draft.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import glob\n",
    "import seaborn as sns\n",
    "\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import json\n",
    "import copy\n",
    "import logging\n",
    "import os\n",
    "import random\n",
    "import re\n",
    "import subprocess\n",
    "from typing import Any, Dict, List\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from tf_agents.environments import py_environment\n",
    "from tf_agents.specs import array_spec\n",
    "from tf_agents.trajectories import time_step as ts\n",
    "from tf_agents.typing import types\n",
    "from tf_agents.policies.random_py_policy import RandomPyPolicy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "class JVMState:\n",
    "    def __init__(self, flags=[], params=[], goals=[]):\n",
    "        self._flags = flags\n",
    "        self._params = params\n",
    "        self._goals = goals\n",
    "    \n",
    "    def get_list(self):\n",
    "        return np.array(\n",
    "            [*self._flags, *self._params, *self._goals])\n",
    "    \n",
    "    def get_flags(self):\n",
    "        return self._flags\n",
    "    \n",
    "    def get_params(self):\n",
    "        return self._params\n",
    "    \n",
    "    def get_goals(self):\n",
    "        return self._goals\n",
    "    \n",
    "    def set_flags(self, flags):\n",
    "        self._flags = flags\n",
    "    \n",
    "    def set_params(self, params):\n",
    "        self._params = params\n",
    "    \n",
    "    def set_goals(self, goals):\n",
    "        self._goals = goals\n",
    "    \n",
    "        \n",
    "\n",
    "class JVMEnv(py_environment.PyEnvironment):\n",
    "\n",
    "    def __init__(\n",
    "        self,\n",
    "        jdk_path: str,\n",
    "        bm_path: str,\n",
    "        gc_viewer_jar: str,\n",
    "        callback_path: str,\n",
    "        goals: List,\n",
    "        flags: List,\n",
    "        params: List,\n",
    "        default_flags: List,\n",
    "        bm_name: str = \"avrora\",\n",
    "        n: int = 5,\n",
    "        verbose: bool = False,\n",
    "    ):\n",
    "\n",
    "        self.jdk = jdk_path\n",
    "        self.bm_path = bm_path\n",
    "        self.gc_viewer_jar = gc_viewer_jar\n",
    "        self.callback_path = callback_path\n",
    "        self._bm = bm_name\n",
    "        self._gc_log_file = f\"gc-{self._bm}.txt\"\n",
    "        self._n = n\n",
    "        self._goals = goals\n",
    "        self._flags = flags\n",
    "        self._params = params\n",
    "        self._default_flags = default_flags\n",
    "        self._verbose = verbose\n",
    "        print(\"jdk_path: \", jdk_path)\n",
    "        \n",
    "        self._env = os.environ.copy()\n",
    "        self._env[\"PATH\"] = f\"{self.jdk}:{self._env['PATH']}\"\n",
    "        \n",
    "        self._episode_ended = False\n",
    "\n",
    "        # ============= F L A G S =============\n",
    "        self._num_flags = len(self._flags)\n",
    "        self._num_params = len(self._params)\n",
    "        self._num_goals = len(self._goals)\n",
    "        # TODO: Find min and max values for flags\n",
    "        # self._goal_idx = self._num_flags\n",
    "        # self._flags = {\n",
    "        #     \"MaxTenuringThreshold\": {\"min\": 1, \"max\": 16, \"step\": 3},\n",
    "        #     \"ParallelGCThreads\": {\"min\": 4, \"max\": 24, \"step\": 4},\n",
    "        # }\n",
    "\n",
    "        # self._action_mapping = {\n",
    "        #     0: self._decrease_MaxTenuringThreshold,\n",
    "        #     1: self._increase_MaxTenuringThreshold,\n",
    "        #     2: self._decrease_ParallelGCThreads,\n",
    "        #     3: self._increase_ParallelGCThreads,\n",
    "        # }\n",
    "        # =====================================\n",
    "\n",
    "        # assert (\n",
    "        #     len(self._action_mapping) == 2 * self._num_flags\n",
    "        # ), \"Each flag should have 2 actions!\"\n",
    "\n",
    "        # self._flags_min_values = [self._flags[i][\"min\"] for i in self._flags.keys()]\n",
    "        # self._flags_max_values = [self._flags[i][\"max\"] for i in self._flags.keys()]\n",
    "        # self._flags_step_values = [self._flags[i][\"step\"] for i in self._flags.keys()]\n",
    "        \n",
    "        self._flags_min_values = [flag[\"min\"] for flag in self._flags]\n",
    "        self._flags_max_values = [flag[\"max\"] for flag in self._flags]\n",
    "        self._flags_step_values = [flag[\"step\"] for flag in self._flags]\n",
    "\n",
    "        self._action_spec = array_spec.BoundedArraySpec(\n",
    "            shape = (self._num_flags, ),\n",
    "            dtype = np.int32,\n",
    "            minimum = np.array(self._flags_min_values),\n",
    "            maximum = np.array(self._flags_max_values),\n",
    "            name = \"action\",\n",
    "        )\n",
    "\n",
    "        self._observation_spec = array_spec.BoundedArraySpec(\n",
    "            shape=(self._num_flags + self._num_goals + self._num_params,),  # 1 goal, {X} external vars\n",
    "            dtype=np.float32,\n",
    "            name=\"observation\",\n",
    "        )\n",
    "\n",
    "        # # For offline RL: if you already have a dataset file with trajectories.\n",
    "        self._new_df = pd.read_csv(f\"datasets/large_{self._bm}_real_saved_states.csv\")\n",
    "\n",
    "        self._default_state = self._get_default_state(mode=\"default\")\n",
    "        # self._default_state = self._get_default_state(mode=\"random\")\n",
    "\n",
    "        self._perf_states = {}\n",
    "        self._perf_states[0] = {\n",
    "            \"flags\": self._default_state.get_flags(),\n",
    "            \"goal\": self._default_state.get_goals(),\n",
    "            \"params\": self._default_state.get_params(),\n",
    "            \"count\": 1,\n",
    "        }\n",
    "\n",
    "        self._print_welcome_msg()\n",
    "\n",
    "    @property\n",
    "    def jdk(self):\n",
    "        return self._jdk\n",
    "\n",
    "    @jdk.setter\n",
    "    def jdk(self, path: str):\n",
    "        if not os.path.exists(path):\n",
    "            raise FileNotFoundError(path)\n",
    "        self._jdk = os.path.join(path, \"bin\")\n",
    "\n",
    "    @property\n",
    "    def bm_path(self):\n",
    "        return self._bm_path\n",
    "\n",
    "    @bm_path.setter\n",
    "    def bm_path(self, path: str):\n",
    "        if not os.path.exists(path):\n",
    "            raise FileNotFoundError(path)\n",
    "        self._bm_path = os.path.join(path)\n",
    "\n",
    "    @property\n",
    "    def gc_viewer_jar(self):\n",
    "        return self._gc_viewer_jar\n",
    "\n",
    "    @gc_viewer_jar.setter\n",
    "    def gc_viewer_jar(self, path: str):\n",
    "        if not os.path.exists(path):\n",
    "            raise FileNotFoundError(path)\n",
    "        self._gc_viewer_jar = os.path.join(path)\n",
    "\n",
    "    @property\n",
    "    def callback_path(self):\n",
    "        return self._callback_path\n",
    "\n",
    "    @callback_path.setter\n",
    "    def callback_path(self, path: str):\n",
    "        if not os.path.exists(path):\n",
    "            raise FileNotFoundError(path)\n",
    "        self._callback_path = os.path.join(path)\n",
    "\n",
    "\n",
    "    def action_spec(self):\n",
    "        \"\"\"Get the actions that should be provided to `step()`.\"\"\"\n",
    "        return self._action_spec\n",
    "\n",
    "    def observation_spec(self):\n",
    "        \"\"\"Get the the observations provided by the environment.\"\"\"\n",
    "        return self._observation_spec\n",
    "\n",
    "    def _reset(self):\n",
    "        \"\"\"\n",
    "        Resets the environment state.\n",
    "\n",
    "        This method must be called before :func:`step()`.\n",
    "\n",
    "        Returns:\n",
    "            A `TimeStep` namedtuple containing:\n",
    "                step_type: A `StepType` of `FIRST`.\n",
    "                reward: 0.0, indicating the reward.\n",
    "                discount: 1.0, indicating the discount.\n",
    "                observation: A NumPy array, or a nested dict,\n",
    "                list or tuple of arrays\n",
    "                corresponding to `observation_spec()`.\n",
    "        \"\"\"\n",
    "        self._episode_ended = False\n",
    "\n",
    "        # To ensure all elements within an object array are copied, use `copy.deepcopy`\n",
    "        self._default_state = self._get_default_state(mode=\"random\")\n",
    "        # self._default_state = self._get_default_state(mode=\"default\")\n",
    "        \n",
    "        self._state = copy.deepcopy(self._default_state)\n",
    "\n",
    "        logging.debug(\n",
    "            f\"[RESET] {self._get_info()}, target: {self._state[self._goal_idx]}\"\n",
    "        )\n",
    "        return ts.restart(np.array(self._state, dtype=np.float32))\n",
    "\n",
    "    def _step(self, action: types.NestedArray):\n",
    "        \"\"\"Updates the environment according to action.\n",
    "\n",
    "        Parameters:\n",
    "            action: A NumPy array, or a nested dict, list or tuple of arrays\n",
    "                corresponding to `action_spec()`.\n",
    "        Returns:\n",
    "            A `TimeStep` namedtuple containing:\n",
    "                step_type: A `StepType` of `FIRST`.\n",
    "                reward: 0.0, indicating the reward.\n",
    "                discount: 1.0, indicating the discount.\n",
    "                observation: A NumPy array, or a nested dict, list or tuple of arrays\n",
    "                corresponding to `observation_spec()`.\n",
    "        \"\"\"\n",
    "        if self._episode_ended:\n",
    "            # The last action ended the episode.\n",
    "            # Ignore the current action and start a new episode.\n",
    "            logging.debug(\n",
    "                f\"[EPISODE ENDED] {self._get_info()}, target: {self._state[self._goal_idx]}\"\n",
    "            )\n",
    "            return self.reset()\n",
    "        # Apply an action based on the mapping: decrease/increase <flag.value>.\n",
    "        self._action_mapping.get(int(action))()\n",
    "\n",
    "        # Make sure we don't leave the boundaries.\n",
    "        self._state = self._clip_state(self._state)\n",
    "\n",
    "        # Check if the current JVM configuration is cached.\n",
    "        # Add `state` to cache if new.\n",
    "        self._state = copy.deepcopy(\n",
    "            self._state_merging(self._state[: self._num_flags])\n",
    "        )\n",
    "\n",
    "        # Termination criteria\n",
    "        if self._state[self._goal_idx] <= self._default_state[self._goal_idx] * 0.04:\n",
    "            self._episode_ended = True\n",
    "\n",
    "        self._reward = self._get_reward(\n",
    "            current_state=self._state,\n",
    "            previous_state=self._default_state,\n",
    "            lower_is_better=True,\n",
    "            beta=0.0,\n",
    "        )  # ! No intrinsic reward\n",
    "\n",
    "        logging.debug(\n",
    "            f\"[STEP] {self._get_info()}, target: {self._state[self._goal_idx]}\"\n",
    "        )\n",
    "\n",
    "        if self._episode_ended:\n",
    "            return ts.termination(np.array(self._state, dtype=np.float32), reward=2.0)\n",
    "        else:\n",
    "            return ts.transition(\n",
    "                np.array(self._state, dtype=np.float32),\n",
    "                reward=self._reward,\n",
    "                discount=0.5,\n",
    "            )\n",
    "\n",
    "    def _state_merging(self, flags):\n",
    "        \"\"\"\n",
    "        Store states' JVM configurations and performance measurements\n",
    "        in a cache. The cache stores the JVM flags and the metric goal.\n",
    "        In each `self._step` iteration, the performance of the same\n",
    "        state is queried and retrieved directly from the cache instead\n",
    "        of re-running the benchmark utility.\n",
    "        Han, Xue & Yu, Tingting. (2020).\n",
    "        Automated Performance Tuning for Highly-Configurable Software Systems.\n",
    "\n",
    "        0: {\"flags\": [10, 10], \"goal\": [234], \"params\": [150, 0.1, 30, 19], \"count\": 1},\n",
    "        1: {\"flags\": [10, 20], \"goal\": [222], \"params\": [140, 0.2, 29, 18], \"count\": 4},\n",
    "        ...\n",
    "        \"\"\"\n",
    "        saved_states = [self._perf_states[i][\"flags\"] for i in self._perf_states.keys()]\n",
    "        state = JVMState()\n",
    "        if flags == self._default_state.get_flags():\n",
    "            state = self._default_state\n",
    "        elif flags in saved_states:\n",
    "            \"\"\"\n",
    "            If current state is stored in a cache,\n",
    "            update the state goal value.\n",
    "            \"\"\"\n",
    "            idx = saved_states.index(flags)\n",
    "            self._perf_states[idx][\"count\"] += 1\n",
    "            \n",
    "            goal = self._perf_states[idx][\"goal\"]\n",
    "            params = self._perf_states[idx][\"params\"]\n",
    "            \n",
    "            state = JVMState(flags, params, goal)\n",
    "        else:\n",
    "            \"\"\"\n",
    "            If current state is not stored in the cache,\n",
    "            measure the performance metric, and save it\n",
    "            in the cache.\n",
    "            \"\"\"\n",
    "            try:\n",
    "                last_index = list(self._perf_states.keys())[-1]\n",
    "            except IndexError:\n",
    "                # If `self._perf_states` is empty\n",
    "                last_index = -1\n",
    "                \n",
    "            # Launch a benchmark with a new JVM configuration\n",
    "            state = self._synthetic_run(flags)\n",
    "            \n",
    "            # Store a new state in the cache, count = 1.\n",
    "            self._perf_states[last_index + 1] = {\n",
    "                \"flags\": state.get_flags(),\n",
    "                \"goal\": state.get_goals(),\n",
    "                \"params\": state.get_params(),\n",
    "                \"count\": 1,\n",
    "            }\n",
    "            \n",
    "        assert len(state.get_list()) != 0, \"State is empty!\"\n",
    "        \n",
    "        return state\n",
    "\n",
    "    def _synthetic_run(self, flag_values: List[int]):\n",
    "        flag_names = [each[\"name\"] for each in self._flags]\n",
    "        goal_names = [each[\"name\"] for each in self._goals]\n",
    "        param_names = self._params\n",
    "        \n",
    "        # Select columns where flags equal to flag_values\n",
    "        param_values = (\n",
    "            self._new_df.set_index(flag_names)\n",
    "            .loc[tuple(flag_values), param_names]\n",
    "            .values\n",
    "        )\n",
    "\n",
    "        goal_values = (\n",
    "            self._new_df.set_index(flag_names)\n",
    "            .loc[tuple(flag_values), goal_names]\n",
    "            .values\n",
    "        )\n",
    "\n",
    "        state = JVMState(\n",
    "            flags=flag_values,\n",
    "            params=param_values,\n",
    "            goals=goal_values,\n",
    "        )\n",
    "        \n",
    "        return state\n",
    "\n",
    "    def _get_JVM_opt_value(self, opt: str):\n",
    "        \"\"\"\n",
    "        Get the default JVM option value from environment\n",
    "        by parsing java PrintFlagsFinal output.\n",
    "        Equals to bash command:\n",
    "        java -XX:+PrintFlagsFinal -version|grep \"$opt\"|\n",
    "\n",
    "        Parameters:\n",
    "        opt (str): JVM option name\n",
    "\n",
    "        Returns:\n",
    "        (int) Default JVM option value\n",
    "        \"\"\"\n",
    "        opt_value = None\n",
    "\n",
    "        flags = subprocess.check_output(\n",
    "            [\"java\", \"-XX:+PrintFlagsFinal\", \"-version\"], text=True, env=self._env\n",
    "        )\n",
    "\n",
    "        assert re.search(opt, flags), f\"Option {opt} was not found in JVM flags\"\n",
    "\n",
    "        for line in flags.split(\"\\n\"):\n",
    "            if re.search(opt, line):\n",
    "                opt_value = re.findall(\"\\d+\", line)[0]\n",
    "                # TODO: Handle boolean values\n",
    "                return int(opt_value)\n",
    "\n",
    "    def _get_info(self):\n",
    "        info = {}\n",
    "        flags = list(self._flags.keys())\n",
    "        for flag in flags:\n",
    "            flag_idx = flags.index(flag)\n",
    "            flag_value = self._state[flag_idx]\n",
    "            info[flag] = flag_value\n",
    "        return info\n",
    "\n",
    "    def _get_default_state(self, mode: str = \"default\"):\n",
    "        \"\"\"\n",
    "        Get default values for each JVM option stored in `self._flags`\n",
    "        from `java -XX:+PrintFlagsFinal -version` command output.\n",
    "        Also runs a benchmark with default JVM_OPTS and get the initial\n",
    "        goal value.\n",
    "\n",
    "        Parameters:\n",
    "        mode (str): The mode of generating JVM options (default/minimum/random)\n",
    "                    where\n",
    "                    - \"default\" sets JVM_OPTS default JDK options,\n",
    "                    - \"minimum\" sets JVM_OPTS minimum possible values,\n",
    "                    - \"random\" sets JVM_OPTS random values within options\n",
    "                    range.\n",
    "\n",
    "        Returns:\n",
    "        (np.array) The initial state of the Agent which stores the default\n",
    "        JVM_OPTS and its performance measurement.\n",
    "        \"\"\"\n",
    "        assert mode in [\"default\", \"random\"], f\"Unknown mode: {mode}\"\n",
    "        \n",
    "        run_flags = []\n",
    "        \n",
    "        if mode == \"default\":\n",
    "            run_flags = [7, 12]\n",
    "        if mode == \"random\":\n",
    "            for i in range(self._num_flags):\n",
    "                \n",
    "                min_val = self._flags_min_values[i]\n",
    "                max_val = self._flags_max_values[i]\n",
    "                step_val = self._flags_step_values[i]\n",
    "                \n",
    "                run_flags.append(\n",
    "                    random.randrange(\n",
    "                        start=min_val,\n",
    "                        stop=max_val + step_val,  # must be included\n",
    "                        step=step_val,\n",
    "                    )\n",
    "                )\n",
    "                \n",
    "        state = self._synthetic_run(run_flags)\n",
    "        return state\n",
    "\n",
    "    def _get_goal_value(self, jvm_opts: List[str] = []):\n",
    "        \"\"\"\n",
    "        Run the benchmark with default values and\n",
    "        get a start point of the goal.\n",
    "\n",
    "        Parameters:\n",
    "        jvm_opts (list): An array of JVM options.\n",
    "\n",
    "        Returns:\n",
    "        (int) A performance measurement (target metric\n",
    "        or goal value).\n",
    "        \"\"\"\n",
    "        # Run benchmark with default values\n",
    "        self._run(\n",
    "            jvm_opts,\n",
    "            self._gc_log_file,\n",
    "            self._bm,\n",
    "            self.bm_path,\n",
    "            self.callback_path,\n",
    "            self._n,\n",
    "        )\n",
    "\n",
    "        if os.path.exists(self._gc_log_file):\n",
    "            # Get goal value from first-time generated GC log\n",
    "            result = self._get_goal_from_file()\n",
    "            # Clean up\n",
    "            os.remove(self._gc_log_file)\n",
    "            return result\n",
    "        else:\n",
    "            raise Exception(\n",
    "                \"GC log file was not generated: please, ensure that benchmark runs correctly!\"\n",
    "            )\n",
    "\n",
    "    def _clip_state(self, state):\n",
    "        for i in range(self._num_flags):\n",
    "            \"\"\"\n",
    "            Iterate through each flag value in `state`\n",
    "            and use `np.clip` to make sure we don't leave\n",
    "            the boundaries\n",
    "            \"\"\"\n",
    "            min_value_i = self._flags_min_values[i]\n",
    "            max_value_i = self._flags_max_values[i]\n",
    "            state[i] = np.clip(state[i], min_value_i, max_value_i)\n",
    "        return state\n",
    "\n",
    "    def _get_goal_from_file(self):\n",
    "        sep = \";\"\n",
    "        summary = \"summary.csv\"\n",
    "        goal_value = None\n",
    "        subprocess.call(\n",
    "            [\n",
    "                \"java\",\n",
    "                \"-cp\",\n",
    "                self.gc_viewer_jar,\n",
    "                \"com.tagtraum.perf.gcviewer.GCViewer\",\n",
    "                self._gc_log_file,\n",
    "                summary,\n",
    "                \"-t\",\n",
    "                \"SUMMARY\",\n",
    "            ],\n",
    "            stdout=subprocess.DEVNULL,\n",
    "            stderr=subprocess.STDOUT,\n",
    "            text=True,\n",
    "            env=self._env,\n",
    "        )\n",
    "\n",
    "        assert os.path.exists(summary), f\"File {summary} does not exist\"\n",
    "\n",
    "        with open(summary, \"+r\") as summary_file:\n",
    "            for line in summary_file.readlines():\n",
    "                if self._goal + sep in line:\n",
    "                    goal_value = float(line.split(sep)[1])\n",
    "\n",
    "        if goal_value == None:\n",
    "            raise Exception(f\"Goal '{self._goal}' was not found in '{summary}'!\")\n",
    "\n",
    "        if os.path.exists(summary):\n",
    "            os.remove(summary)\n",
    "\n",
    "        return goal_value\n",
    "\n",
    "    def _get_reward(\n",
    "        self,\n",
    "        current_state: JVMState,\n",
    "        previous_state: JVMState,\n",
    "        lower_is_better: bool = False,\n",
    "        beta: float = 1.0,\n",
    "    ):\n",
    "        \"\"\"\n",
    "        Get the environment reward. The reward is composed of two terms:\n",
    "            `reward = reward_ex + beta * reward_in`,\n",
    "        where `beta` is a hyperparameter adjusting the balance between\n",
    "        exploitation and exploration.\n",
    "        - reward_ex is an extrinsic reward from the environment at time `t`.\n",
    "        - reward_in is an intrinsic exploration bonus.\n",
    "        Parameters:\n",
    "            current_state (np.array):   Current state containing JVM flags\n",
    "                                        and goal value.\n",
    "            previous_state (np.array):  Previous state containing JVM flags\n",
    "                                        and goal value.\n",
    "            lower_is_better (bool):     Whether to consider lower goal values\n",
    "                                        better than larger ones. So that reward\n",
    "                                        is positive.\n",
    "            beta (float):               Importance of instrinsic rewards ([0.0, 1.0]).\n",
    "        Returns:\n",
    "            reward (float):             An environment reward value at current\n",
    "                                        time step.\n",
    "        \"\"\"\n",
    "        coef = -1 if lower_is_better else 1\n",
    "        reward_in = 0  # Intrinsic reward.\n",
    "        reward_ex = 0  # Extrinsic reward.\n",
    "\n",
    "        # TODO: Check if beta is in range [0, 1]\n",
    "\n",
    "        assert len(current_state.get_goals()) == 1, \"Multi-objective is not supported\"\n",
    "\n",
    "        # JVMState supports multiple goal values,\n",
    "        # so we paramsct a single number from 1D array\n",
    "        # TODO: Add multi-objective support\n",
    "        current_goal = current_state.get_goals()[0]\n",
    "        previous_goal = previous_state.get_goals()[0]\n",
    "\n",
    "        if coef * current_goal <= previous_goal:\n",
    "            for i in self._perf_states.keys():\n",
    "                if self._perf_states[i][\"flags\"] == current_state.get_flags():\n",
    "                    # First, we add the state to cache with count=1,\n",
    "                    # but we haven't run the benchmark with it yet,\n",
    "                    # so it is fair to say that actually count is 0.\n",
    "                    count = self._perf_states[i][\"count\"] - 1\n",
    "                    reward_in = (count + 0.01) ** (-1 / 2)\n",
    "            reward_ex = coef * (current_goal - previous_goal) / previous_goal\n",
    "            reward = reward_ex + beta * reward_in\n",
    "            reward = round(reward, 4)\n",
    "        else:\n",
    "            reward = -1\n",
    "\n",
    "        return reward\n",
    "\n",
    "\n",
    "    def _get_jvm_opts(self, flags):\n",
    "        jvm_opts = []\n",
    "        for flag_name in self._flags.keys():\n",
    "            i = list(self._flags.keys()).index(flag_name)\n",
    "            # `state[:self._num_flags]` stores an array of current\n",
    "            # flag values [<MaxHeapSize>, <InitialHeapSize>]\n",
    "            flag_value = int(flags[i])\n",
    "            # TODO: Add a condition for flags with boolean values\n",
    "            # TODO: Add a unit!\n",
    "            jvm_opts.append(f\"-XX:{flag_name}={flag_value}\")\n",
    "        return jvm_opts\n",
    "\n",
    "    def _run(\n",
    "        self,\n",
    "        jvm_opts: List[str],\n",
    "        gc_log_file: str,\n",
    "        bm: str,\n",
    "        bm_path: str,\n",
    "        callback_path: str,\n",
    "        n: int = 5,\n",
    "        verbose: bool = False,\n",
    "    ):\n",
    "        \"\"\"\n",
    "        Run a benchmark with the specified JVM options\n",
    "        such as MaxHeapSize.\n",
    "\n",
    "        Parameters:\n",
    "        jvm_opts (array of str):  JVM options (e.g., MaxHeapSize, GCTimeRation,\n",
    "                                ParallelGCThreads, etc).\n",
    "        gc_log_file (str):        Path to the garbage collector log file (txt).\n",
    "        bm (str):                 DaCapo benchmark name.\n",
    "        bm_path (str):            DaCapo benchmark path.\n",
    "        n (int):                  Total number of benchmark's iterations.\n",
    "        verbose (bool):           Print debug messages.\n",
    "        \"\"\"\n",
    "\n",
    "        if not os.path.exists(callback_path):\n",
    "            raise FileNotFoundError(callback_path)\n",
    "\n",
    "        # Clean up before running the benchmark\n",
    "        try:\n",
    "            os.remove(gc_log_file)\n",
    "        except FileNotFoundError:\n",
    "            pass\n",
    "\n",
    "        # Default flags\n",
    "        jvm_opts.append(*self._default_flags)\n",
    "        # jvm_opts.append(\"-XX:+UseParallelGC\")\n",
    "        # jvm_opts.append(\"-Xmx16G\")\n",
    "        # jvm_opts.append(\"-Xms16G\")\n",
    "        # jvm_opts.append(\"-XX:SurvivorRatio=130\")\n",
    "        # jvm_opts.append(\"-XX:TargetSurvivorRatio=66\")\n",
    "\n",
    "        # Run the benchmark (hide output)\n",
    "        try:\n",
    "            subprocess.check_output(\n",
    "                [\n",
    "                    \"java\",\n",
    "                    \"-cp\",\n",
    "                    f\"{callback_path}:{bm_path}\",\n",
    "                    f\"-Xlog:gc*=trace:file={gc_log_file}:tags,time,uptime,level\",\n",
    "                    *jvm_opts,\n",
    "                    \"-Dvmstat.enable_jfr=yes\",\n",
    "                    \"-Dvmstat.csv=yes\",\n",
    "                    \"Harness\",\n",
    "                    \"-v\",\n",
    "                    \"-n\",\n",
    "                    f\"{n}\",\n",
    "                    bm,\n",
    "                ],\n",
    "                stderr=subprocess.STDOUT,\n",
    "                text=True,\n",
    "                env=self._env,\n",
    "            )\n",
    "        except subprocess.CalledProcessError as e:\n",
    "            raise Exception(\"Command failed with return code\", e.returncode, e.output)\n",
    "\n",
    "        return\n",
    "\n",
    "    def _render(self):\n",
    "        raise NotImplementedError(\"This environment has not implemented `render().'\")\n",
    "\n",
    "    def _print_welcome_msg(self):\n",
    "        print(\n",
    "            \"Successfully initialized a JVM Environment!\\n\",\n",
    "            f\"JDK: {self.jdk},\\n\",\n",
    "            f\"Benchmark: {self._bm} ({self.bm_path}),\\n\",\n",
    "            f\"Number of iterations: {self._n},\\n\",\n",
    "            f\"Goals: {self._goals},\\n\",\n",
    "            f\"Number of JVM options: {self._num_flags},\\n\",\n",
    "            f\"JVM options: {self._flags},\\n\",\n",
    "            # f\"Env. default state: {self._default_state},\\n\",\n",
    "            # f\"Env. default goal value: {self._default_state[self._goal_idx]},\\n\",\n",
    "        )\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([], dtype=float64)"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "JVMState().get_list()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "current_state = JVMState(\n",
    "    flags = [7, 12],\n",
    "    params= [0.005, 123.45, 99.0],\n",
    "    goals = [0.005]\n",
    ")\n",
    "\n",
    "previous_state = JVMState(\n",
    "    flags = [7, 14],\n",
    "    params= [0.0055, 126.45, 101.0],\n",
    "    goals = [0.01]\n",
    ")\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "jdk_path:  jdk-11.0.20.1.jdk\n",
      "Successfully initialized a JVM Environment!\n",
      " JDK: jdk-11.0.20.1.jdk/bin,\n",
      " Benchmark: avrora (dacapo-bench.jar),\n",
      " Number of iterations: 5,\n",
      " Goals: [{'name': 'footprint', 'minimize': True}],\n",
      " Number of JVM options: 3,\n",
      " JVM options: [{'name': 'MaxTenuringThreshold', 'min': 1, 'max': 16, 'step': 3}, {'name': 'ParallelGCThreads', 'min': 4, 'max': 24, 'step': 4, 'unit': ''}, {'name': 'MaxHeapSize', 'min': 2, 'max': 24, 'step': 4, 'unit': 'G'}],\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/dn/p1_3j9fs78g4wn8r0hcrqks40000gn/T/ipykernel_94487/2109101939.py:324: PerformanceWarning: indexing past lexsort depth may impact performance.\n",
      "  self._new_df.set_index(flag_names)\n",
      "/var/folders/dn/p1_3j9fs78g4wn8r0hcrqks40000gn/T/ipykernel_94487/2109101939.py:330: PerformanceWarning: indexing past lexsort depth may impact performance.\n",
      "  self._new_df.set_index(flag_names)\n"
     ]
    }
   ],
   "source": [
    "with open(\"env_config_3flags.json\", \"r\") as config:\n",
    "    env_properties = json.load(config)\n",
    "\n",
    "env = JVMEnv(**env_properties)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.005]"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "current_state.get_goals()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.01]"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "previous_state.get_goals()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10.5"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "env._get_reward(\n",
    "    current_state,\n",
    "    previous_state,\n",
    "    lower_is_better=True\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "random_policy = RandomPyPolicy(\n",
    "    None,\n",
    "    action_spec\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 3,  6, 15], dtype=int32)"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "random_policy.action(None).action"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "class JVMAction:\n",
    "    def __init__(self, flag_name):\n",
    "        self._flag_name = flag_name\n",
    "    \n",
    "    def "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "jdk_path:  jdk-11.0.20.1.jdk\n"
     ]
    },
    {
     "ename": "KeyError",
     "evalue": "'Key length (7) exceeds index depth (3)'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[48], line 4\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m \u001b[38;5;28mopen\u001b[39m(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124menv_config_multiflags.json\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mr\u001b[39m\u001b[38;5;124m\"\u001b[39m) \u001b[38;5;28;01mas\u001b[39;00m config:\n\u001b[1;32m      2\u001b[0m     env_properties \u001b[38;5;241m=\u001b[39m json\u001b[38;5;241m.\u001b[39mload(config)\n\u001b[0;32m----> 4\u001b[0m env \u001b[38;5;241m=\u001b[39m \u001b[43mJVMEnv\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43menv_properties\u001b[49m\u001b[43m)\u001b[49m\n",
      "Cell \u001b[0;32mIn[47], line 106\u001b[0m, in \u001b[0;36mJVMEnv.__init__\u001b[0;34m(self, jdk_path, bm_path, gc_viewer_jar, callback_path, goals, flags, params, default_flags, bm_name, n, verbose)\u001b[0m\n\u001b[1;32m    103\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_new_df \u001b[38;5;241m=\u001b[39m pd\u001b[38;5;241m.\u001b[39mread_csv(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mdatasets/\u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_bm\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m_real_saved_states.csv\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m    105\u001b[0m \u001b[38;5;66;03m# self._default_state = self._get_default_state(mode=\"default\")\u001b[39;00m\n\u001b[0;32m--> 106\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_default_state \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_get_default_state\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmode\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mrandom\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m)\u001b[49m\n\u001b[1;32m    108\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_perf_states \u001b[38;5;241m=\u001b[39m {}\n\u001b[1;32m    109\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_perf_states[\u001b[38;5;241m0\u001b[39m] \u001b[38;5;241m=\u001b[39m {\n\u001b[1;32m    110\u001b[0m     \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124margs\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_default_state[: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_num_flags],\n\u001b[1;32m    111\u001b[0m     \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mgoal\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_default_state[\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_goal_idx],\n\u001b[1;32m    112\u001b[0m     \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mextra\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_default_state[\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_goal_idx \u001b[38;5;241m+\u001b[39m \u001b[38;5;241m1\u001b[39m :],\n\u001b[1;32m    113\u001b[0m     \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcount\u001b[39m\u001b[38;5;124m\"\u001b[39m: \u001b[38;5;241m1\u001b[39m,\n\u001b[1;32m    114\u001b[0m }\n",
      "Cell \u001b[0;32mIn[47], line 389\u001b[0m, in \u001b[0;36mJVMEnv._get_default_state\u001b[0;34m(self, mode)\u001b[0m\n\u001b[1;32m    381\u001b[0m         step_val \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_flags_step_values[i]\n\u001b[1;32m    382\u001b[0m         rand_flags\u001b[38;5;241m.\u001b[39mappend(\n\u001b[1;32m    383\u001b[0m             random\u001b[38;5;241m.\u001b[39mrandrange(\n\u001b[1;32m    384\u001b[0m                 start\u001b[38;5;241m=\u001b[39mmin_val,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    387\u001b[0m             )\n\u001b[1;32m    388\u001b[0m         )\n\u001b[0;32m--> 389\u001b[0m     state \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_synthetic_run\u001b[49m\u001b[43m(\u001b[49m\u001b[43mrand_flags\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    390\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m state\n",
      "Cell \u001b[0;32mIn[47], line 312\u001b[0m, in \u001b[0;36mJVMEnv._synthetic_run\u001b[0;34m(self, flag_values)\u001b[0m\n\u001b[1;32m    309\u001b[0m param_names \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mlist\u001b[39m(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_new_df\u001b[38;5;241m.\u001b[39mcolumns[\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_num_flags :])\n\u001b[1;32m    310\u001b[0m \u001b[38;5;66;03m# Select columns where flags equal to flag_values\u001b[39;00m\n\u001b[1;32m    311\u001b[0m param_values \u001b[38;5;241m=\u001b[39m (\n\u001b[0;32m--> 312\u001b[0m     \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_new_df\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mset_index\u001b[49m\u001b[43m(\u001b[49m\u001b[43mflag_names\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    313\u001b[0m \u001b[43m    \u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mloc\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;28;43mtuple\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mflag_values\u001b[49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mparam_names\u001b[49m\u001b[43m]\u001b[49m\n\u001b[1;32m    314\u001b[0m     \u001b[38;5;241m.\u001b[39mvalues\n\u001b[1;32m    315\u001b[0m )\n\u001b[1;32m    316\u001b[0m state \u001b[38;5;241m=\u001b[39m [\u001b[38;5;241m*\u001b[39mflag_values, \u001b[38;5;241m*\u001b[39mparam_values]\n\u001b[1;32m    317\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m state\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexing.py:1097\u001b[0m, in \u001b[0;36m_LocationIndexer.__getitem__\u001b[0;34m(self, key)\u001b[0m\n\u001b[1;32m   1095\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_is_scalar_access(key):\n\u001b[1;32m   1096\u001b[0m         \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mobj\u001b[38;5;241m.\u001b[39m_get_value(\u001b[38;5;241m*\u001b[39mkey, takeable\u001b[38;5;241m=\u001b[39m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_takeable)\n\u001b[0;32m-> 1097\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_getitem_tuple\u001b[49m\u001b[43m(\u001b[49m\u001b[43mkey\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1098\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m   1099\u001b[0m     \u001b[38;5;66;03m# we by definition only have the 0th axis\u001b[39;00m\n\u001b[1;32m   1100\u001b[0m     axis \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39maxis \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;241m0\u001b[39m\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexing.py:1280\u001b[0m, in \u001b[0;36m_LocIndexer._getitem_tuple\u001b[0;34m(self, tup)\u001b[0m\n\u001b[1;32m   1278\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m suppress(IndexingError):\n\u001b[1;32m   1279\u001b[0m     tup \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_expand_ellipsis(tup)\n\u001b[0;32m-> 1280\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_getitem_lowerdim\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtup\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1282\u001b[0m \u001b[38;5;66;03m# no multi-index, so validate all of the indexers\u001b[39;00m\n\u001b[1;32m   1283\u001b[0m tup \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_validate_tuple_indexer(tup)\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexing.py:976\u001b[0m, in \u001b[0;36m_LocationIndexer._getitem_lowerdim\u001b[0;34m(self, tup)\u001b[0m\n\u001b[1;32m    974\u001b[0m \u001b[38;5;66;03m# we may have a nested tuples indexer here\u001b[39;00m\n\u001b[1;32m    975\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_is_nested_tuple_indexer(tup):\n\u001b[0;32m--> 976\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_getitem_nested_tuple\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtup\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    978\u001b[0m \u001b[38;5;66;03m# we maybe be using a tuple to represent multiple dimensions here\u001b[39;00m\n\u001b[1;32m    979\u001b[0m ax0 \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mobj\u001b[38;5;241m.\u001b[39m_get_axis(\u001b[38;5;241m0\u001b[39m)\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexing.py:1077\u001b[0m, in \u001b[0;36m_LocationIndexer._getitem_nested_tuple\u001b[0;34m(self, tup)\u001b[0m\n\u001b[1;32m   1074\u001b[0m     axis \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[38;5;241m1\u001b[39m\n\u001b[1;32m   1075\u001b[0m     \u001b[38;5;28;01mcontinue\u001b[39;00m\n\u001b[0;32m-> 1077\u001b[0m obj \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mgetattr\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mobj\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mname\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_getitem_axis\u001b[49m\u001b[43m(\u001b[49m\u001b[43mkey\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43maxis\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43maxis\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1078\u001b[0m axis \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[38;5;241m1\u001b[39m\n\u001b[1;32m   1080\u001b[0m \u001b[38;5;66;03m# if we have a scalar, we are done\u001b[39;00m\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexing.py:1343\u001b[0m, in \u001b[0;36m_LocIndexer._getitem_axis\u001b[0;34m(self, key, axis)\u001b[0m\n\u001b[1;32m   1341\u001b[0m \u001b[38;5;66;03m# fall thru to straight lookup\u001b[39;00m\n\u001b[1;32m   1342\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_validate_key(key, axis)\n\u001b[0;32m-> 1343\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_get_label\u001b[49m\u001b[43m(\u001b[49m\u001b[43mkey\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43maxis\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43maxis\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexing.py:1293\u001b[0m, in \u001b[0;36m_LocIndexer._get_label\u001b[0;34m(self, label, axis)\u001b[0m\n\u001b[1;32m   1291\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21m_get_label\u001b[39m(\u001b[38;5;28mself\u001b[39m, label, axis: AxisInt):\n\u001b[1;32m   1292\u001b[0m     \u001b[38;5;66;03m# GH#5567 this will fail if the label is not present in the axis.\u001b[39;00m\n\u001b[0;32m-> 1293\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mobj\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mxs\u001b[49m\u001b[43m(\u001b[49m\u001b[43mlabel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43maxis\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43maxis\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/generic.py:4088\u001b[0m, in \u001b[0;36mNDFrame.xs\u001b[0;34m(self, key, axis, level, drop_level)\u001b[0m\n\u001b[1;32m   4085\u001b[0m     index \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mindex\n\u001b[1;32m   4087\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(index, MultiIndex):\n\u001b[0;32m-> 4088\u001b[0m     loc, new_index \u001b[38;5;241m=\u001b[39m \u001b[43mindex\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_get_loc_level\u001b[49m\u001b[43m(\u001b[49m\u001b[43mkey\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mlevel\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m0\u001b[39;49m\u001b[43m)\u001b[49m\n\u001b[1;32m   4089\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m drop_level:\n\u001b[1;32m   4090\u001b[0m         \u001b[38;5;28;01mif\u001b[39;00m lib\u001b[38;5;241m.\u001b[39mis_integer(loc):\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexes/multi.py:2998\u001b[0m, in \u001b[0;36mMultiIndex._get_loc_level\u001b[0;34m(self, key, level)\u001b[0m\n\u001b[1;32m   2995\u001b[0m         \u001b[38;5;28;01mpass\u001b[39;00m\n\u001b[1;32m   2997\u001b[0m \u001b[38;5;66;03m# partial selection\u001b[39;00m\n\u001b[0;32m-> 2998\u001b[0m indexer \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mget_loc\u001b[49m\u001b[43m(\u001b[49m\u001b[43mkey\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   2999\u001b[0m ilevels \u001b[38;5;241m=\u001b[39m [i \u001b[38;5;28;01mfor\u001b[39;00m i \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mrange\u001b[39m(\u001b[38;5;28mlen\u001b[39m(key)) \u001b[38;5;28;01mif\u001b[39;00m key[i] \u001b[38;5;241m!=\u001b[39m \u001b[38;5;28mslice\u001b[39m(\u001b[38;5;28;01mNone\u001b[39;00m, \u001b[38;5;28;01mNone\u001b[39;00m)]\n\u001b[1;32m   3000\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mlen\u001b[39m(ilevels) \u001b[38;5;241m==\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mnlevels:\n",
      "File \u001b[0;32m~/projects/gc-ml/gc-ml-env/lib/python3.8/site-packages/pandas/core/indexes/multi.py:2817\u001b[0m, in \u001b[0;36mMultiIndex.get_loc\u001b[0;34m(self, key)\u001b[0m\n\u001b[1;32m   2815\u001b[0m keylen \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mlen\u001b[39m(key)\n\u001b[1;32m   2816\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mnlevels \u001b[38;5;241m<\u001b[39m keylen:\n\u001b[0;32m-> 2817\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mKeyError\u001b[39;00m(\n\u001b[1;32m   2818\u001b[0m         \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mKey length (\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mkeylen\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m) exceeds index depth (\u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mnlevels\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m)\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m   2819\u001b[0m     )\n\u001b[1;32m   2821\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m keylen \u001b[38;5;241m==\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mnlevels \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mis_unique:\n\u001b[1;32m   2822\u001b[0m     \u001b[38;5;66;03m# TODO: what if we have an IntervalIndex level?\u001b[39;00m\n\u001b[1;32m   2823\u001b[0m     \u001b[38;5;66;03m#  i.e. do we need _index_as_unique on that level?\u001b[39;00m\n\u001b[1;32m   2824\u001b[0m     \u001b[38;5;28;01mtry\u001b[39;00m:\n",
      "\u001b[0;31mKeyError\u001b[0m: 'Key length (7) exceeds index depth (3)'"
     ]
    }
   ],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "include_cols = [\n",
    "    \"fullGCPause\",\n",
    "    \"avgPause\",\n",
    "    \"fullGcPauseCount\",\n",
    "    \"footprint\",\n",
    "    \"gcPerformance\",\n",
    "    \"totalTenuredUsedMax\",\n",
    "    \"avgPromotion\",\n",
    "    \"fullGCPerformance\",\n",
    "    \"throughput\"\n",
    "]\n",
    "\n",
    "all_dfs = []\n",
    "\n",
    "def get_processed_df(csv_file):\n",
    "    processed_df = pd.read_csv(csv_file, sep=';', skiprows=1)\n",
    "    processed_df = processed_df.replace(',','', regex=True)\n",
    "    processed_df = processed_df.replace('n.a.','NaN', regex=True)\n",
    "    processed_df = processed_df[processed_df.iloc[:, 0].isin(include_cols)]\n",
    "    processed_df.index = processed_df.iloc[:, 0]\n",
    "    processed_df = processed_df.transpose().iloc[1:2, :] # .astype(float).values \n",
    "    return processed_df\n",
    "\n",
    "for summary in glob.glob(\"summaries_avrora/*.csv\"): # glob.glob(\"summaries_avrora/*.csv\") + \n",
    "    df = get_processed_df(summary)\n",
    "    \n",
    "    all_dfs.append(df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "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>totalHeapAllocMax</th>\n",
       "      <th>totalTenuredUsedMax</th>\n",
       "      <th>avgPromotion</th>\n",
       "      <th>avgPause</th>\n",
       "      <th>fullGcPauseCount</th>\n",
       "      <th>fullGCPause</th>\n",
       "      <th>footprint</th>\n",
       "      <th>throughput</th>\n",
       "      <th>gcPerformance</th>\n",
       "      <th>fullGCPerformance</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3355</td>\n",
       "      <td>64.8</td>\n",
       "      <td>0.03712</td>\n",
       "      <td>5</td>\n",
       "      <td>0.33</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.7</td>\n",
       "      <td>14920.584</td>\n",
       "      <td>6134.62</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3355</td>\n",
       "      <td>103.2</td>\n",
       "      <td>0.04522</td>\n",
       "      <td>5</td>\n",
       "      <td>0.41</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.64</td>\n",
       "      <td>14959.635</td>\n",
       "      <td>4953.668</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3409</td>\n",
       "      <td>-66765</td>\n",
       "      <td>0.04362</td>\n",
       "      <td>5</td>\n",
       "      <td>0.4</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.71</td>\n",
       "      <td>16238.272</td>\n",
       "      <td>5117.301</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3355</td>\n",
       "      <td>-150528</td>\n",
       "      <td>0.04677</td>\n",
       "      <td>5</td>\n",
       "      <td>0.42</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.7</td>\n",
       "      <td>13366.383</td>\n",
       "      <td>4824.204</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3354</td>\n",
       "      <td>-236954</td>\n",
       "      <td>0.05045</td>\n",
       "      <td>5</td>\n",
       "      <td>0.46</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.67</td>\n",
       "      <td>13693.046</td>\n",
       "      <td>4424.913</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>4748</td>\n",
       "      <td>19</td>\n",
       "      <td>0.04687</td>\n",
       "      <td>5</td>\n",
       "      <td>0.42</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.69</td>\n",
       "      <td>13714.105</td>\n",
       "      <td>9793.49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3411</td>\n",
       "      <td>-89702</td>\n",
       "      <td>0.04654</td>\n",
       "      <td>5</td>\n",
       "      <td>0.42</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.68</td>\n",
       "      <td>14191.244</td>\n",
       "      <td>4830.485</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3380</td>\n",
       "      <td>-137421</td>\n",
       "      <td>0.0536</td>\n",
       "      <td>5</td>\n",
       "      <td>0.49</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.24</td>\n",
       "      <td>13323.378</td>\n",
       "      <td>4152.97</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3355</td>\n",
       "      <td>-114893</td>\n",
       "      <td>0.04662</td>\n",
       "      <td>5</td>\n",
       "      <td>0.42</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.69</td>\n",
       "      <td>13341.199</td>\n",
       "      <td>4851.689</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1,508</th>\n",
       "      <td>3376</td>\n",
       "      <td>142.6</td>\n",
       "      <td>0.03331</td>\n",
       "      <td>5</td>\n",
       "      <td>0.3</td>\n",
       "      <td>1508</td>\n",
       "      <td>99.62</td>\n",
       "      <td>16745.099</td>\n",
       "      <td>6862.604</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>180 rows × 9 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "totalHeapAllocMax totalTenuredUsedMax avgPromotion  avgPause fullGcPauseCount  \\\n",
       " 1,508                           3355         64.8   0.03712                5   \n",
       " 1,508                           3355        103.2   0.04522                5   \n",
       " 1,508                           3409       -66765   0.04362                5   \n",
       " 1,508                           3355      -150528   0.04677                5   \n",
       " 1,508                           3354      -236954   0.05045                5   \n",
       "...                               ...          ...       ...              ...   \n",
       " 1,508                           4748           19   0.04687                5   \n",
       " 1,508                           3411       -89702   0.04654                5   \n",
       " 1,508                           3380      -137421    0.0536                5   \n",
       " 1,508                           3355      -114893   0.04662                5   \n",
       " 1,508                           3376        142.6   0.03331                5   \n",
       "\n",
       "totalHeapAllocMax fullGCPause footprint throughput gcPerformance  \\\n",
       " 1,508                   0.33      1508       99.7     14920.584   \n",
       " 1,508                   0.41      1508      99.64     14959.635   \n",
       " 1,508                    0.4      1508      99.71     16238.272   \n",
       " 1,508                   0.42      1508       99.7     13366.383   \n",
       " 1,508                   0.46      1508      99.67     13693.046   \n",
       "...                       ...       ...        ...           ...   \n",
       " 1,508                   0.42      1508      99.69     13714.105   \n",
       " 1,508                   0.42      1508      99.68     14191.244   \n",
       " 1,508                   0.49      1508      99.24     13323.378   \n",
       " 1,508                   0.42      1508      99.69     13341.199   \n",
       " 1,508                    0.3      1508      99.62     16745.099   \n",
       "\n",
       "totalHeapAllocMax fullGCPerformance  \n",
       " 1,508                      6134.62  \n",
       " 1,508                     4953.668  \n",
       " 1,508                     5117.301  \n",
       " 1,508                     4824.204  \n",
       " 1,508                     4424.913  \n",
       "...                             ...  \n",
       " 1,508                      9793.49  \n",
       " 1,508                     4830.485  \n",
       " 1,508                      4152.97  \n",
       " 1,508                     4851.689  \n",
       " 1,508                     6862.604  \n",
       "\n",
       "[180 rows x 9 columns]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "frame = pd.concat(all_dfs,)\n",
    "frame"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total number of selected parameters: 9\n",
      "Number of filtered parameters: 7\n"
     ]
    }
   ],
   "source": [
    "# Transpose.\n",
    "# frame = frame.T\n",
    "\n",
    "# Remove constant values.\n",
    "frame = frame.loc[:, (frame != frame.iloc[0]).any()]\n",
    "\n",
    "print(\"Total number of selected parameters:\", len(include_cols))\n",
    "print(\"Number of filtered parameters:\", len(frame.columns))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "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>totalHeapAllocMax</th>\n",
       "      <th>totalHeapUsedMax</th>\n",
       "      <th>totalHeapUsedMaxpc</th>\n",
       "      <th>totalTenuredUsedMax</th>\n",
       "      <th>totalYoungUsedMax</th>\n",
       "      <th>totalYoungUsedMaxpc</th>\n",
       "      <th>avgPromotion</th>\n",
       "      <th>promotionTotal</th>\n",
       "      <th>avgGCPause</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>totalHeapAllocMax</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>totalHeapUsedMax</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.163403</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>0.033075</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>totalHeapUsedMaxpc</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.163403</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>0.033075</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>totalTenuredUsedMax</th>\n",
       "      <td>0.163403</td>\n",
       "      <td>0.163403</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.163403</td>\n",
       "      <td>0.163403</td>\n",
       "      <td>-0.047795</td>\n",
       "      <td>-0.047794</td>\n",
       "      <td>0.146599</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>totalYoungUsedMax</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.163403</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>0.033075</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>totalYoungUsedMaxpc</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.163403</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>0.033075</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>avgPromotion</th>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.047795</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-0.361351</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>promotionTotal</th>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.047794</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>-0.023649</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>-0.361351</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>avgGCPause</th>\n",
       "      <td>0.033075</td>\n",
       "      <td>0.033075</td>\n",
       "      <td>0.146599</td>\n",
       "      <td>0.033075</td>\n",
       "      <td>0.033075</td>\n",
       "      <td>-0.361351</td>\n",
       "      <td>-0.361351</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "totalHeapAllocMax    totalHeapUsedMax  totalHeapUsedMaxpc  \\\n",
       "totalHeapAllocMax                                           \n",
       "totalHeapUsedMax             1.000000            1.000000   \n",
       "totalHeapUsedMaxpc           1.000000            1.000000   \n",
       "totalTenuredUsedMax          0.163403            0.163403   \n",
       "totalYoungUsedMax            1.000000            1.000000   \n",
       "totalYoungUsedMaxpc          1.000000            1.000000   \n",
       "avgPromotion                -0.023649           -0.023649   \n",
       "promotionTotal              -0.023649           -0.023649   \n",
       "avgGCPause                   0.033075            0.033075   \n",
       "\n",
       "totalHeapAllocMax    totalTenuredUsedMax  totalYoungUsedMax  \\\n",
       "totalHeapAllocMax                                             \n",
       "totalHeapUsedMax                0.163403           1.000000   \n",
       "totalHeapUsedMaxpc              0.163403           1.000000   \n",
       "totalTenuredUsedMax             1.000000           0.163403   \n",
       "totalYoungUsedMax               0.163403           1.000000   \n",
       "totalYoungUsedMaxpc             0.163403           1.000000   \n",
       "avgPromotion                   -0.047795          -0.023649   \n",
       "promotionTotal                 -0.047794          -0.023649   \n",
       "avgGCPause                      0.146599           0.033075   \n",
       "\n",
       "totalHeapAllocMax    totalYoungUsedMaxpc  avgPromotion  promotionTotal  \\\n",
       "totalHeapAllocMax                                                        \n",
       "totalHeapUsedMax                1.000000     -0.023649       -0.023649   \n",
       "totalHeapUsedMaxpc              1.000000     -0.023649       -0.023649   \n",
       "totalTenuredUsedMax             0.163403     -0.047795       -0.047794   \n",
       "totalYoungUsedMax               1.000000     -0.023649       -0.023649   \n",
       "totalYoungUsedMaxpc             1.000000     -0.023649       -0.023649   \n",
       "avgPromotion                   -0.023649      1.000000        1.000000   \n",
       "promotionTotal                 -0.023649      1.000000        1.000000   \n",
       "avgGCPause                      0.033075     -0.361351       -0.361351   \n",
       "\n",
       "totalHeapAllocMax    avgGCPause  \n",
       "totalHeapAllocMax                \n",
       "totalHeapUsedMax       0.033075  \n",
       "totalHeapUsedMaxpc     0.033075  \n",
       "totalTenuredUsedMax    0.146599  \n",
       "totalYoungUsedMax      0.033075  \n",
       "totalYoungUsedMaxpc    0.033075  \n",
       "avgPromotion          -0.361351  \n",
       "promotionTotal        -0.361351  \n",
       "avgGCPause             1.000000  "
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "frame.corr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: xlabel='totalHeapAllocMax', ylabel='totalHeapAllocMax'>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots() # figsize=(8, 8)\n",
    "\n",
    "# Generate a mask for the upper triangle\n",
    "mask = np.triu(np.ones_like(frame.corr(), dtype=bool))\n",
    "\n",
    "sns.heatmap(frame.corr(), ax=ax, fmt=\".2f\", annot=True, mask=mask)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of correlated parameters: 7\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "totalHeapAllocMax\n",
       "avgPromotion          -0.361351\n",
       "promotionTotal        -0.361351\n",
       "totalHeapUsedMaxpc     0.033075\n",
       "totalHeapUsedMax       0.033075\n",
       "totalYoungUsedMax      0.033075\n",
       "totalYoungUsedMaxpc    0.033075\n",
       "totalTenuredUsedMax    0.146599\n",
       "avgGCPause             1.000000\n",
       "Name: avgGCPause, dtype: float64"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Correlation with output variable\n",
    "cor_target = frame.corr()[\"avgGCPause\"].sort_values()\n",
    "# cor_target = frame.corr()[\"gcPerformance\"].sort_values()\n",
    "print(\"Number of correlated parameters:\", len(cor_target) - 1)\n",
    "cor_target"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Extra parameters (Avrora)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial list of parameters:\n",
      " totalHeapAllocMax\n",
      "totalHeapUsedMax\n",
      "totalHeapUsedMaxpc\n",
      "totalTenuredAllocMax\n",
      "totalTenuredUsedMax\n",
      "totalTenuredUsedMaxpc\n",
      "totalYoungAllocMax\n",
      "totalYoungUsedMax\n",
      "totalYoungUsedMaxpc\n",
      "avgPromotion\n",
      "promotionTotal\n",
      "avgGCPause\n"
     ]
    }
   ],
   "source": [
    "print(\"Initial list of parameters:\\n\", \"\\n\".join(include_cols))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following parameters did not change over time and were excluded from the list.\n",
    "- totalHeapAllocMax\n",
    "- totalYoungAllocMax\n",
    "- totalTenuredUsedMaxpc\n",
    "- totalTenuredAllocMax\n",
    "\n",
    "\n",
    "1. avgPromotion (-0.36, negative correlation)\n",
    "2. promotionTotal (-0.36, negative correlation)\n",
    "3. totalHeapUsedMaxpc (0.033075)\n",
    "4. totalHeapUsedMax (0.033075)\n",
    "5. totalYoungUsedMax (0.033075)\n",
    "6. totalYoungUsedMaxpc (0.033075)\n",
    "7. totalTenuredUsedMax (0.146599)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Extra parameters (Kafka)\n",
    "\n",
    "1. avgPromotion           0.183438\n",
    "2. promotionTotal         0.183791\n",
    "3. totalHeapUsedMaxpc    -0.021756\n",
    "4. totalHeapUsedMax      -0.011937\n",
    "5. totalYoungUsedMax     -0.011882\n",
    "6. totalYoungUsedMaxpc   -0.007519\n",
    "7. totalTenuredUsedMax    0.080765"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "flags = {\n",
    "    \"MaxTenuringThreshold\": {\"min\": 1, \"max\": 16, \"step\": 3},\n",
    "    \"ParallelGCThreads\": {\"min\": 4, \"max\": 24, \"step\": 4},\n",
    "}\n",
    "\n",
    "flags_min_values = [flags[i][\"min\"] for i in flags.keys()]\n",
    "flags_max_values = [flags[i][\"max\"] for i in flags.keys()]\n",
    "flags_step_values = [flags[i][\"step\"] for i in flags.keys()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "_num_variables=2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[4, 12]"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import random\n",
    "\n",
    "random_state = []\n",
    "for i in range(_num_variables):\n",
    "    random_state.append(\n",
    "        random.randrange(flags_min_values[i], flags_max_values[i], flags_step_values[i])\n",
    "    )\n",
    "random_state\n",
    "# [random.randint(flags_min_values[i], flags_max_values[i]) for i in range(_num_variables)]\n",
    "# random.randint(flags_min_values[0], flags_max_values[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "[]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[1, 4]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "flags_min_values"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "gc-ml-env",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}