qlib/examples/tutorial/detailed_workflow.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2ac18d8f",
   "metadata": {},
   "outputs": [],
   "source": [
    "#  Copyright (c) Microsoft Corporation.\n",
    "#  Licensed under the MIT License."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "096c6260",
   "metadata": {},
   "source": [
    "# Introduction\n",
    "Though users can automatically run the whole Quant research worklfow based on configurations with Qlib.\n",
    "\n",
    "Some advanced users usally would like to carefully customize each component to explore more in Quant.\n",
    "\n",
    "If you just want a simple example of Qlib. [Quick start](https://github.com/microsoft/qlib#quick-start) and [workflow_by_code](https://github.com/microsoft/qlib/blob/main/examples/workflow_by_code.ipynb) may be a better choice for you.\n",
    "\n",
    "If you want to know more details about Quant research, this notebook may be a better place for you to start.\n",
    "\n",
    "We hope this script could be a tutorial for users who are interested in the details of Quant.\n",
    "\n",
    "This notebook tries to demonstrate how can we use Qlib to build components step by step. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "b96a4196",
   "metadata": {},
   "outputs": [],
   "source": [
    "from pprint import pprint\n",
    "from pathlib import Path\n",
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9e707694",
   "metadata": {},
   "outputs": [],
   "source": [
    "MARKET = \"csi300\"\n",
    "BENCHMARK = \"SH000300\"\n",
    "EXP_NAME = \"tutorial_exp\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ff16a42b",
   "metadata": {},
   "source": [
    "# Data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df055d7d",
   "metadata": {},
   "source": [
    "## Get data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e9898c23",
   "metadata": {},
   "source": [
    "Users can follow [the steps](https://github.com/microsoft/qlib/tree/main/scripts#download-qlib-data) to download data with CLI.\n",
    "\n",
    "In this example we use the underlying API to automatically download data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a0bcfa97",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.tests.data import GetData\n",
    "\n",
    "GetData().qlib_data(exists_skip=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "42b89646",
   "metadata": {},
   "outputs": [],
   "source": [
    "import qlib\n",
    "\n",
    "qlib.init()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "90080d29",
   "metadata": {},
   "source": [
    "## Inspect raw data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "413e99b8",
   "metadata": {},
   "source": [
    "Currently, Qlib support several kinds of data source."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ecf241e9",
   "metadata": {},
   "source": [
    "### Calendar"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0c386f38",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.data import D\n",
    "\n",
    "print(D.calendar(start_time=\"2010-01-01\", end_time=\"2017-12-31\", freq=\"day\")[:2])  # calendar data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f436b49d",
   "metadata": {},
   "source": [
    "### Basic data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a889b763",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = D.features(\n",
    "    [\"SH601216\"],\n",
    "    [\"$open\", \"$high\", \"$low\", \"$close\", \"$factor\"],\n",
    "    start_time=\"2020-05-01\",\n",
    "    end_time=\"2020-05-31\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "eceb43c8",
   "metadata": {},
   "outputs": [],
   "source": [
    "import plotly.graph_objects as go\n",
    "import plotly.io as pio\n",
    "\n",
    "pio.renderers.default = \"notebook\"\n",
    "fig = go.Figure(\n",
    "    data=[\n",
    "        go.Candlestick(\n",
    "            x=df.index.get_level_values(\"datetime\"),\n",
    "            open=df[\"$open\"],\n",
    "            high=df[\"$high\"],\n",
    "            low=df[\"$low\"],\n",
    "            close=df[\"$close\"],\n",
    "        )\n",
    "    ]\n",
    ")\n",
    "fig.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "768ef188",
   "metadata": {},
   "source": [
    "### price adjustment"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d9a536a4",
   "metadata": {},
   "source": [
    "Maybe you think the price is not what it looks like in real world.\n",
    "\n",
    "Due to the price adjustment, the price will be different from the real trading data ."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "45df33b5",
   "metadata": {},
   "outputs": [],
   "source": [
    "import plotly.graph_objects as go\n",
    "\n",
    "fig = go.Figure(\n",
    "    data=[\n",
    "        go.Candlestick(\n",
    "            x=df.index.get_level_values(\"datetime\"),\n",
    "            open=df[\"$open\"] / df[\"$factor\"],\n",
    "            high=df[\"$high\"] / df[\"$factor\"],\n",
    "            low=df[\"$low\"] / df[\"$factor\"],\n",
    "            close=df[\"$close\"] / df[\"$factor\"],\n",
    "        )\n",
    "    ]\n",
    ")\n",
    "fig.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d6acffc3",
   "metadata": {},
   "source": [
    "Please notice the price gap on [2020-05-26](http://vip.stock.finance.sina.com.cn/corp/view/vISSUE_ShareBonusDetail.php?stockid=601216&type=1&end_date=2020-05-20)\n",
    "\n",
    "If we want to represent the change of assets value by price, adjust prices are necesary.\n",
    "By default, Qlib stores the adjusted prices."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "af5f063e",
   "metadata": {},
   "source": [
    "### Static universe V.S. dynamic universe"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8b9b8ce5",
   "metadata": {},
   "source": [
    "Dynamic universe"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "50d3ab70",
   "metadata": {},
   "outputs": [],
   "source": [
    "# dynamic universe\n",
    "universe = D.list_instruments(D.instruments(\"csi100\"), start_time=\"2010-01-01\", end_time=\"2020-12-31\")\n",
    "pprint(universe)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6be08f23",
   "metadata": {},
   "outputs": [],
   "source": [
    "print(len(universe))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "28e7dd04",
   "metadata": {},
   "source": [
    "Qlib use dynamic universe by default.\n",
    "\n",
    "csi100 has around 100 stocks each day(it is not that accurate due to the low precision of data)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ad8b8503",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = D.features(D.instruments(\"csi100\"), [\"$close\"], start_time=\"2010-01-01\", end_time=\"2020-12-31\")\n",
    "df.groupby(\"datetime\").size().plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f37f2ef",
   "metadata": {},
   "source": [
    "### Point-In-Time data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5dfa9e9d",
   "metadata": {},
   "source": [
    "#### download data\n",
    "NOTE: To run the test faster, we only download the data of two stocks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "da0a9564",
   "metadata": {},
   "outputs": [],
   "source": [
    "p = Path(\"~/.qlib/qlib_data/cn_data/financial\").expanduser()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4657fe13",
   "metadata": {},
   "outputs": [],
   "source": [
    "if not p.exists():\n",
    "    !cd ../../scripts/data_collector/pit/ && pip install -r requirements.txt\n",
    "    !cd ../../scripts/data_collector/pit/ && python collector.py download_data --source_dir ~/.qlib/stock_data/source/pit --start 2000-01-01 --end 2020-01-01 --interval quarterly --symbol_regex \"^(600519|000725).*\"\n",
    "    !cd ../../scripts/data_collector/pit/ && python collector.py normalize_data --interval quarterly --source_dir ~/.qlib/stock_data/source/pit --normalize_dir ~/.qlib/stock_data/source/pit_normalized\n",
    "    !cd ../../scripts/ && python dump_pit.py dump --csv_path ~/.qlib/stock_data/source/pit_normalized --qlib_dir ~/.qlib/qlib_data/cn_data --interval quarterly"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9358cb89",
   "metadata": {},
   "source": [
    "#### querying data\n",
    "using `roewa(performanceExpressROEWa,业绩快报净资产收益率ROE-加权)` as an example\n",
    "\n",
    "If we want to get fundamental data `in the most recent quarter` daily, we can use following example.\n",
    "\n",
    "Maitai release part of its fundamental data on [2019-07-13](http://www.cninfo.com.cn/new/disclosure/detail?stockCode=600519&announcementId=1206443183&orgId=gssh0600519&announcementTime=2019-07-13) and  release others on [2019-07-18](http://www.cninfo.com.cn/new/disclosure/detail?stockCode=600519&announcementId=1206456129&orgId=gssh0600519&announcementTime=2019-07-18)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "47ee1621",
   "metadata": {},
   "outputs": [],
   "source": [
    "instruments = [\"sh600519\"]\n",
    "data = D.features(\n",
    "    instruments,\n",
    "    [\"P($$roewa_q)\"],\n",
    "    start_time=\"2019-01-01\",\n",
    "    end_time=\"2019-07-19\",\n",
    "    freq=\"day\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "752f4ffe",
   "metadata": {},
   "outputs": [],
   "source": [
    "data.tail(15)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0e370d2d",
   "metadata": {},
   "source": [
    "### experss engine\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a46d166c",
   "metadata": {},
   "outputs": [],
   "source": [
    "D.features(\n",
    "    [\"sh600519\"],\n",
    "    [\"(EMA($close, 12) - EMA($close, 26))/$close - EMA((EMA($close, 12) - EMA($close, 26))/$close, 9)/$close\"],\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5ddcd1ea",
   "metadata": {},
   "source": [
    "\n",
    "## Dataset loading and preprocessing "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f54c6804",
   "metadata": {},
   "source": [
    "Some heuristic principles of create features\n",
    "- make the features comparable between instrumets: remove unit from the features.\n",
    "- try to keep the distribution invariant\n",
    "- keep the scale of features similar"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "93013dcd",
   "metadata": {},
   "source": [
    "### data loader\n",
    "\n",
    "It's interface can be found [here](https://github.com/microsoft/qlib/blob/main/qlib/data/dataset/loader.py#L24) \n",
    "\n",
    "QlibDataLoader is an implementation which load data from Qlib's data source"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dcfa44a6",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.data.dataset.loader import QlibDataLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5d78b4bf",
   "metadata": {},
   "outputs": [],
   "source": [
    "qdl = QlibDataLoader(config=([\"$close / Ref($close, 10)\"], [\"RET10\"]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3fb29d8e",
   "metadata": {},
   "outputs": [],
   "source": [
    "qdl.load(instruments=[\"sh600519\"], start_time=\"20190101\", end_time=\"20191231\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0dded8f5",
   "metadata": {},
   "source": [
    "### data handler"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "78d6d2b0",
   "metadata": {},
   "source": [
    "finance data can't be perfect.\n",
    "\n",
    "We have to process them before feeding them into Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c078fa3b",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = qdl.load(instruments=[\"sh600519\"], start_time=\"20190101\", end_time=\"20191231\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "45e5adf9",
   "metadata": {},
   "outputs": [],
   "source": [
    "df.isna().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "514b85e6",
   "metadata": {},
   "outputs": [],
   "source": [
    "df.plot(kind=\"hist\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "db1625f7",
   "metadata": {},
   "source": [
    "Datahander is responsible for data preprocessing and provides data fetching interface \n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "43b35c17",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.data.dataset.handler import DataHandlerLP\n",
    "from qlib.data.dataset.processor import ZScoreNorm, Fillna"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "38a5f4b2",
   "metadata": {},
   "outputs": [],
   "source": [
    "# NOTE: normally, the training & validation time range will be  `fit_start_time` ， `fit_end_time`\n",
    "# however，all the components are decomposed, so the training & validation time range is unknown when preprocessing.\n",
    "dh = DataHandlerLP(\n",
    "    instruments=[\"sh600519\"],\n",
    "    start_time=\"20170101\",\n",
    "    end_time=\"20191231\",\n",
    "    infer_processors=[\n",
    "        ZScoreNorm(fit_start_time=\"20170101\", fit_end_time=\"20181231\"),\n",
    "        Fillna(),\n",
    "    ],\n",
    "    data_loader=qdl,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9469fd1e",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = dh.fetch()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fc35b3c0",
   "metadata": {},
   "outputs": [],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1dd8e11b",
   "metadata": {},
   "outputs": [],
   "source": [
    "df.isna().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7208efc3",
   "metadata": {},
   "outputs": [],
   "source": [
    "df.plot(kind=\"hist\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1e0fb32c",
   "metadata": {},
   "source": [
    "### dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0302a801",
   "metadata": {},
   "source": [
    "#### basic dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "96ef76e4",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.data.dataset import DatasetH, TSDatasetH"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "12fd8296",
   "metadata": {},
   "outputs": [],
   "source": [
    "ds = DatasetH(dh, segments={\"train\": (\"20180101\", \"20181231\"), \"valid\": (\"20190101\", \"20191231\")})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4cc4c199",
   "metadata": {},
   "outputs": [],
   "source": [
    "ds.prepare(\"train\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4639b6a7",
   "metadata": {},
   "outputs": [],
   "source": [
    "ds.prepare(\"valid\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a56001e4",
   "metadata": {},
   "source": [
    "#### Time Series Dataset\n",
    "\n",
    "For different model, the required dataset format will be different.\n",
    "\n",
    "For example, Qlib provides a Time Series Dataset(TSDatasetH) to help users to create time-series dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "425135e1",
   "metadata": {},
   "outputs": [],
   "source": [
    "ds = TSDatasetH(\n",
    "    step_len=10,\n",
    "    handler=dh,\n",
    "    segments={\"train\": (\"20180101\", \"20181231\"), \"valid\": (\"20190101\", \"20191231\")},\n",
    ")\n",
    "train_sampler = ds.prepare(\"train\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8f724041",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_sampler"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e5aa762c",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_sampler[0]  # Retrieving the first example"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "eb64112c",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_sampler[\"2018-01-08\", \"sh600519\"]  # get the time series by <'timestamp', 'instrument_id'> index"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4e6ab197",
   "metadata": {},
   "source": [
    "### Off-the-shelf dataset\n",
    "\n",
    "Qlib integrated some dataset alreadly"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2e21b45c",
   "metadata": {
    "code_folding": []
   },
   "outputs": [],
   "source": [
    "handler_kwargs = {\n",
    "    \"start_time\": \"2008-01-01\",\n",
    "    \"end_time\": \"2020-08-01\",\n",
    "    \"fit_start_time\": \"2008-01-01\",\n",
    "    \"fit_end_time\": \"2014-12-31\",\n",
    "    \"instruments\": MARKET,\n",
    "}\n",
    "handler_conf = {\n",
    "    \"class\": \"Alpha158\",\n",
    "    \"module_path\": \"qlib.contrib.data.handler\",\n",
    "    \"kwargs\": handler_kwargs,\n",
    "}\n",
    "pprint(handler_conf)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "17077f0c",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.utils import init_instance_by_config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6a20d9b2",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd = init_instance_by_config(handler_conf)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5aa02379",
   "metadata": {},
   "source": [
    "Using config to create instance is a highly frequently used practice in Qlib (e.g. the [workflows configurations](https://github.com/microsoft/qlib/blob/main/examples/benchmarks/LightGBM/workflow_config_lightgbm_Alpha158.yaml) are based on it).\n",
    "\n",
    "\n",
    "The above configuration is the same as the code below"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "480d35bf",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.contrib.data.handler import Alpha158\n",
    "\n",
    "hd = Alpha158(**handler_kwargs)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b170153d",
   "metadata": {},
   "source": [
    "This dataset has the same structure as the simple one with 1 column  we created just now."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "735758e5",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = hd.fetch()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2b6de50c",
   "metadata": {},
   "outputs": [],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cb1bc4fb",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd.data_loader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5a927f1c",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd.data_loader.fields"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f5a15747",
   "metadata": {},
   "source": [
    "#### some details\n",
    "\n",
    "The training data may not be the same as the test data.\n",
    "\n",
    "e.g.\n",
    "- the training dataset and test dataset use a different fitlering rules,  data processing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cf2defa0",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd.learn_processors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ef55a881",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd.infer_processors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7af3b077",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd.process_type  # appending type"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "20127e19",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd.fetch(col_set=\"label\", data_key=hd.DK_L)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "09c37d41",
   "metadata": {},
   "outputs": [],
   "source": [
    "hd.fetch(col_set=\"label\", data_key=hd.DK_I)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ec2cfb12",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_conf = {\n",
    "    \"class\": \"DatasetH\",\n",
    "    \"module_path\": \"qlib.data.dataset\",\n",
    "    \"kwargs\": {\n",
    "        \"handler\": hd,\n",
    "        \"segments\": {\n",
    "            \"train\": (\"2008-01-01\", \"2014-12-31\"),\n",
    "            \"valid\": (\"2015-01-01\", \"2016-12-31\"),\n",
    "            \"test\": (\"2017-01-01\", \"2020-08-01\"),\n",
    "        },\n",
    "    },\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aca33c3c",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = init_instance_by_config(dataset_conf)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0c89c15d",
   "metadata": {},
   "source": [
    "# Model Training & Inference\n",
    "\n",
    "[Model interface](https://github.com/microsoft/qlib/blob/main/qlib/model/base.py)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0e916286",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.workflow import R\n",
    "from qlib.workflow.record_temp import SignalRecord, PortAnaRecord, SigAnaRecord"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f6975911",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = init_instance_by_config(\n",
    "    {\n",
    "        \"class\": \"LGBModel\",\n",
    "        \"module_path\": \"qlib.contrib.model.gbdt\",\n",
    "        \"kwargs\": {\n",
    "            \"loss\": \"mse\",\n",
    "            \"colsample_bytree\": 0.8879,\n",
    "            \"learning_rate\": 0.0421,\n",
    "            \"subsample\": 0.8789,\n",
    "            \"lambda_l1\": 205.6999,\n",
    "            \"lambda_l2\": 580.9768,\n",
    "            \"max_depth\": 8,\n",
    "            \"num_leaves\": 210,\n",
    "            \"num_threads\": 20,\n",
    "        },\n",
    "    }\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2e2dafb6",
   "metadata": {},
   "outputs": [],
   "source": [
    "# start exp to train model\n",
    "with R.start(experiment_name=EXP_NAME):\n",
    "    model.fit(dataset)\n",
    "    R.save_objects(trained_model=model)\n",
    "\n",
    "    rec = R.get_recorder()\n",
    "    rid = rec.id  # save the record id\n",
    "\n",
    "    # Inference and saving signal\n",
    "    sr = SignalRecord(model, dataset, rec)\n",
    "    sr.generate()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b6b9f3d",
   "metadata": {},
   "source": [
    "# Evaluation:\n",
    "- Signal-based\n",
    "- Portfolio-based: backtest "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4328f881",
   "metadata": {},
   "outputs": [],
   "source": [
    "###################################\n",
    "# prediction, backtest & analysis\n",
    "###################################\n",
    "port_analysis_config = {\n",
    "    \"executor\": {\n",
    "        \"class\": \"SimulatorExecutor\",\n",
    "        \"module_path\": \"qlib.backtest.executor\",\n",
    "        \"kwargs\": {\n",
    "            \"time_per_step\": \"day\",\n",
    "            \"generate_portfolio_metrics\": True,\n",
    "        },\n",
    "    },\n",
    "    \"strategy\": {\n",
    "        \"class\": \"TopkDropoutStrategy\",\n",
    "        \"module_path\": \"qlib.contrib.strategy.signal_strategy\",\n",
    "        \"kwargs\": {\n",
    "            \"signal\": \"<PRED>\",\n",
    "            \"topk\": 50,\n",
    "            \"n_drop\": 5,\n",
    "        },\n",
    "    },\n",
    "    \"backtest\": {\n",
    "        \"start_time\": \"2017-01-01\",\n",
    "        \"end_time\": \"2020-08-01\",\n",
    "        \"account\": 100000000,\n",
    "        \"benchmark\": BENCHMARK,\n",
    "        \"exchange_kwargs\": {\n",
    "            \"freq\": \"day\",\n",
    "            \"limit_threshold\": 0.095,\n",
    "            \"deal_price\": \"close\",\n",
    "            \"open_cost\": 0.0005,\n",
    "            \"close_cost\": 0.0015,\n",
    "            \"min_cost\": 5,\n",
    "        },\n",
    "    },\n",
    "}\n",
    "\n",
    "# backtest and analysis\n",
    "with R.start(experiment_name=EXP_NAME, recorder_id=rid, resume=True):\n",
    "    # signal-based analysis\n",
    "    rec = R.get_recorder()\n",
    "    sar = SigAnaRecord(rec)\n",
    "    sar.generate()\n",
    "\n",
    "    #  portfolio-based analysis: backtest\n",
    "    par = PortAnaRecord(rec, port_analysis_config, \"day\")\n",
    "    par.generate()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d66ad59d",
   "metadata": {},
   "source": [
    "# Loading results & Analysis"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d7d7dea",
   "metadata": {},
   "source": [
    "## loading data\n",
    "Because Qlib leverage MLflow to save model & data.\n",
    "All the data can be access by `mlflow ui`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1ec9dbb6",
   "metadata": {},
   "outputs": [],
   "source": [
    "# load recorder\n",
    "recorder = R.get_recorder(recorder_id=rid, experiment_name=EXP_NAME)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "25e72b0d",
   "metadata": {},
   "outputs": [],
   "source": [
    "# load previous results\n",
    "pred_df = recorder.load_object(\"pred.pkl\")\n",
    "report_normal_df = recorder.load_object(\"portfolio_analysis/report_normal_1day.pkl\")\n",
    "positions = recorder.load_object(\"portfolio_analysis/positions_normal_1day.pkl\", trusted=True)\n",
    "analysis_df = recorder.load_object(\"portfolio_analysis/port_analysis_1day.pkl\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dce3696b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Previous Model can be loaded. but it is not used.\n",
    "loaded_model = recorder.load_object(\"trained_model\", trusted=True)\n",
    "loaded_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cf8eca78",
   "metadata": {},
   "outputs": [],
   "source": [
    "from qlib.contrib.report import analysis_model, analysis_position"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8d34b347",
   "metadata": {},
   "source": [
    "## analysis position"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1cae642f",
   "metadata": {},
   "source": [
    "### report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "47c727b2",
   "metadata": {},
   "outputs": [],
   "source": [
    "analysis_position.report_graph(report_normal_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "15b7ca14",
   "metadata": {},
   "source": [
    "### risk analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3f100690",
   "metadata": {},
   "outputs": [],
   "source": [
    "analysis_position.risk_analysis_graph(analysis_df, report_normal_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2ed48aee",
   "metadata": {},
   "source": [
    "## analysis model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aec13561",
   "metadata": {},
   "outputs": [],
   "source": [
    "label_df = dataset.prepare(\"test\", col_set=\"label\")\n",
    "label_df.columns = [\"label\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "56d57363",
   "metadata": {},
   "source": [
    "### score IC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7612533d",
   "metadata": {},
   "outputs": [],
   "source": [
    "pred_label = pd.concat([label_df, pred_df], axis=1, sort=True).reindex(label_df.index)\n",
    "analysis_position.score_ic_graph(pred_label)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "589664c4",
   "metadata": {},
   "source": [
    "### model performance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "40258655",
   "metadata": {},
   "outputs": [],
   "source": [
    "analysis_model.model_performance_graph(pred_label)"
   ]
  }
 ],
 "metadata": {
  "jupytext": {
   "encoding": "# -*- coding: utf-8 -*-",
   "formats": "ipynb,py:percent"
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "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": 5
}