xgb_custom/xgb_custom.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Custom Objective and Evaluation Functions\n",
    "\n",
    "This demo was adapted from **[xgboost's custom metric tutorial](https://xgboost.readthedocs.io/en/latest/tutorials/custom_metric_obj.html)**.  We demonstrate how to use a custom objective and a custom evaluation function using an xgboost trainer.\n",
    "\n",
    "This function differs from `xgb_trainer` by exposing the low-level xgboost python api, `train`.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# nuclio: ignore\n",
    "import nuclio"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from os import path\n",
    "import numpy as np\n",
    "from numpy.random import randint, randn, seed\n",
    "import pandas as pd\n",
    "from xgboost import DMatrix, train\n",
    "import matplotlib.pyplot as plt\n",
    "from mlrun.execution import MLClientCtx\n",
    "from mlrun.datastore import DataItem\n",
    "from mlrun.artifacts import PlotArtifact\n",
    "from mlrun.mlutils.data import get_splits, get_sample\n",
    "\n",
    "from cloudpickle import dumps\n",
    "\n",
    "from typing import (Tuple, Dict, List, Union, Callable)\n",
    "\n",
    "seed(seed=1994)\n",
    "\n",
    "## UNCOMMENT THIS LINE TO TEST CALCULATED VALUES\n",
    "DEBUG_ERROR = 0 # this will be added to the custom eval function--set it to some value like 999 "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### generate data with outliers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gen_outliers(context: MLClientCtx, nrows=4096, feats=16, \n",
    "                 outs=64, omax=10_000, labels_col=\"labels\",\n",
    "                 header=[], label_type=\"int32\", key=\"xgb-outs\",\n",
    "                 local_path=\"xgb_custom\"):\n",
    "    \"\"\"simulate data with outliers\n",
    "    \n",
    "    :param context:    the function's execution context\n",
    "    :param nrows:      (4096) number of data points\n",
    "    :param feats:      (16) number of features\n",
    "    :param outs:       (64) number of outliers\n",
    "    :param omax:       (10_100) max value of outliers\n",
    "    :param labels_col: (labels) name of labels column\n",
    "    :param header:     () header for dataset, will default to\n",
    "                       `feat_`\n",
    "    :param label_type: (int32) data type for the label column\n",
    "    :param key:        key of datset in artifact store\n",
    "    :param local_path: path in artifact store where data will be\n",
    "                       serialized\n",
    "    \"\"\"\n",
    "    x = randn(nrows, feats)\n",
    "    y = randn(nrows)\n",
    "    y += np.abs(np.min(y))\n",
    "\n",
    "    for i in range(0, outs):\n",
    "        ind = randint(0, len(y)-1)\n",
    "        y[ind] += randint(0, omax)\n",
    "    \n",
    "    if not header:\n",
    "        header = [f\"feat_{j}\" for j in range(feats)]\n",
    "        header.append(labels_col)\n",
    "\n",
    "    data = pd.DataFrame(data=np.concatenate((x,y.reshape(-1,1)),axis=-1),\n",
    "                        columns=header)\n",
    "    data = data.astype({labels_col: label_type})\n",
    "    \n",
    "    context.log_dataset(key, df=data, local_path=local_path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## custom objective and eval"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "toc-hr-collapsed": true,
    "toc-nb-collapsed": true
   },
   "source": [
    "The following code was adapted from xgboost's documentation **[Custom Objective and Evaluation Metric](https://xgboost.readthedocs.io/en/latest/tutorials/custom_metric_obj.html?highlight=tree_method#custom-objective-and-evaluation-metric)**."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gradient(predt: np.ndarray, dtrain: DMatrix) -> np.ndarray:\n",
    "    \"\"\"gradient of squared log error\"\"\"\n",
    "    y = dtrain.get_label()\n",
    "    return (np.log1p(predt) - np.log1p(y)) / (predt + 1)\n",
    "\n",
    "\n",
    "def hessian(predt: np.ndarray, dtrain: DMatrix) -> np.ndarray:\n",
    "    \"\"\"hessian of squared log error\"\"\"\n",
    "    y = dtrain.get_label()\n",
    "    return ((-np.log1p(predt) + np.log1p(y) + 1) /\n",
    "            np.power(predt + 1, 2))\n",
    "\n",
    "\n",
    "def squared_log(predt: np.ndarray, dtrain: DMatrix) -> Tuple[np.ndarray,\n",
    "                                                             np.ndarray]:\n",
    "    \"\"\"squared log error objective\n",
    "\n",
    "    simplified version for RMSLE used as objective function\n",
    "    \"\"\"\n",
    "    predt[predt < -1] = -1 + 1e-6\n",
    "    grad = gradient(predt, dtrain)\n",
    "    hess = hessian(predt, dtrain)\n",
    "    return grad, hess\n",
    "\n",
    "def rmsle(predt: np.ndarray, dtrain: DMatrix) -> Tuple[str, float]:\n",
    "    \"\"\" Root mean squared log error metric.\n",
    "    \"\"\"\n",
    "    y = dtrain.get_label()\n",
    "    predt[predt < -1] = -1 + 1e-6\n",
    "    elements = np.power(np.log1p(y) - np.log1p(predt), 2)\n",
    "    return \"my_rmsle\", float(np.sqrt(np.sum(elements) / len(y))) + DEBUG_ERROR"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## learning curves"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "def learning_curves(\n",
    "    context: MLClientCtx,\n",
    "    results: dict,\n",
    "    figsz: Tuple[int,int]=(10,10),\n",
    "    plots_dest: str = \"plots\"\n",
    ") -> None:\n",
    "    \"\"\"plot xgb learning curves\n",
    "    \n",
    "    this will also log a model's learning curves\n",
    "    \"\"\"\n",
    "    plt.clf()\n",
    "    plt.figure(figsize=figsz)\n",
    "    plt.plot(results[\"train\"][\"my_rmsle\"], label=\"train-my-rmsle\")\n",
    "    plt.plot(results[\"valid\"][\"my_rmsle\"], label=\"valid-my-rmsle\")\n",
    "    plt.title(f\"learning curves\")\n",
    "    plt.legend()\n",
    " \n",
    "    context.log_artifact(\n",
    "        PlotArtifact(f\"learning-curves\", body=plt.gcf()),\n",
    "        local_path=f\"{plots_dest}/learning-curves.html\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## fit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def fit(\n",
    "    context: MLClientCtx,\n",
    "    dataset: DataItem,\n",
    "    num_boost_round: int = 10,\n",
    "    evals: List[Tuple[DMatrix, str]] = [],\n",
    "    obj: Union[Callable, str] = \"\",\n",
    "    feval: Union[Callable, str] = None,\n",
    "    maximize: bool = False,\n",
    "    early_stopping_rounds: int = None,\n",
    "    evals_result: dict = {},\n",
    "    verbose_eval: bool = True,\n",
    "    xgb_model: DataItem = None,\n",
    "    callbacks: List[Callable] = [],\n",
    "    label_column: str = \"labels\",\n",
    "    encode_cols: dict = {},\n",
    "    sample: int = -1,\n",
    "    test_size: float = 0.25,\n",
    "    valid_size: float = 0.75,\n",
    "    random_state: int = 1994,\n",
    "    models_dest: str = \"models\",\n",
    "    plots_dest: str = \"plots\",\n",
    "    file_ext: str = \"csv\",\n",
    "    test_set_key: str = \"test-set\",\n",
    "    gpus: bool = False\n",
    ") -> None:\n",
    "    \"\"\"low level xgboost train api\n",
    "    \n",
    "    for the xgboost `train` params see:\n",
    "    https://xgboost.readthedocs.io/en/latest/python/python_api.html#xgboost.train\n",
    "\n",
    "    Note:  the first parameter of xgboost's `train` method is a dict of parameters\n",
    "           supplied to the booster (engine).  To modify one of those simply\n",
    "           add a task parameter (when running you supply an mlrun NewTask) with the\n",
    "           prefix \"XGB_\". So for example, to set the 'tree_method' parameter to 'approx',\n",
    "           add {\"XGB_tree_method\":\"approx\"} to the task params key.\n",
    "    \n",
    "    :param context:           the function context\n",
    "    :param dataset:           the full data set, train, valid and test will be extracted and\n",
    "                              each converted to a DMatrix for input to xgboost's `train`\n",
    "    :param label_column:      ground-truth (y) labels\n",
    "    :param encode_cols:       dictionary of names and prefixes for columns that are\n",
    "                              to hot be encoded.\n",
    "    :param sample:            Selects the first n rows, or select a sample\n",
    "                              starting from the first. If negative <-1, select\n",
    "                              a random sample\n",
    "    :param test_size:         (0.05) test set size\n",
    "    :param valid_size:        (0.75) Once the test set has been removed the\n",
    "                              training set gets this proportion.\n",
    "    :param random_state:      (1) sklearn rng seed\n",
    "    :param models_dest:       destination subfolder for model artifacts\n",
    "    :param plots_dest:        destination subfolder for plot artifacts\n",
    "    :param file_ext:          format for test_set_key hold out data\n",
    "    :param test_set_key:      (test-set), key of held out data in artifact store\n",
    "    :param gpus:              (False), run on gpus\n",
    "    \"\"\"\n",
    "    raw, labels, header = get_sample(dataset, sample, label_column)\n",
    "    \n",
    "    # hot-encode\n",
    "    if encode_cols:\n",
    "        raw = pd.get_dummies(raw, \n",
    "                             columns=list(encode_cols.keys()), \n",
    "                             prefix=list(encode_cols.values()), \n",
    "                             drop_first=True)\n",
    "    \n",
    "    # split the sample into train validate, test and calibration sets:\n",
    "    (xtrain, ytrain), (xvalid, yvalid), (xtest, ytest) = \\\n",
    "        get_splits(raw, labels, 3, test_size, valid_size, random_state)\n",
    "    \n",
    "    # save test data as regular dataframe as it may be used by other process\n",
    "    context.log_dataset(test_set_key, df=pd.concat([xtest, ytest], axis=1),\n",
    "                        format=file_ext, index=False)\n",
    "    \n",
    "    # convert to xgboost DMatrix (todo - dask, gpu)\n",
    "    dtrain = DMatrix(xtrain, label=ytrain)\n",
    "    dvalid = DMatrix(xvalid, label=yvalid)\n",
    "    \n",
    "    boost_params = {\n",
    "        \"tree_method\": \"gpu_hist\" if gpus else \"hist\", \n",
    "        \"seed\": random_state,\n",
    "        \"disable_default_eval_metric\": 1,\n",
    "        \"objective\": \"reg:squaredlogerror\",\n",
    "        \"eval_metric\": \"rmsle\"}\n",
    "\n",
    "    # enable user to customize `booster param` parameters\n",
    "    for k, v in context.parameters.items():\n",
    "        if k.startswith('XGB_'):\n",
    "            boost_params[k[4:]] = v\n",
    "    \n",
    "    # collect learning curves / training history\n",
    "    results = dict()\n",
    "    \n",
    "    booster = train(\n",
    "        boost_params,\n",
    "        dtrain=dtrain,\n",
    "        num_boost_round=num_boost_round,\n",
    "        evals=[(dtrain, \"train\"), (dvalid, \"valid\")],\n",
    "        evals_result=results,\n",
    "        obj=squared_log,\n",
    "        feval=rmsle,\n",
    "        maximize=maximize,\n",
    "        early_stopping_rounds=early_stopping_rounds,\n",
    "        verbose_eval=verbose_eval,\n",
    "        # xgb_model=xgb_model,\n",
    "        # callbacks: List[Callable] = []\n",
    "    )\n",
    "    \n",
    "    context.log_model(\"model\", \n",
    "                      body=dumps(booster),\n",
    "                      model_file = \"model.pkl\",\n",
    "                      artifact_path='/User/artifacts/tttt')\n",
    "    \n",
    "    learning_curves(context, results)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# nuclio: end-code"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### run locally"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "from mlrun import NewTask, run_local"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[mlrun] 2020-06-14 13:30:21,675 starting run gen_outliers uid=29ae7c944a184de881acc81206a92a48  -> http://mlrun-api:8080\n",
      "[mlrun] 2020-06-14 13:30:22,141 log artifact xgb-outs at /User/artifacts/xgb_custom.csv, size: 2762858, db: Y\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<style> \n",
       ".dictlist {\n",
       "  background-color: #b3edff; \n",
       "  text-align: center; \n",
       "  margin: 4px; \n",
       "  border-radius: 3px; padding: 0px 3px 1px 3px; display: inline-block;}\n",
       ".artifact {\n",
       "  cursor: pointer; \n",
       "  background-color: #ffe6cc; \n",
       "  text-align: left; \n",
       "  margin: 4px; border-radius: 3px; padding: 0px 3px 1px 3px; display: inline-block;\n",
       "}\n",
       "div.block.hidden {\n",
       "  display: none;\n",
       "}\n",
       ".clickable {\n",
       "  cursor: pointer;\n",
       "}\n",
       ".ellipsis {\n",
       "  display: inline-block;\n",
       "  max-width: 60px;\n",
       "  white-space: nowrap;\n",
       "  overflow: hidden;\n",
       "  text-overflow: ellipsis;\n",
       "}\n",
       ".master-wrapper {\n",
       "  display: flex;\n",
       "  flex-flow: row nowrap;\n",
       "  justify-content: flex-start;\n",
       "  align-items: stretch;\n",
       "}\n",
       ".master-tbl {\n",
       "  flex: 3\n",
       "}\n",
       ".master-wrapper > div {\n",
       "  margin: 4px;\n",
       "  padding: 10px;\n",
       "}\n",
       "iframe.fileview {\n",
       "  border: 0 none;\n",
       "  height: 100%;\n",
       "  width: 100%;\n",
       "  white-space: pre-wrap;\n",
       "}\n",
       ".pane-header-title {\n",
       "  width: 80%;\n",
       "  font-weight: 500;\n",
       "}\n",
       ".pane-header {\n",
       "  line-height: 1;\n",
       "  background-color: #ffe6cc;\n",
       "  padding: 3px;\n",
       "}\n",
       ".pane-header .close {\n",
       "  font-size: 20px;\n",
       "  font-weight: 700;\n",
       "  float: right;\n",
       "  margin-top: -5px;\n",
       "}\n",
       ".master-wrapper .right-pane {\n",
       "  border: 1px inset silver;\n",
       "  width: 40%;\n",
       "  min-height: 300px;\n",
       "  flex: 3\n",
       "  min-width: 500px;\n",
       "}\n",
       ".master-wrapper * {\n",
       "  box-sizing: border-box;\n",
       "}\n",
       "</style><script>\n",
       "function copyToClipboard(fld) {\n",
       "    if (document.queryCommandSupported && document.queryCommandSupported('copy')) {\n",
       "        var textarea = document.createElement('textarea');\n",
       "        textarea.textContent = fld.innerHTML;\n",
       "        textarea.style.position = 'fixed';\n",
       "        document.body.appendChild(textarea);\n",
       "        textarea.select();\n",
       "\n",
       "        try {\n",
       "            return document.execCommand('copy'); // Security exception may be thrown by some browsers.\n",
       "        } catch (ex) {\n",
       "\n",
       "        } finally {\n",
       "            document.body.removeChild(textarea);\n",
       "        }\n",
       "    }\n",
       "}\n",
       "function expandPanel(el) {\n",
       "  const panelName = \"#\" + el.getAttribute('paneName');\n",
       "  console.log(el.title);\n",
       "\n",
       "  document.querySelector(panelName + \"-title\").innerHTML = el.title\n",
       "  iframe = document.querySelector(panelName + \"-body\");\n",
       "  \n",
       "  const tblcss = `<style> body { font-family: Arial, Helvetica, sans-serif;}\n",
       "    #csv { margin-bottom: 15px; }\n",
       "    #csv table { border-collapse: collapse;}\n",
       "    #csv table td { padding: 4px 8px; border: 1px solid silver;} </style>`;\n",
       "\n",
       "  function csvToHtmlTable(str) {\n",
       "    return '<div id=\"csv\"><table><tr><td>' +  str.replace(/[\\n\\r]+$/g, '').replace(/[\\n\\r]+/g, '</td></tr><tr><td>')\n",
       "      .replace(/,/g, '</td><td>') + '</td></tr></table></div>';\n",
       "  }\n",
       "  \n",
       "  function reqListener () {\n",
       "    if (el.title.endsWith(\".csv\")) {\n",
       "      iframe.setAttribute(\"srcdoc\", tblcss + csvToHtmlTable(this.responseText));\n",
       "    } else {\n",
       "      iframe.setAttribute(\"srcdoc\", this.responseText);\n",
       "    }  \n",
       "    console.log(this.responseText);\n",
       "  }\n",
       "\n",
       "  const oReq = new XMLHttpRequest();\n",
       "  oReq.addEventListener(\"load\", reqListener);\n",
       "  oReq.open(\"GET\", el.title);\n",
       "  oReq.send();\n",
       "  \n",
       "  \n",
       "  //iframe.src = el.title;\n",
       "  const resultPane = document.querySelector(panelName + \"-pane\");\n",
       "  if (resultPane.classList.contains(\"hidden\")) {\n",
       "    resultPane.classList.remove(\"hidden\");\n",
       "  }\n",
       "}\n",
       "function closePanel(el) {\n",
       "  const panelName = \"#\" + el.getAttribute('paneName')\n",
       "  const resultPane = document.querySelector(panelName + \"-pane\");\n",
       "  if (!resultPane.classList.contains(\"hidden\")) {\n",
       "    resultPane.classList.add(\"hidden\");\n",
       "  }\n",
       "}\n",
       "\n",
       "</script>\n",
       "<div class=\"master-wrapper\">\n",
       "  <div class=\"block master-tbl\"><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>project</th>\n",
       "      <th>uid</th>\n",
       "      <th>iter</th>\n",
       "      <th>start</th>\n",
       "      <th>state</th>\n",
       "      <th>name</th>\n",
       "      <th>labels</th>\n",
       "      <th>inputs</th>\n",
       "      <th>parameters</th>\n",
       "      <th>results</th>\n",
       "      <th>artifacts</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>default</td>\n",
       "      <td><div title=\"29ae7c944a184de881acc81206a92a48\"><a href=\"https://mlrun-ui.default-tenant.app.flobdnzhgtgt.iguazio-cd2.com/projects/default/jobs/29ae7c944a184de881acc81206a92a48/info\" target=\"_blank\" >...06a92a48</a></div></td>\n",
       "      <td>0</td>\n",
       "      <td>Jun 14 13:30:21</td>\n",
       "      <td>completed</td>\n",
       "      <td>gen_outliers</td>\n",
       "      <td><div class=\"dictlist\">v3io_user=admin</div><div class=\"dictlist\">kind=handler</div><div class=\"dictlist\">owner=admin</div><div class=\"dictlist\">host=jupyter-7b44c8d958-kklf7</div></td>\n",
       "      <td></td>\n",
       "      <td><div class=\"dictlist\">nrows=8192</div><div class=\"dictlist\">label_type=float</div></td>\n",
       "      <td></td>\n",
       "      <td><div class=\"artifact\" onclick=\"expandPanel(this)\" paneName=\"resultf958e165\" title=\"/files/artifacts/xgb_custom.csv\">xgb-outs</div></td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div></div>\n",
       "  <div id=\"resultf958e165-pane\" class=\"right-pane block hidden\">\n",
       "    <div class=\"pane-header\">\n",
       "      <span id=\"resultf958e165-title\" class=\"pane-header-title\">Title</span>\n",
       "      <span onclick=\"closePanel(this)\" paneName=\"resultf958e165\" class=\"close clickable\">&times;</span>\n",
       "    </div>\n",
       "    <iframe class=\"fileview\" id=\"resultf958e165-body\"></iframe>\n",
       "  </div>\n",
       "</div>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "to track results use .show() or .logs() or in CLI: \n",
      "!mlrun get run 29ae7c944a184de881acc81206a92a48 --project default , !mlrun logs 29ae7c944a184de881acc81206a92a48 --project default\n",
      "[mlrun] 2020-06-14 13:30:22,218 run executed, status=completed\n"
     ]
    }
   ],
   "source": [
    "gen_outs_tsk = NewTask(name='gen_outliers',\n",
    "                       handler=gen_outliers, \n",
    "                       params={'nrows': 8192, 'label_type': 'float'})\n",
    "\n",
    "outliers_run = run_local(gen_outs_tsk)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[mlrun] 2020-06-14 13:30:23,011 starting run fit model uid=489d2d26007e444cb81ffe7b6bc9d4a2  -> http://mlrun-api:8080\n",
      "[mlrun] 2020-06-14 13:30:23,405 log artifact test-set at /User/artifacts/test-set.csv, size: 689366, db: Y\n",
      "[mlrun] 2020-06-14 13:30:23,545 log artifact model at /User/artifacts/tttt/, size: 17052, db: Y\n",
      "[mlrun] 2020-06-14 13:30:23,712 log artifact learning-curves at /User/artifacts/plots/learning-curves.html, size: 31641, db: Y\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<style> \n",
       ".dictlist {\n",
       "  background-color: #b3edff; \n",
       "  text-align: center; \n",
       "  margin: 4px; \n",
       "  border-radius: 3px; padding: 0px 3px 1px 3px; display: inline-block;}\n",
       ".artifact {\n",
       "  cursor: pointer; \n",
       "  background-color: #ffe6cc; \n",
       "  text-align: left; \n",
       "  margin: 4px; border-radius: 3px; padding: 0px 3px 1px 3px; display: inline-block;\n",
       "}\n",
       "div.block.hidden {\n",
       "  display: none;\n",
       "}\n",
       ".clickable {\n",
       "  cursor: pointer;\n",
       "}\n",
       ".ellipsis {\n",
       "  display: inline-block;\n",
       "  max-width: 60px;\n",
       "  white-space: nowrap;\n",
       "  overflow: hidden;\n",
       "  text-overflow: ellipsis;\n",
       "}\n",
       ".master-wrapper {\n",
       "  display: flex;\n",
       "  flex-flow: row nowrap;\n",
       "  justify-content: flex-start;\n",
       "  align-items: stretch;\n",
       "}\n",
       ".master-tbl {\n",
       "  flex: 3\n",
       "}\n",
       ".master-wrapper > div {\n",
       "  margin: 4px;\n",
       "  padding: 10px;\n",
       "}\n",
       "iframe.fileview {\n",
       "  border: 0 none;\n",
       "  height: 100%;\n",
       "  width: 100%;\n",
       "  white-space: pre-wrap;\n",
       "}\n",
       ".pane-header-title {\n",
       "  width: 80%;\n",
       "  font-weight: 500;\n",
       "}\n",
       ".pane-header {\n",
       "  line-height: 1;\n",
       "  background-color: #ffe6cc;\n",
       "  padding: 3px;\n",
       "}\n",
       ".pane-header .close {\n",
       "  font-size: 20px;\n",
       "  font-weight: 700;\n",
       "  float: right;\n",
       "  margin-top: -5px;\n",
       "}\n",
       ".master-wrapper .right-pane {\n",
       "  border: 1px inset silver;\n",
       "  width: 40%;\n",
       "  min-height: 300px;\n",
       "  flex: 3\n",
       "  min-width: 500px;\n",
       "}\n",
       ".master-wrapper * {\n",
       "  box-sizing: border-box;\n",
       "}\n",
       "</style><script>\n",
       "function copyToClipboard(fld) {\n",
       "    if (document.queryCommandSupported && document.queryCommandSupported('copy')) {\n",
       "        var textarea = document.createElement('textarea');\n",
       "        textarea.textContent = fld.innerHTML;\n",
       "        textarea.style.position = 'fixed';\n",
       "        document.body.appendChild(textarea);\n",
       "        textarea.select();\n",
       "\n",
       "        try {\n",
       "            return document.execCommand('copy'); // Security exception may be thrown by some browsers.\n",
       "        } catch (ex) {\n",
       "\n",
       "        } finally {\n",
       "            document.body.removeChild(textarea);\n",
       "        }\n",
       "    }\n",
       "}\n",
       "function expandPanel(el) {\n",
       "  const panelName = \"#\" + el.getAttribute('paneName');\n",
       "  console.log(el.title);\n",
       "\n",
       "  document.querySelector(panelName + \"-title\").innerHTML = el.title\n",
       "  iframe = document.querySelector(panelName + \"-body\");\n",
       "  \n",
       "  const tblcss = `<style> body { font-family: Arial, Helvetica, sans-serif;}\n",
       "    #csv { margin-bottom: 15px; }\n",
       "    #csv table { border-collapse: collapse;}\n",
       "    #csv table td { padding: 4px 8px; border: 1px solid silver;} </style>`;\n",
       "\n",
       "  function csvToHtmlTable(str) {\n",
       "    return '<div id=\"csv\"><table><tr><td>' +  str.replace(/[\\n\\r]+$/g, '').replace(/[\\n\\r]+/g, '</td></tr><tr><td>')\n",
       "      .replace(/,/g, '</td><td>') + '</td></tr></table></div>';\n",
       "  }\n",
       "  \n",
       "  function reqListener () {\n",
       "    if (el.title.endsWith(\".csv\")) {\n",
       "      iframe.setAttribute(\"srcdoc\", tblcss + csvToHtmlTable(this.responseText));\n",
       "    } else {\n",
       "      iframe.setAttribute(\"srcdoc\", this.responseText);\n",
       "    }  \n",
       "    console.log(this.responseText);\n",
       "  }\n",
       "\n",
       "  const oReq = new XMLHttpRequest();\n",
       "  oReq.addEventListener(\"load\", reqListener);\n",
       "  oReq.open(\"GET\", el.title);\n",
       "  oReq.send();\n",
       "  \n",
       "  \n",
       "  //iframe.src = el.title;\n",
       "  const resultPane = document.querySelector(panelName + \"-pane\");\n",
       "  if (resultPane.classList.contains(\"hidden\")) {\n",
       "    resultPane.classList.remove(\"hidden\");\n",
       "  }\n",
       "}\n",
       "function closePanel(el) {\n",
       "  const panelName = \"#\" + el.getAttribute('paneName')\n",
       "  const resultPane = document.querySelector(panelName + \"-pane\");\n",
       "  if (!resultPane.classList.contains(\"hidden\")) {\n",
       "    resultPane.classList.add(\"hidden\");\n",
       "  }\n",
       "}\n",
       "\n",
       "</script>\n",
       "<div class=\"master-wrapper\">\n",
       "  <div class=\"block master-tbl\"><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>project</th>\n",
       "      <th>uid</th>\n",
       "      <th>iter</th>\n",
       "      <th>start</th>\n",
       "      <th>state</th>\n",
       "      <th>name</th>\n",
       "      <th>labels</th>\n",
       "      <th>inputs</th>\n",
       "      <th>parameters</th>\n",
       "      <th>results</th>\n",
       "      <th>artifacts</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>default</td>\n",
       "      <td><div title=\"489d2d26007e444cb81ffe7b6bc9d4a2\"><a href=\"https://mlrun-ui.default-tenant.app.flobdnzhgtgt.iguazio-cd2.com/projects/default/jobs/489d2d26007e444cb81ffe7b6bc9d4a2/info\" target=\"_blank\" >...6bc9d4a2</a></div></td>\n",
       "      <td>0</td>\n",
       "      <td>Jun 14 13:30:23</td>\n",
       "      <td>completed</td>\n",
       "      <td>fit model</td>\n",
       "      <td><div class=\"dictlist\">v3io_user=admin</div><div class=\"dictlist\">kind=handler</div><div class=\"dictlist\">owner=admin</div><div class=\"dictlist\">host=jupyter-7b44c8d958-kklf7</div></td>\n",
       "      <td><div title=\"store://default/gen_outliers_xgb-outs#29ae7c944a184de881acc81206a92a48\">dataset</div></td>\n",
       "      <td><div class=\"dictlist\">num_boost_round=40</div><div class=\"dictlist\">verbose_eval=False</div><div class=\"dictlist\">XGB_max_depth=2</div><div class=\"dictlist\">XGB_subsample=0.9</div></td>\n",
       "      <td></td>\n",
       "      <td><div class=\"artifact\" onclick=\"expandPanel(this)\" paneName=\"result7d21aacc\" title=\"/files/artifacts/test-set.csv\">test-set</div><div title=\"/User/artifacts/tttt\">model</div><div class=\"artifact\" onclick=\"expandPanel(this)\" paneName=\"result7d21aacc\" title=\"/files/artifacts/plots/learning-curves.html\">learning-curves</div></td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div></div>\n",
       "  <div id=\"result7d21aacc-pane\" class=\"right-pane block hidden\">\n",
       "    <div class=\"pane-header\">\n",
       "      <span id=\"result7d21aacc-title\" class=\"pane-header-title\">Title</span>\n",
       "      <span onclick=\"closePanel(this)\" paneName=\"result7d21aacc\" class=\"close clickable\">&times;</span>\n",
       "    </div>\n",
       "    <iframe class=\"fileview\" id=\"result7d21aacc-body\"></iframe>\n",
       "  </div>\n",
       "</div>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "to track results use .show() or .logs() or in CLI: \n",
      "!mlrun get run 489d2d26007e444cb81ffe7b6bc9d4a2 --project default , !mlrun logs 489d2d26007e444cb81ffe7b6bc9d4a2 --project default\n",
      "[mlrun] 2020-06-14 13:30:23,790 run executed, status=completed\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# THIS IS SETUP SO THAT YOU CAN COMPARE THE RESULTS AND SEE THAT THEY ARE EXACT.  \n",
    "# UNCOMMENT LINE `DEBUG_ERROR` AT TOP OF THIS NOTEBOOK TO TEST THAT THESE ARE \n",
    "# TRULY CALCULATED VALUES\n",
    "\n",
    "fit_tsk = NewTask(\n",
    "    name='fit model', \n",
    "    handler=fit,\n",
    "    params={\"num_boost_round\" : 40, \n",
    "            \"verbose_eval\"    : False,\n",
    "            \"XGB_max_depth\"   : 2,\n",
    "            \"XGB_subsample\"   : 0.9}\n",
    ")\n",
    "\n",
    "fit_run = run_local(fit_tsk, inputs={\"dataset\":outliers_run.outputs[\"xgb-outs\"]})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# export"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[mlrun] 2020-06-14 13:30:50,033 function spec saved to path: function.yaml\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<mlrun.runtimes.kubejob.KubejobRuntime at 0x7f59845b33c8>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from mlrun import code_to_function\n",
    "from mlrun.platforms.other import auto_mount\n",
    "\n",
    "gpus = False\n",
    "\n",
    "# create job function object from notebook code\n",
    "fn_params = {\n",
    "    \"name\"        : \"xgb_custom\",\n",
    "    \"handler\"     : \"fit\",\n",
    "    \"kind\"        : \"job\",\n",
    "    \"image\"       : \"mlrun/ml-models\" if not gpus else \"mlrun/ml-models-gpu\",\n",
    "    \"description\" : \"train an xgboost model using the low-level api\",\n",
    "    \"categories\"  : [\"analysis\"],\n",
    "    \"labels\"      : {\"author\": \"yjb\"}\n",
    "}\n",
    "\n",
    "xgb_fn = code_to_function(**fn_params)\n",
    "\n",
    "xgb_fn.export(\"function.yaml\")\n",
    "xgb_fn.apply(auto_mount())"
   ]
  }
 ],
 "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": "3.6.8"
  },
  "toc-autonumbering": false,
  "toc-showcode": false,
  "toc-showmarkdowntxt": false,
  "toc-showtags": false
 },
 "nbformat": 4,
 "nbformat_minor": 4
}