AFML 4.2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sample Weights\n",
    "\n",
    "This notebook will cover exercise answer.\n",
    "\n",
    "* Exercise 4.5\n",
    "\n",
    "As we go along, there will be some explanations.\n",
    "\n",
    "More importantly, this method can be applied not just within mean-reversion strategy but also other strategies as well. \n",
    "\n",
    "Most of the functions below can be found under research/Sampling.\n",
    "\n",
    "Contact: boyboi86@gmail.com"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Num of CPU core:  4\n",
      "Machine info:  Windows-10-10.0.18362-SP0\n",
      "Python 3.7.4 (default, Aug  9 2019, 18:34:13) [MSC v.1915 64 bit (AMD64)]\n",
      "Numpy 1.17.3\n",
      "Pandas 1.0.3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Wei_X\\Anaconda3\\lib\\site-packages\\statsmodels\\tools\\_testing.py:19: FutureWarning: pandas.util.testing is deprecated. Use the functions in the public API at pandas.testing instead.\n",
      "  import pandas.util.testing as tm\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 1500x800 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import research as rs\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Note**\n",
    "\n",
    "Instead of trend strategy, we will continue to use mean-reversion strategy.\n",
    "\n",
    "This part will deviate slightly from the exercise, but I assure you the outcome is somewhat similiar.\n",
    "\n",
    "However using trend strategy, the idea which Dr Marco Lopez wish to express would be obvious since meta-labels would suffer from imbalance sample."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "dollar = pd.read_csv('./research/Sample_data/dollar_bars.txt', \n",
    "                 sep=',', \n",
    "                 header=0, \n",
    "                 parse_dates = True, \n",
    "                 index_col=['date_time'])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Bollinger Band results:\n",
      "\n",
      "Num of times upper limit touched: 10195\n",
      "Num of times lower limit touched: 8987\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Wei_X\\Desktop\\Python\\research\\Labels\\triple_barrier_method.py:75: UserWarning: Data and events index shape must be same, reindex data to fit events\n",
      "  warnings.warn('Data and events index shape must be same, reindex data to fit events')\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[                                             t1  sl  pt\n",
      "2016-04-14 14:56:40.929 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 15:26:56.280 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 15:35:15.960 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 16:14:10.181 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 16:33:11.080 2016-04-15 17:10:42.500 NaT NaT\n",
      "...                                         ...  ..  ..\n",
      "2016-12-30 20:59:58.848                     NaT NaT NaT\n",
      "2016-12-30 21:00:00.352                     NaT NaT NaT\n",
      "2016-12-30 21:00:24.294                     NaT NaT NaT\n",
      "2016-12-30 21:03:03.027                     NaT NaT NaT\n",
      "2016-12-30 21:13:31.990                     NaT NaT NaT\n",
      "\n",
      "[6043 rows x 3 columns]] this out\n",
      "[                                             t1  sl  pt\n",
      "2016-04-14 14:56:40.929 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 15:26:56.280 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 15:35:15.960 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 16:14:10.181 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 16:33:11.080 2016-04-15 17:10:42.500 NaT NaT\n",
      "...                                         ...  ..  ..\n",
      "2016-12-30 20:59:58.848                     NaT NaT NaT\n",
      "2016-12-30 21:00:00.352                     NaT NaT NaT\n",
      "2016-12-30 21:00:24.294                     NaT NaT NaT\n",
      "2016-12-30 21:03:03.027                     NaT NaT NaT\n",
      "2016-12-30 21:13:31.990                     NaT NaT NaT\n",
      "\n",
      "[6043 rows x 3 columns],                                              t1  sl  pt\n",
      "2015-09-10 12:36:39.559 2015-09-11 12:38:05.265 NaT NaT\n",
      "2015-09-10 13:01:38.292 2015-09-11 13:30:09.634 NaT NaT\n",
      "2015-09-10 13:30:20.359 2015-09-11 13:38:02.648 NaT NaT\n",
      "2015-09-10 13:35:08.110 2015-09-11 13:38:02.648 NaT NaT\n",
      "2015-09-10 13:39:55.114 2015-09-11 13:48:13.785 NaT NaT\n",
      "...                                         ...  ..  ..\n",
      "2016-04-14 14:07:28.549 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:17:07.857 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:25:57.715 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:36:02.258 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:49:13.017 2016-04-15 17:10:42.500 NaT NaT\n",
      "\n",
      "[6043 rows x 3 columns]]"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2020-05-26 15:10:01.079582 33.33% _pt_sl_t1 done after 0.51 minutes. Remaining 1.01 minutes.\r"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " this out\n",
      "[                                             t1  sl  pt\n",
      "2016-04-14 14:56:40.929 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 15:26:56.280 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 15:35:15.960 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 16:14:10.181 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 16:33:11.080 2016-04-15 17:10:42.500 NaT NaT\n",
      "...                                         ...  ..  ..\n",
      "2016-12-30 20:59:58.848                     NaT NaT NaT\n",
      "2016-12-30 21:00:00.352                     NaT NaT NaT\n",
      "2016-12-30 21:00:24.294                     NaT NaT NaT\n",
      "2016-12-30 21:03:03.027                     NaT NaT NaT\n",
      "2016-12-30 21:13:31.990                     NaT NaT NaT\n",
      "\n",
      "[6043 rows x 3 columns],                                              t1  sl  pt\n",
      "2015-09-10 12:36:39.559 2015-09-11 12:38:05.265 NaT NaT\n",
      "2015-09-10 13:01:38.292 2015-09-11 13:30:09.634 NaT NaT\n",
      "2015-09-10 13:30:20.359 2015-09-11 13:38:02.648 NaT NaT\n",
      "2015-09-10 13:35:08.110 2015-09-11 13:38:02.648 NaT NaT\n",
      "2015-09-10 13:39:55.114 2015-09-11 13:48:13.785 NaT NaT\n",
      "...                                         ...  ..  ..\n",
      "2016-04-14 14:07:28.549 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:17:07.857 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:25:57.715 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:36:02.258 2016-04-15 17:10:42.500 NaT NaT\n",
      "2016-04-14 14:49:13.017 2016-04-15 17:10:42.500 NaT NaT\n",
      "\n",
      "[6043 rows x 3 columns],                                              t1                      sl  pt\n",
      "2015-01-05 03:37:42.912 2015-01-06 09:15:06.306 2015-01-05 15:45:52.901 NaT\n",
      "2015-01-05 08:21:37.801 2015-01-06 09:15:06.306 2015-01-05 15:40:45.114 NaT\n",
      "2015-01-05 10:27:41.882 2015-01-06 11:18:19.521 2015-01-05 15:35:09.056 NaT\n",
      "2015-01-05 13:42:46.868 2015-01-06 13:48:17.273                     NaT NaT\n",
      "2015-01-05 14:30:57.874 2015-01-06 14:39:52.557                     NaT NaT\n",
      "...                                         ...                     ...  ..\n",
      "2015-09-10 08:26:20.902 2015-09-11 10:14:00.027                     NaT NaT\n",
      "2015-09-10 09:55:02.979 2015-09-11 10:14:00.027                     NaT NaT\n",
      "2015-09-10 10:27:02.613 2015-09-11 12:38:05.265                     NaT NaT\n",
      "2015-09-10 11:43:14.539 2015-09-11 12:38:05.265                     NaT NaT\n",
      "2015-09-10 12:19:25.229 2015-09-11 12:38:05.265                     NaT NaT\n",
      "\n",
      "[6043 rows x 3 columns]] this out\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2020-05-26 15:10:01.277418 66.67% _pt_sl_t1 done after 0.51 minutes. Remaining 0.25 minutes.\r",
      "2020-05-26 15:10:01.322038 100.0% _pt_sl_t1 done after 0.51 minutes. Remaining 0.0 minutes.\n"
     ]
    }
   ],
   "source": [
    "dollar = rs.bband_as_side(data = dollar, \n",
    "                          window = 50, \n",
    "                          width = 0.001)\n",
    "\n",
    "dollar['volatility'] = rs.vol(dollar['close'], span0 = 50)\n",
    "\n",
    "events = rs.cs_filter(dollar['close'], \n",
    "                    limit = dollar['volatility'].mean())\n",
    "\n",
    "vb = rs.vert_barrier(data = dollar['close'], \n",
    "                 events = events, \n",
    "                 period = 'days', \n",
    "                 freq = 1)\n",
    "\n",
    "tb = rs.tri_barrier(data = dollar['close'], \n",
    "                    events = events, \n",
    "                    trgt = dollar['volatility'] * 5, \n",
    "                    min_req = 0.002, \n",
    "                    num_threads = 3, \n",
    "                    ptSl = [0,2],\n",
    "                    t1 = vb, \n",
    "                    side = dollar['side'])\n",
    "\n",
    "mlabel = rs.meta_label(data = dollar['close'], \n",
    "                       events = tb, \n",
    "                       drop = False) # when you have a side binary, you won't have rare labels usually\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.0    9126\n",
       "1.0    8967\n",
       "Name: bin, dtype: int64"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mlabel['bin'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " Classification Report\n",
      "=======================================================\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "         0.0       0.00      0.00      0.00      9126\n",
      "         1.0       0.50      1.00      0.66      8967\n",
      "\n",
      "    accuracy                           0.50     18093\n",
      "   macro avg       0.25      0.50      0.33     18093\n",
      "weighted avg       0.25      0.50      0.33     18093\n",
      "\n",
      "\n",
      " Confusion Matrix\n",
      "=======================================================\n",
      "[[TN, FP]\n",
      " [FN, TP]]\n",
      "\n",
      "[[   0 9126]\n",
      " [   0 8967]]\n",
      "\n",
      " Accuracy Score\n",
      "=======================================================\n",
      "0.4956060354833361\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Wei_X\\Anaconda3\\lib\\site-packages\\sklearn\\metrics\\_classification.py:1221: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n"
     ]
    }
   ],
   "source": [
    "# original data matrix report w/o random forest\n",
    "\n",
    "rs.report_matrix(actual_data = mlabel, \n",
    "                 prediction_data = None, \n",
    "                 ROC = None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " Classification Report\n",
      "=======================================================\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "         0.0       0.53      0.33      0.41      2687\n",
      "         1.0       0.52      0.71      0.60      2741\n",
      "\n",
      "    accuracy                           0.52      5428\n",
      "   macro avg       0.52      0.52      0.50      5428\n",
      "weighted avg       0.52      0.52      0.50      5428\n",
      "\n",
      "\n",
      " Confusion Matrix\n",
      "=======================================================\n",
      "[[TN, FP]\n",
      " [FN, TP]]\n",
      "\n",
      "[[ 897 1790]\n",
      " [ 804 1937]]\n",
      "\n",
      " Accuracy Score\n",
      "=======================================================\n",
      "0.5221075902726603\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Exercise 4.5a\n",
    "# for convienience sake, we will just go for crossing average & volatility as features\n",
    "\n",
    "X = dollar.drop(['open', 'high', 'low', 'close','cum_vol', 'cum_dollar', 'cum_ticks'], axis = 1)\n",
    "X.dropna(inplace = True) # we lost quite abit of data\n",
    "\n",
    "X = X.reindex(mlabel.index)\n",
    "y = mlabel['bin']\n",
    "\n",
    "n_estimators, max_depth, c_random_state = 500, 7, 42\n",
    "\n",
    "# Random Forest Model\n",
    "rf = RandomForestClassifier(max_depth = max_depth, \n",
    "                            n_estimators = n_estimators,\n",
    "                            criterion = 'entropy',\n",
    "                            #bootstrap=True,\n",
    "                            #max_samples = av_uniqueness_by_coevent['tW'].mean()\n",
    "                            oob_score = False,\n",
    "                            class_weight = None, # We did not use class weight\n",
    "                            random_state = c_random_state)\n",
    "\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, shuffle=False)\n",
    "\n",
    "rf.fit(X_train, y_train.values.ravel())\n",
    "\n",
    "y_prob_test = rf.predict_proba(X_test)[:, 1] #here we are only interested in True positive\n",
    "y_pred_test = rf.predict(X_test)\n",
    "\n",
    "# We go for test data straight\n",
    "rs.report_matrix(actual_data = y_test, \n",
    "                 prediction_data = y_pred_test, \n",
    "                 ROC = y_prob_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " Classification Report\n",
      "=======================================================\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "         0.0       0.54      0.31      0.40      2687\n",
      "         1.0       0.52      0.74      0.61      2741\n",
      "\n",
      "    accuracy                           0.53      5428\n",
      "   macro avg       0.53      0.53      0.51      5428\n",
      "weighted avg       0.53      0.53      0.51      5428\n",
      "\n",
      "\n",
      " Confusion Matrix\n",
      "=======================================================\n",
      "[[TN, FP]\n",
      " [FN, TP]]\n",
      "\n",
      "[[ 844 1843]\n",
      " [ 714 2027]]\n",
      "\n",
      " Accuracy Score\n",
      "=======================================================\n",
      "0.5289240972733972\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "rf_cls_wght_bln = RandomForestClassifier(max_depth = max_depth, \n",
    "                                    n_estimators = n_estimators,\n",
    "                                    criterion = 'entropy',\n",
    "                                    #bootstrap=True,\n",
    "                                    #max_samples = av_uniqueness_by_coevent['tW'].mean()\n",
    "                                    oob_score = False,\n",
    "                                    class_weight = 'balanced', # We try balance weight only\n",
    "                                    random_state = c_random_state)\n",
    "\n",
    "rf_cls_wght_bln.fit(X_train, y_train.values.ravel())\n",
    "\n",
    "y_prob_test0 = rf_cls_wght_bln.predict_proba(X_test)[:, 1] #here we are only interested in True positive\n",
    "y_pred_test0 = rf_cls_wght_bln.predict(X_test)\n",
    "\n",
    "# We go for test data straight\n",
    "rs.report_matrix(actual_data = y_test, \n",
    "                 prediction_data = y_pred_test0, \n",
    "                 ROC = y_prob_test0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " Classification Report\n",
      "=======================================================\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "         0.0       0.54      0.33      0.41      2687\n",
      "         1.0       0.53      0.73      0.61      2741\n",
      "\n",
      "    accuracy                           0.53      5428\n",
      "   macro avg       0.53      0.53      0.51      5428\n",
      "weighted avg       0.53      0.53      0.51      5428\n",
      "\n",
      "\n",
      " Confusion Matrix\n",
      "=======================================================\n",
      "[[TN, FP]\n",
      " [FN, TP]]\n",
      "\n",
      "[[ 878 1809]\n",
      " [ 737 2004]]\n",
      "\n",
      " Accuracy Score\n",
      "=======================================================\n",
      "0.5309506263817244\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "rf_cls_wght_sub_bln = RandomForestClassifier(max_depth = max_depth, \n",
    "                                    n_estimators = n_estimators,\n",
    "                                    criterion = 'entropy',\n",
    "                                    #bootstrap=True,\n",
    "                                    #max_samples = av_uniqueness_by_coevent['tW'].mean()\n",
    "                                    oob_score = False,\n",
    "                                    class_weight = 'balanced_subsample', # We try balance weight only\n",
    "                                    random_state = c_random_state)\n",
    "\n",
    "rf_cls_wght_sub_bln.fit(X_train, y_train.values.ravel())\n",
    "\n",
    "y_prob_test1 = rf_cls_wght_sub_bln.predict_proba(X_test)[:, 1] #here we are only interested in True positive\n",
    "y_pred_test1 = rf_cls_wght_sub_bln.predict(X_test)\n",
    "\n",
    "# We go for test data straight\n",
    "rs.report_matrix(actual_data = y_test, \n",
    "                 prediction_data = y_pred_test1, \n",
    "                 ROC = y_prob_test1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Conclusion:\n",
    "\n",
    "#### Accuracy\n",
    "\n",
    "Class_weight = 'None'.\n",
    "The Accuracy was about 0.5221075902726603.\n",
    "\n",
    "Class_weight = 'balance'.\n",
    "The Accuracy was about 0.5289240972733972.\n",
    "\n",
    "Class_weight = 'balance_subsample'.\n",
    "The Accuracy was about 0.5309506263817244 (Highest improvement).\n",
    "\n",
    "#### Confusion Matrix\n",
    "\n",
    "The power of a hypothesis test is the probability of making the correct decision if the alternative hypothesis is true, hence to reject a null hypothesis if found to be false.\n",
    "\n",
    "In our case, null hypothesis is FP (Not True until proven, easily rejected). By focusing on TP, we will increase our ML model's accuracy.\n",
    "\n",
    "The distribution Postive (TP and FP) experience an increase, when class_weight was set to 'balance' and 'balance_subsample'.\n",
    "\n",
    "Again, with 'balance_subsample' experience higher allocation to Positive distributions within condusion matrix, which is reflected in their respective accuracy score.\n",
    "\n",
    "Comparing class weight = 'None' and class_weight = 'balance_subsample', True Positive (TP) experienced a jump from 1937 to 2004. In short, when we set class_weight, we are asking ML model to focus on True region specifically TP.\n",
    "\n",
    "This is correct and especially true when we have a high imbalance dataset with TP being a minority, this is also the reason why Dr Marco Lopez wanted to demostrate with an imbalance dataset with 2/3 of negatives meta-labels.\n",
    "\n",
    "For more details refer to Advances in Financial Machine Learning, page 71 - 72, section 4.8."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}