pytorch_predictor_hyperparameter.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true,
    "pycharm": {
     "name": "#%% md\n"
    },
    "id": "g3hAeYMcVtK3"
   },
   "source": [
    "## Example curve prediction script\n",
    "In this notebook we will try to predict the function coefficients."
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "!pip install --quiet ray pickle5\n",
    "from google.colab import drive\n",
    "drive.mount('/content/gdrive/')\n",
    "!cp -r /content/gdrive/MyDrive/Colab\\ Notebooks/otfb-lhc-ml/* ./\n",
    "import pickle5 as pickle"
   ],
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "j_ZiM72EWIJG",
    "outputId": "e8a10d28-7eb4-4b49-f6f2-b342a32c0f6a",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 1,
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "\u001B[K     |████████████████████████████████| 54.5 MB 57.9 MB/s \n",
      "\u001B[K     |████████████████████████████████| 256 kB 47.9 MB/s \n",
      "\u001B[K     |████████████████████████████████| 144 kB 37.5 MB/s \n",
      "\u001B[K     |████████████████████████████████| 8.8 MB 39.3 MB/s \n",
      "\u001B[K     |████████████████████████████████| 4.1 MB 32.8 MB/s \n",
      "\u001B[K     |████████████████████████████████| 461 kB 34.4 MB/s \n",
      "\u001B[?25hMounted at /content/gdrive/\n"
     ]
    }
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ],
   "metadata": {
    "id": "cPxYkyDim6va",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 11,
   "outputs": []
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "id": "D37NGt9bVtK4"
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import os\n",
    "from matplotlib import pyplot as plt\n",
    "from scipy.optimize import curve_fit\n",
    "from prettytable import PrettyTable\n",
    "\n",
    "from generate_TF import GenerateTF\n",
    "\n",
    "# from sklearn.model_selection import RepeatedKFold\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn import preprocessing\n",
    "\n",
    "import torch\n",
    "import torchvision.transforms as transforms\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "import ray\n",
    "from ray import tune\n",
    "from ray.tune.schedulers import ASHAScheduler\n",
    "from ray.tune import CLIReporter\n",
    "from pytorchClassifiers import Net\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 424
    },
    "id": "kg3bVRqCVtK6",
    "outputId": "c4685760-a68c-4cbf-da7b-cdc2566c11d6"
   },
   "outputs": [
    {
     "data": {
      "text/plain": "    phi      g_oo                                                  x  \\\n0 -20.0  0.001000  [-1619443.0, -1614543.0211800302, -1609643.042...   \n1 -20.0  0.001088  [-1619443.0, -1614543.0211800302, -1609643.042...   \n2 -20.0  0.001185  [-1619443.0, -1614543.0211800302, -1609643.042...   \n3 -20.0  0.001289  [-1619443.0, -1614543.0211800302, -1609643.042...   \n4 -20.0  0.001403  [-1619443.0, -1614543.0211800302, -1609643.042...   \n\n                                                   y  \n0  [-19.88111092306877, -24.65165651883909, -18.7...  \n1  [-22.868232696282362, -17.698964676040426, -16...  \n2  [-17.300864575069816, -25.87749055893209, -28....  \n3  [-15.239813657422143, -28.608066399706654, -24...  \n4  [-18.40240405493654, -20.57374582325774, -24.8...  ",
      "text/html": "<div>\n<style scoped>\n    .dataframe tbody tr th:only-of-type {\n        vertical-align: middle;\n    }\n\n    .dataframe tbody tr th {\n        vertical-align: top;\n    }\n\n    .dataframe thead th {\n        text-align: right;\n    }\n</style>\n<table border=\"1\" class=\"dataframe\">\n  <thead>\n    <tr style=\"text-align: right;\">\n      <th></th>\n      <th>phi</th>\n      <th>g_oo</th>\n      <th>x</th>\n      <th>y</th>\n    </tr>\n  </thead>\n  <tbody>\n    <tr>\n      <th>0</th>\n      <td>-20.0</td>\n      <td>0.001000</td>\n      <td>[-1619443.0, -1614543.0211800302, -1609643.042...</td>\n      <td>[-19.88111092306877, -24.65165651883909, -18.7...</td>\n    </tr>\n    <tr>\n      <th>1</th>\n      <td>-20.0</td>\n      <td>0.001088</td>\n      <td>[-1619443.0, -1614543.0211800302, -1609643.042...</td>\n      <td>[-22.868232696282362, -17.698964676040426, -16...</td>\n    </tr>\n    <tr>\n      <th>2</th>\n      <td>-20.0</td>\n      <td>0.001185</td>\n      <td>[-1619443.0, -1614543.0211800302, -1609643.042...</td>\n      <td>[-17.300864575069816, -25.87749055893209, -28....</td>\n    </tr>\n    <tr>\n      <th>3</th>\n      <td>-20.0</td>\n      <td>0.001289</td>\n      <td>[-1619443.0, -1614543.0211800302, -1609643.042...</td>\n      <td>[-15.239813657422143, -28.608066399706654, -24...</td>\n    </tr>\n    <tr>\n      <th>4</th>\n      <td>-20.0</td>\n      <td>0.001403</td>\n      <td>[-1619443.0, -1614543.0211800302, -1609643.042...</td>\n      <td>[-18.40240405493654, -20.57374582325774, -24.8...</td>\n    </tr>\n  </tbody>\n</table>\n</div>"
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# with open('data/transfer-functions-82000.pkl', \"rb\") as fh:\n",
    "#     raw_data = pickle.load(fh)\n",
    "# Load the data\n",
    "# df = pd.DataFrame(raw_data)\n",
    "df = pd.read_pickle('./data/transfer-functions-82000.pkl')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "\n",
    "phi_encoder = preprocessing.LabelEncoder()\n",
    "df['phi_label'] = phi_encoder.fit_transform(df['phi'])\n",
    "print(phi_encoder.classes_)\n",
    "\n",
    "gain_encoder = preprocessing.LabelEncoder()\n",
    "df['g_oo_label'] = gain_encoder.fit_transform(df['g_oo'])\n",
    "print(gain_encoder.classes_)\n",
    "\n",
    "print(df.dtypes)\n"
   ],
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "nRj9ku5GYsZ2",
    "outputId": "2fb6e580-b242-45f8-e98a-b261a8a6de4c",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 3,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[-20. -19. -18. -17. -16. -15. -14. -13. -12. -11. -10.  -9.  -8.  -7.\n",
      "  -6.  -5.  -4.  -3.  -2.  -1.   0.   1.   2.   3.   4.   5.   6.   7.\n",
      "   8.   9.  10.  11.  12.  13.  14.  15.  16.  17.  18.  19.  20.]\n",
      "[0.001      0.0010884  0.00118461 0.00128933 0.0014033  0.00152735\n",
      " 0.00166237 0.00180932 0.00196926 0.00214334 0.00233281 0.00253902\n",
      " 0.00276347 0.00300776 0.00327364 0.00356302 0.00387799 0.00422079\n",
      " 0.00459391 0.005     ]\n",
      "phi           float64\n",
      "g_oo          float64\n",
      "x              object\n",
      "y              object\n",
      "phi_label       int64\n",
      "g_oo_label      int64\n",
      "dtype: object\n"
     ]
    }
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 432x288 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Visualize the data\n",
    "for index, row in df.sample(n=3).iterrows():\n",
    "    y = row['y']\n",
    "    x = row['x']\n",
    "    plt.plot(x, y)\n",
    "plt.show()"
   ],
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 278
    },
    "id": "gARJOZXKVtK6",
    "outputId": "96f9e684-316f-47f7-daa4-b1cb22384176"
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "id": "x13UDFMxVtK7"
   },
   "outputs": [],
   "source": [
    "# Extract the target variables\n",
    "param_a_label = df.pop('phi_label')\n",
    "param_b_label = df.pop('g_oo_label')\n",
    "\n",
    "param_a = df.pop('phi')\n",
    "param_b = df.pop('g_oo')\n",
    "\n",
    "# All x should be equal\n",
    "x = df.iloc[0].x.astype(np.float32)\n",
    "df.drop(columns='x', inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "DcAQMh1qVtK8",
    "outputId": "7a1a8082-df5c-48c5-f43c-5d619e47a9ed"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[ 0,  5],\n",
      "        [ 0,  6],\n",
      "        [ 0,  7],\n",
      "        [ 0,  8],\n",
      "        [ 0,  9],\n",
      "        [ 0, 10],\n",
      "        [ 0, 11],\n",
      "        [ 0, 12],\n",
      "        [ 0, 13],\n",
      "        [ 0, 14]], dtype=torch.uint8)\n"
     ]
    }
   ],
   "source": [
    "# target is the target vector\n",
    "target = np.array((param_a_label, param_b_label), dtype=np.uint8).T\n",
    "\n",
    "target_original = np.array((param_a, param_b)).T\n",
    "\n",
    "target = torch.tensor(target, dtype=torch.uint8)\n",
    "print(target[5:15])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "RlaDD98GVtK9",
    "outputId": "4bd40931-28b1-4ec5-aa63-544e1ed20db2"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(82000, 662)\n",
      "<class 'numpy.ndarray'>\n"
     ]
    }
   ],
   "source": [
    "values = np.zeros((len(df), len(df.iloc[0, 0])), dtype=np.float32)\n",
    "# print(values.shape)\n",
    "index = 0\n",
    "for _, row in df.iterrows():\n",
    "    values[index, :] = row.values[0]\n",
    "    index += 1\n",
    "print(values.shape)\n",
    "print(type(values))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "0cT_JvJIVtK-",
    "outputId": "d519e14a-bf5d-4b5c-8a55-17ca47b54e66"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[-0.5378, -0.8267, -0.2371,  ..., -0.1169, -0.1908, -1.1691],\n",
      "        [-0.8251, -0.1174, -0.0605,  ..., -0.5932, -1.5411, -1.1238],\n",
      "        [-0.2897, -0.9517, -1.2612,  ..., -0.4935, -0.5153, -0.2053],\n",
      "        ...,\n",
      "        [ 2.1191,  1.8198,  1.9688,  ...,  2.6293,  2.1272,  2.4447],\n",
      "        [ 1.5838,  1.2171,  0.5005,  ...,  0.6547,  1.2669,  1.4611],\n",
      "        [ 1.1176,  0.9647,  1.1369,  ...,  1.5023,  1.1669,  0.6056]])\n"
     ]
    }
   ],
   "source": [
    "# Convert data to torch tensor\n",
    "data_scaler = preprocessing.StandardScaler().fit(values)\n",
    "scaled_values = data_scaler.transform(values)\n",
    "data = torch.tensor(scaled_values)\n",
    "print(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([65600, 662])\n",
      "torch.Size([13120, 662])\n",
      "torch.Size([3280, 662])\n",
      "torch.Size([65600, 2])\n",
      "torch.Size([13120, 2])\n",
      "torch.Size([3280, 2])\n"
     ]
    }
   ],
   "source": [
    "# Split in train and test\n",
    "X_train, X_test, y_train, y_test, y_train_original, y_test_original = train_test_split(\n",
    "    data, target, target_original, test_size=0.2, random_state=2)\n",
    "\n",
    "# further divide X_test in test + validate\n",
    "X_test, X_validate, y_test, y_validate, y_test_original, y_validate_original = train_test_split(\n",
    "    X_test, y_test, y_test_original, test_size=0.2, random_state=1)\n",
    "print(X_train.shape)\n",
    "print(X_test.shape)\n",
    "print(X_validate.shape)\n",
    "print(y_train.shape)\n",
    "print(y_test.shape)\n",
    "print(y_validate.shape)"
   ],
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "quk4_WrcVtK_",
    "outputId": "1e776c17-864d-4ca7-c6cf-5808048d9897"
   }
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "outputs": [],
   "source": [
    "# Initialize model, loss function and optimizer\n",
    "def train_model(config, X_train, y_train, X_test, y_test, name, epochs,\n",
    "                checkpoint_dir=None):\n",
    "    \n",
    "    # if 'stride' not in config and 'kernel_size' in config:\n",
    "    #     config['stride'] = config['kernel_size']\n",
    "\n",
    "    n_inputs = X_train.shape[1]\n",
    "    \n",
    "    # we will build one model per output target\n",
    "    n_outputs = len(np.unique(y_test))\n",
    "    model = Net(n_inputs, n_outputs, name=name,\n",
    "                with_batch_norm=True, save_best=False, \n",
    "                verbose=0, **config)\n",
    "\n",
    "    # specify loss function (regression)\n",
    "    criterion = nn.NLLLoss()\n",
    "\n",
    "    # specify optimizer\n",
    "    optimizer = torch.optim.Adam(model.parameters(), lr=0.005)\n",
    "\n",
    "    if checkpoint_dir:\n",
    "        model_state, optimizer_state = torch.load(\n",
    "            os.path.join(checkpoint_dir, \"checkpoint\"))\n",
    "        model.load_state_dict(model_state)\n",
    "        optimizer.load_state_dict(optimizer_state)\n",
    "    # print(model)\n",
    "\n",
    "\n",
    "    # train the network\n",
    "    for epoch in range(epochs):\n",
    "        train_loss, train_accuracy, test_loss, test_accuracy = model.train_net(\n",
    "            optimizer, criterion, X_train, y_train, X_test, y_test, 1)\n",
    "        \n",
    "        with tune.checkpoint_dir(epoch) as checkpoint_dir:\n",
    "            path = os.path.join(checkpoint_dir, \"checkpoint\")\n",
    "            torch.save((model.state_dict(), optimizer.state_dict()), path)\n",
    "\n",
    "        tune.report(accuracy=test_accuracy[-1], loss=test_loss[-1])\n",
    "\n",
    "    return model, train_loss, test_loss, train_accuracy, test_accuracy"
   ],
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "id": "QDWCDYWoVtLA"
   }
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-06-30 11:15:41,521\tINFO trial_runner.py:803 -- starting train_model_367ce_00000\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "== Status ==\n",
      "Current time: 2022-06-30 11:15:41 (running for 00:00:00.31)\n",
      "Memory usage on this node: 14.9/14.9 GiB: ***LOW MEMORY*** less than 10% of the memory on this node is available for use. This can cause unexpected crashes. Consider reducing the memory used by your application or reducing the Ray object store size by setting `object_store_memory` when calling `ray.init`.\n",
      "Using AsyncHyperBand: num_stopped=0\n",
      "Bracket: Iter 64.000: None | Iter 32.000: None | Iter 16.000: None | Iter 8.000: None | Iter 4.000: None | Iter 2.000: None | Iter 1.000: None\n",
      "Resources requested: 1.0/12 CPUs, 0/0 GPUs, 0.0/1.22 GiB heap, 0.0/0.61 GiB objects\n",
      "Result logdir: /home/kiliakis/ray_results/train_model_2022-06-30_11-15-41\n",
      "Number of trials: 16/50 (15 PENDING, 1 RUNNING)\n",
      "+-------------------------+----------+-----------------------+-----------------------------------+------------+------------+------------+--------------+\n",
      "| Trial name              | status   | loc                   | activation_func                   | avg_pool   |   hidden_1 |   hidden_2 |   trim_edges |\n",
      "|-------------------------+----------+-----------------------+-----------------------------------+------------+------------+------------+--------------|\n",
      "| train_model_367ce_00000 | RUNNING  | 128.141.159.40:944218 | <built-in function gelu>          | (6, 6)     |        520 |        220 |          135 |\n",
      "| train_model_367ce_00001 | PENDING  |                       | <built-in function gelu>          | (6, 6)     |        500 |        220 |          135 |\n",
      "| train_model_367ce_00002 | PENDING  |                       | <built-in function gelu>          | (6, 6)     |        480 |        180 |          135 |\n",
      "| train_model_367ce_00003 | PENDING  |                       | <built-in function gelu>          | (6, 6)     |        500 |        220 |          140 |\n",
      "| train_model_367ce_00004 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        480 |        220 |          130 |\n",
      "| train_model_367ce_00005 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        520 |        220 |          140 |\n",
      "| train_model_367ce_00006 | PENDING  |                       | <built-in function gelu>          | (6, 6)     |        500 |        220 |          135 |\n",
      "| train_model_367ce_00007 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        520 |        220 |          135 |\n",
      "| train_model_367ce_00008 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        500 |        200 |          140 |\n",
      "| train_model_367ce_00009 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        480 |        180 |          135 |\n",
      "| train_model_367ce_00010 | PENDING  |                       | <built-in function gelu>          | (6, 6)     |        520 |        180 |          130 |\n",
      "| train_model_367ce_00011 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        500 |        180 |          135 |\n",
      "| train_model_367ce_00012 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        480 |        220 |          140 |\n",
      "| train_model_367ce_00013 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        480 |        180 |          140 |\n",
      "| train_model_367ce_00014 | PENDING  |                       | <function relu at 0x7f9112867820> | (6, 6)     |        480 |        180 |          140 |\n",
      "| train_model_367ce_00015 | PENDING  |                       | <built-in function gelu>          | (6, 6)     |        500 |        180 |          135 |\n",
      "+-------------------------+----------+-----------------------+-----------------------------------+------------+------------+------------+--------------+\n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-06-30 11:15:47,046\tINFO trial_runner.py:803 -- starting train_model_367ce_00001\n"
     ]
    }
   ],
   "source": [
    "# configuring the search space\n",
    "config = {\n",
    "    \"activation_func\": tune.choice([F.gelu, F.relu]),\n",
    "    \"avg_pool\": tune.choice([(6,6)]),\n",
    "    \"hidden_1\": tune.choice([480, 500, 520]),\n",
    "    \"hidden_2\": tune.choice([180, 200, 220]),\n",
    "    # \"hidden_3\": tune.choice([0]),\n",
    "    \"trim_edges\": tune.choice([130, 135, 140]),\n",
    "}\n",
    "\n",
    "scheduler = ASHAScheduler(\n",
    "    max_t=100,\n",
    "    grace_period=1,\n",
    "    reduction_factor=2\n",
    ")\n",
    "\n",
    "reporter = CLIReporter(\n",
    "    # parameter_columns=[\"l1\", \"l2\", \"lr\", \"batch_size\"],\n",
    "    metric_columns=[\"loss\", \"accuracy\", \"training_iteration\"])\n",
    "\n",
    "result = tune.run(\n",
    "    tune.with_parameters(train_model,\n",
    "                         X_train=X_train, y_train=y_train[:,0].view(-1,),\n",
    "                         X_test=X_test, y_test=y_test[:,0].view(-1,),\n",
    "                         name='model1', epochs=50),\n",
    "    resources_per_trial={\"cpu\": 1},\n",
    "    config=config,\n",
    "    metric='accuracy',\n",
    "    mode='max',\n",
    "    num_samples=50,\n",
    "    scheduler=scheduler,\n",
    "    progress_reporter=reporter)\n"
   ],
   "metadata": {
    "pycharm": {
     "name": "#%%\n",
     "is_executing": true
    },
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "lm5AkGUDVtLC",
    "outputId": "90c7b50f-9fd0-4c66-d4c4-7cb30a0cf36d"
   }
  },
  {
   "cell_type": "code",
   "source": [
    "results_df = result.results_df.sort_values('accuracy', ascending=False)\n",
    "results_df.drop(columns=['time_this_iter_s', 'should_checkpoint', 'done', \n",
    "                        'timesteps_total', 'episodes_total', 'experiment_id', \n",
    "                        'time_since_restore', 'timesteps_since_restore', \n",
    "                        'iterations_since_restore', 'warmup_time', \n",
    "                        'experiment_tag'], inplace=True)\n",
    "results_df.drop(columns=['timestamp', 'time_total_s', 'pid', 'hostname', \n",
    "                         'node_ip'], inplace=True)\n",
    "print(results_df)"
   ],
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "h3LB6Kil7kvc",
    "outputId": "45d14b9b-1a7b-43fc-919a-7780c1e6c05e",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 89,
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "              accuracy      loss  training_iteration                 date  \\\n",
      "trial_id                                                                    \n",
      "ae718_00045  32.240854  1.787318                  50  2022-06-24_09-05-18   \n",
      "ae718_00000  31.257622  1.799332                  50  2022-06-24_09-00-43   \n",
      "ae718_00006  19.641768  2.352236                  16  2022-06-24_08-59-15   \n",
      "ae718_00008  14.283537  2.674124                   8  2022-06-24_08-59-48   \n",
      "ae718_00021  13.452744  2.728689                   8  2022-06-24_09-01-36   \n",
      "ae718_00023  13.140244  2.742292                   8  2022-06-24_09-01-46   \n",
      "ae718_00031  13.003049  2.727978                   8  2022-06-24_09-02-37   \n",
      "ae718_00043  12.881098  2.733721                   8  2022-06-24_09-03-22   \n",
      "ae718_00012  11.913110  2.789290                   8  2022-06-24_09-00-43   \n",
      "ae718_00017   8.399390  3.091795                   4  2022-06-24_09-01-06   \n",
      "ae718_00042   7.804878  3.155393                   4  2022-06-24_09-03-06   \n",
      "ae718_00011   7.782012  3.142257                   4  2022-06-24_09-00-12   \n",
      "ae718_00028   7.637195  3.149252                   4  2022-06-24_09-02-06   \n",
      "ae718_00029   7.507622  3.150575                   4  2022-06-24_09-02-07   \n",
      "ae718_00048   7.202744  3.143762                   4  2022-06-24_09-03-51   \n",
      "ae718_00001   6.730183  3.334598                   2  2022-06-24_08-57-51   \n",
      "ae718_00019   6.707317  3.359427                   2  2022-06-24_09-01-07   \n",
      "ae718_00040   6.554878  3.422982                   2  2022-06-24_09-02-51   \n",
      "ae718_00033   6.539634  3.326439                   2  2022-06-24_09-02-23   \n",
      "ae718_00005   6.509146  3.354213                   2  2022-06-24_08-58-15   \n",
      "ae718_00037   6.135671  3.367187                   2  2022-06-24_09-02-43   \n",
      "ae718_00014   6.128049  3.379769                   2  2022-06-24_09-00-51   \n",
      "ae718_00047   6.067073  3.427908                   2  2022-06-24_09-03-34   \n",
      "ae718_00041   6.059451  3.380683                   2  2022-06-24_09-02-53   \n",
      "ae718_00026   5.960366  3.368273                   2  2022-06-24_09-01-52   \n",
      "ae718_00002   5.823171  3.418684                   2  2022-06-24_08-57-59   \n",
      "ae718_00013   5.426829  3.666084                   1  2022-06-24_09-00-47   \n",
      "ae718_00020   5.419207  3.717121                   1  2022-06-24_09-01-11   \n",
      "ae718_00007   5.411585  3.740146                   1  2022-06-24_08-59-19   \n",
      "ae718_00039   5.381098  3.705697                   1  2022-06-24_09-02-45   \n",
      "ae718_00036   5.365854  3.713324                   1  2022-06-24_09-02-35   \n",
      "ae718_00027   5.365854  3.739639                   1  2022-06-24_09-01-50   \n",
      "ae718_00004   5.297256  3.732092                   1  2022-06-24_08-58-08   \n",
      "ae718_00044   5.205793  3.764152                   1  2022-06-24_09-03-10   \n",
      "ae718_00015   5.190549  3.706306                   1  2022-06-24_09-00-51   \n",
      "ae718_00032   5.160061  3.694118                   1  2022-06-24_09-02-15   \n",
      "ae718_00038   5.144817  3.680515                   1  2022-06-24_09-02-41   \n",
      "ae718_00046   5.106707  3.748698                   1  2022-06-24_09-03-26   \n",
      "ae718_00049   5.091463  3.661986                   1  2022-06-24_09-03-55   \n",
      "ae718_00024   5.068598  3.680593                   1  2022-06-24_09-01-40   \n",
      "ae718_00018   5.068598  3.743485                   1  2022-06-24_09-00-59   \n",
      "ae718_00003   5.060976  3.685821                   1  2022-06-24_08-58-03   \n",
      "ae718_00035   5.038110  3.728057                   1  2022-06-24_09-02-31   \n",
      "ae718_00016   4.984756  3.813178                   1  2022-06-24_09-00-55   \n",
      "ae718_00034   4.939024  3.706637                   1  2022-06-24_09-02-27   \n",
      "ae718_00010   4.916159  3.724045                   1  2022-06-24_08-59-56   \n",
      "ae718_00030   4.862805  3.727757                   1  2022-06-24_09-02-10   \n",
      "ae718_00009   4.763720  3.715804                   1  2022-06-24_08-59-52   \n",
      "ae718_00022   4.222561  3.745652                   1  2022-06-24_09-01-15   \n",
      "ae718_00025   4.214939  3.766903                   1  2022-06-24_09-01-44   \n",
      "\n",
      "               config.activation_func config.avg_pool  config.hidden_1  \\\n",
      "trial_id                                                                 \n",
      "ae718_00045  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00000  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00006  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00008  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00021  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00023  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00031  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00043  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00012  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00017  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00042  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00011  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00028  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00029  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00048  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00001  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00019  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00040  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00033  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00005  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00037  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00014  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00047  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00041  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00026  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00002  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00013  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00020  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00007  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00039  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00036  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00027  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00004  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00044  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00015  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00032  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00038  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00046  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00049  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00024  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00018  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00003  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00035  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00016  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00034  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00010  <built-in function gelu>          (7, 7)              500   \n",
      "ae718_00030  <built-in function gelu>          (5, 5)              500   \n",
      "ae718_00009  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00022  <built-in function gelu>          (6, 6)              500   \n",
      "ae718_00025  <built-in function gelu>          (6, 6)              500   \n",
      "\n",
      "             config.hidden_2  config.trim_edges  \n",
      "trial_id                                         \n",
      "ae718_00045              200                135  \n",
      "ae718_00000              200                138  \n",
      "ae718_00006              180                135  \n",
      "ae718_00008              180                138  \n",
      "ae718_00021              180                138  \n",
      "ae718_00023              210                138  \n",
      "ae718_00031              190                135  \n",
      "ae718_00043              180                138  \n",
      "ae718_00012              210                138  \n",
      "ae718_00017              190                135  \n",
      "ae718_00042              180                138  \n",
      "ae718_00011              200                135  \n",
      "ae718_00028              220                135  \n",
      "ae718_00029              180                133  \n",
      "ae718_00048              180                138  \n",
      "ae718_00001              220                133  \n",
      "ae718_00019              180                135  \n",
      "ae718_00040              220                133  \n",
      "ae718_00033              220                133  \n",
      "ae718_00005              190                135  \n",
      "ae718_00037              190                133  \n",
      "ae718_00014              190                138  \n",
      "ae718_00047              190                135  \n",
      "ae718_00041              210                135  \n",
      "ae718_00026              200                135  \n",
      "ae718_00002              200                133  \n",
      "ae718_00013              190                135  \n",
      "ae718_00020              200                135  \n",
      "ae718_00007              180                138  \n",
      "ae718_00039              200                133  \n",
      "ae718_00036              200                138  \n",
      "ae718_00027              210                135  \n",
      "ae718_00004              220                135  \n",
      "ae718_00044              200                135  \n",
      "ae718_00015              190                138  \n",
      "ae718_00032              180                133  \n",
      "ae718_00038              180                133  \n",
      "ae718_00046              190                135  \n",
      "ae718_00049              210                133  \n",
      "ae718_00024              180                135  \n",
      "ae718_00018              180                138  \n",
      "ae718_00003              190                133  \n",
      "ae718_00035              210                135  \n",
      "ae718_00016              190                138  \n",
      "ae718_00034              190                133  \n",
      "ae718_00010              220                138  \n",
      "ae718_00030              190                135  \n",
      "ae718_00009              200                135  \n",
      "ae718_00022              210                135  \n",
      "ae718_00025              200                135  \n"
     ]
    }
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "columns = results_df.columns.values.astype(str).tolist()\n",
    "values = results_df.values\n",
    "# values = np.sort(values, lambda a: -a[0])\n",
    "# Prepare the values\n",
    "values[:, 5] = np.apply_along_axis(lambda v: ''.join(\n",
    "    str(v[5])).replace('(', '').replace(')', ''), 1, values)\n",
    "values = values.astype(str)\n",
    "values[:, 4] = np.apply_along_axis(lambda e: e[4].replace(\n",
    "    '<built-in function ', '').replace('>', ''), 1, values)\n"
   ],
   "metadata": {
    "id": "I22z7pnQtioz",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 90,
   "outputs": []
  },
  {
   "cell_type": "code",
   "source": [
    "base_path = '/content/gdrive/MyDrive/Colab Notebooks/otfb-lhc-ml/hyperopt/'\n",
    "if not os.path.exists(base_path):\n",
    "    os.makedirs(base_path)\n",
    "\n",
    "import csv\n",
    "with open(os.path.join(base_path, f'{values[0, 3]}.csv'), 'w') as file:\n",
    "    writer = csv.writer(file)\n",
    "    writer.writerow(columns)\n",
    "    writer.writerows(values)\n"
   ],
   "metadata": {
    "id": "knVbu9_d4Gyc",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 91,
   "outputs": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "pycharm": {
     "name": "#%%\n"
    },
    "id": "gDnJjY1IVtLE"
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "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",
   "version": "3.9.7"
  },
  "colab": {
   "name": "pytorch-predictor-hyperparameter.ipynb",
   "provenance": [],
   "collapsed_sections": []
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}