv(Path(setup.temp_path).joinpath('12_REPORT_metrics.csv'), index=True)\n",
    "metrics_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "52e382b5-e9dc-47a1-9bf4-6d0b6b4a0c61",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8368677a-f21c-49bc-811b-a994008e9959",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tf_p39",
   "language": "python",
   "name": "tf_p39"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}