18_convolutional_neural_nets/06_cnn_for_trading_features_to_clustered_image_format.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# CNN for Trading - Part 2: From Time-Series Features to Clustered Images"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T16:39:56.783513Z",
     "start_time": "2021-02-23T16:39:56.774804Z"
    }
   },
   "source": [
    "To exploit the grid-like structure of time-series data, we can use CNN architectures for univariate and multivariate time series. In the latter case, we consider different time series as channels, similar to the different color signals.\n",
    "\n",
    "An alternative approach converts a time series of alpha factors into a two-dimensional format to leverage the ability of CNNs to detect local patterns. [Sezer and Ozbayoglu (2018)](https://www.researchgate.net/publication/324802031_Algorithmic_Financial_Trading_with_Deep_Convolutional_Neural_Networks_Time_Series_to_Image_Conversion_Approach) propose CNN-TA, which computes 15 technical indicators for different intervals and uses hierarchical clustering (see Chapter 13, Data-Driven Risk Factors and Asset Allocation with Unsupervised Learning) to locate indicators that behave similarly close to each other in a two-dimensional grid.\n",
    "\n",
    "The authors train a CNN similar to the CIFAR-10 example we used earlier to predict whether to buy, hold, or sell an asset on a given day. They compare the CNN performance to \"buy-and-hold\" and other models and find that it outperforms all alternatives using daily price series for Dow 30 stocks and the nine most-traded ETFs over the 2007-2017 time period.\n",
    "\n",
    "The section on *CNN for Trading* consists of three notebooks that experiment with this approach using daily US equity price data. They demonstrate \n",
    "1. How to compute relevant financial features\n",
    "2. How to convert a similar set of indicators into image format and cluster them by similarity\n",
    "3. How to train a CNN to predict daily returns and evaluate a simple long-short strategy based on the resulting signals."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Selecting and Clustering Features"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The next steps that we will tackle in this notebook are \n",
    "1. Select the 15 most relevant features from the 20 candidates to fill the 15×15 input grid.\n",
    "2. Apply hierarchical clustering to identify features that behave similarly and order the columns and the rows of the grid accordingly."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Imports & Settings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:23.064329Z",
     "start_time": "2021-02-23T18:59:23.062565Z"
    }
   },
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:23.721271Z",
     "start_time": "2021-02-23T18:59:23.065808Z"
    }
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "from pathlib import Path\n",
    "import pandas as pd\n",
    "from tqdm import tqdm\n",
    "\n",
    "from scipy.spatial.distance import pdist\n",
    "from scipy.cluster.hierarchy import dendrogram, linkage, cophenet\n",
    "\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.feature_selection import mutual_info_regression\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:23.723998Z",
     "start_time": "2021-02-23T18:59:23.722349Z"
    }
   },
   "outputs": [],
   "source": [
    "MONTH = 21\n",
    "YEAR = 12 * MONTH"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:23.737796Z",
     "start_time": "2021-02-23T18:59:23.724937Z"
    }
   },
   "outputs": [],
   "source": [
    "START = '2001-01-01'\n",
    "END = '2017-12-31'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:23.750549Z",
     "start_time": "2021-02-23T18:59:23.738814Z"
    }
   },
   "outputs": [],
   "source": [
    "sns.set_style('white')\n",
    "idx = pd.IndexSlice"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:23.758227Z",
     "start_time": "2021-02-23T18:59:23.751920Z"
    }
   },
   "outputs": [],
   "source": [
    "results_path = Path('results', 'cnn_for_trading')\n",
    "if not results_path.exists():\n",
    "    results_path.mkdir(parents=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Model Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:30.844081Z",
     "start_time": "2021-02-23T18:59:23.759439Z"
    }
   },
   "outputs": [],
   "source": [
    "with pd.HDFStore('data.h5') as store:\n",
    "    features = store.get('features')\n",
    "    targets = store.get('targets')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:30.896504Z",
     "start_time": "2021-02-23T18:59:30.845003Z"
    },
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "MultiIndex: 2378728 entries, ('A', Timestamp('2001-01-02 00:00:00')) to ('ZTS', Timestamp('2017-12-29 00:00:00'))\n",
      "Columns: 300 entries, 06_RSI to 85_CMA\n",
      "dtypes: float64(300)\n",
      "memory usage: 5.3+ GB\n"
     ]
    }
   ],
   "source": [
    "features.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T18:59:30.938773Z",
     "start_time": "2021-02-23T18:59:30.897581Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "MultiIndex: 2378728 entries, ('A', Timestamp('2001-01-02 00:00:00')) to ('ZTS', Timestamp('2017-12-29 00:00:00'))\n",
      "Data columns (total 4 columns):\n",
      " #   Column      Dtype  \n",
      "---  ------      -----  \n",
      " 0   r01_fwd     float64\n",
      " 1   r01dec_fwd  float64\n",
      " 2   r05_fwd     float64\n",
      " 3   r05dec_fwd  float64\n",
      "dtypes: float64(4)\n",
      "memory usage: 81.8+ MB\n"
     ]
    }
   ],
   "source": [
    "targets.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Select Features using Mutual Information"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To this end, we estimate the mutual information for each indicator and the 15 intervals with respect to our target, the one-day forward returns. As discussed in Chapter 4, Financial Feature Engineering – How to Research Alpha Factors, scikit-learn provides the `mutual_info_regression()` function that makes this straightforward, albeit time-consuming and memory-intensive. \n",
    "\n",
    "To accelerate the process, we randomly sample 100,000 observations:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.366522Z",
     "start_time": "2021-02-23T18:59:30.939773Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 2/2 [21:21<00:00, 640.71s/it]\n"
     ]
    }
   ],
   "source": [
    "mi = {}\n",
    "for t in tqdm([1, 5]):\n",
    "    target = f'r{t:02}_fwd'\n",
    "    # sample a smaller number to speed up the computation\n",
    "    df = features.join(targets[target]).dropna().sample(n=100000)\n",
    "    X = df.drop(target, axis=1)\n",
    "    y = df[target]\n",
    "    mi[t] = pd.Series(mutual_info_regression(X=X, y=y),\n",
    "                      index=X.columns).sort_values(ascending=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.376177Z",
     "start_time": "2021-02-23T19:20:52.367645Z"
    }
   },
   "outputs": [],
   "source": [
    "mutual_info = pd.DataFrame(mi)\n",
    "mutual_info.to_hdf('data.h5', 'mutual_info')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.391153Z",
     "start_time": "2021-02-23T19:20:52.377122Z"
    }
   },
   "outputs": [],
   "source": [
    "mutual_info = pd.read_hdf('data.h5', 'mutual_info')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.405177Z",
     "start_time": "2021-02-23T19:20:52.392078Z"
    }
   },
   "outputs": [],
   "source": [
    "mi_by_indicator = (mutual_info.groupby(mutual_info.\n",
    "                                       index.to_series()\n",
    "                                       .str.split('_').str[-1])\n",
    "                   .mean()\n",
    "                   .rank(ascending=False)\n",
    "                   .sort_values(by=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.550095Z",
     "start_time": "2021-02-23T19:20:52.406148Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "mutual_info.boxplot()\n",
    "sns.despine();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The below figure shows the mutual information, averaged across the 15 intervals for each indicator. NATR, PPO, and Bollinger Bands are most important from this metric's perspective:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.937686Z",
     "start_time": "2021-02-23T19:20:52.551300Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "(mutual_info.groupby(mutual_info.index.to_series().str.split('_').str[-1])[1]\n",
    " .mean()\n",
    " .sort_values().plot.barh(title='Mutual Information with 1-Day Forward Returns'))\n",
    "sns.despine()\n",
    "plt.tight_layout()\n",
    "plt.savefig(results_path / 'mutual_info_cnn_features', dpi=300)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.941987Z",
     "start_time": "2021-02-23T19:20:52.938868Z"
    }
   },
   "outputs": [],
   "source": [
    "best_features = mi_by_indicator.head(15).index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:52.971896Z",
     "start_time": "2021-02-23T19:20:52.943614Z"
    }
   },
   "outputs": [],
   "source": [
    "size = len(best_features)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Hierarchical Feature Clustering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:20:56.199036Z",
     "start_time": "2021-02-23T19:20:52.973214Z"
    }
   },
   "outputs": [],
   "source": [
    "features = pd.concat([features.filter(like=f'_{f}') for f in best_features], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:21:00.180174Z",
     "start_time": "2021-02-23T19:20:56.200032Z"
    }
   },
   "outputs": [],
   "source": [
    "new_cols = {}\n",
    "for feature in best_features:\n",
    "    fnames = sorted(features.filter(like=f'_{feature}').columns.tolist())\n",
    "    renamed = [f'{i:02}_{feature}' for i in range(1, len(fnames)+ 1)]\n",
    "    new_cols.update(dict(zip(fnames, renamed)))\n",
    "features = features.rename(columns=new_cols).sort_index(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:21:00.194831Z",
     "start_time": "2021-02-23T19:21:00.181519Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "MultiIndex: 2378728 entries, ('A', Timestamp('2001-01-02 00:00:00')) to ('ZTS', Timestamp('2017-12-29 00:00:00'))\n",
      "Columns: 225 entries, 01_BBH to 15_WMA\n",
      "dtypes: float64(225)\n",
      "memory usage: 4.1+ GB\n"
     ]
    }
   ],
   "source": [
    "features.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Hierarchical Clustering"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As discussed in the first section of this chapter, CNNs rely on the locality of relevant patterns that is typically found in images where nearby pixels are closely related and changes from one pixel to the next are often gradual.\n",
    "\n",
    "To organize our indicators in a similar fashion, we will follow Sezer and Ozbayoglu's approach of applying hierarchical clustering. The goal is to identify features that behave similarly and order the columns and the rows of the grid accordingly.\n",
    "\n",
    "We can build on SciPy's `pairwise_distance()`, `linkage()`, and `dendrogram()` functions that we introduced in *Chapter 13, Data-Driven Risk Factors and Asset Allocation with Unsupervised Learning* alongside other forms of clustering. \n",
    "\n",
    "We create a helper function that standardizes the input column-wise to avoid distorting distances among features due to differences in scale, and use the Ward criterion that merges clusters to minimize variance. The function\n",
    "returns the order of the leaf nodes in the dendrogram that in turn displays the successive formation of larger clusters:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:21:00.214677Z",
     "start_time": "2021-02-23T19:21:00.196205Z"
    }
   },
   "outputs": [],
   "source": [
    "def cluster_features(data, labels, ax, title):\n",
    "    data = StandardScaler().fit_transform(data)\n",
    "    pairwise_distance = pdist(data)\n",
    "    Z = linkage(data, 'ward')\n",
    "    c, coph_dists = cophenet(Z, pairwise_distance)\n",
    "    dend = dendrogram(Z,\n",
    "                      labels=labels,\n",
    "                      orientation='top',\n",
    "                      leaf_rotation=0.,\n",
    "                      leaf_font_size=8.,\n",
    "                      ax=ax)\n",
    "    ax.set_title(title)\n",
    "    return dend['ivl']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To obtain the optimized order of technical indicators in the columns and the different intervals in the rows, we use NumPy's `.reshape()` method to ensure that the dimension we would like to cluster appears in the columns of the two-dimensional array we pass to `cluster_features()`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:21:57.861182Z",
     "start_time": "2021-02-23T19:21:00.215792Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = plt.subplots(figsize=(15, 4), ncols=2)\n",
    "\n",
    "labels = sorted(best_features)\n",
    "title = 'Column Features: Indicators'\n",
    "col_order = cluster_features(features.dropna().values.reshape(-1, 15).T,\n",
    "                             labels,\n",
    "                             axes[0],\n",
    "                             title)\n",
    "\n",
    "labels = list(range(1, 16))\n",
    "title = 'Row Features: Indicator Parameters'\n",
    "row_order = cluster_features(\n",
    "    features.dropna().values.reshape(-1, 15, 15).transpose((0, 2, 1)).reshape(-1, 15).T,\n",
    "    labels, axes[1], title)\n",
    "axes[0].set_xlabel('Indicators')\n",
    "axes[1].set_xlabel('Parameters')\n",
    "sns.despine()\n",
    "fig.tight_layout()\n",
    "fig.savefig(results_path / 'cnn_clustering', dpi=300)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We reorder the features accordingly and store the result as inputs for the CNN that we will create in the next step.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:21:57.869035Z",
     "start_time": "2021-02-23T19:21:57.863039Z"
    }
   },
   "outputs": [],
   "source": [
    "feature_order = [f'{i:02}_{j}' for i in row_order for j in col_order]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:21:58.596134Z",
     "start_time": "2021-02-23T19:21:57.871605Z"
    }
   },
   "outputs": [],
   "source": [
    "features = features.loc[:, feature_order]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:22:56.029721Z",
     "start_time": "2021-02-23T19:21:58.597104Z"
    }
   },
   "outputs": [],
   "source": [
    "features = features.apply(pd.to_numeric, downcast='float')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:22:56.069245Z",
     "start_time": "2021-02-23T19:22:56.030777Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "MultiIndex: 2378728 entries, ('A', Timestamp('2001-01-02 00:00:00')) to ('ZTS', Timestamp('2017-12-29 00:00:00'))\n",
      "Columns: 225 entries, 01_CMO to 11_WMA\n",
      "dtypes: float32(225)\n",
      "memory usage: 2.0+ GB\n"
     ]
    }
   ],
   "source": [
    "features.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2021-02-23T19:22:58.021376Z",
     "start_time": "2021-02-23T19:22:56.070350Z"
    }
   },
   "outputs": [],
   "source": [
    "features.to_hdf('data.h5', 'img_data')"
   ]
  }
 ],
 "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": "3.8.5"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": true,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": true
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}