demo.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This onboarding notebook will demonstrate how to use this codebase. The different sections correspond to different parts of the overall pipeline and also roughly correspond to the different files in `utils`.\n",
    "\n",
    "Before running this notebook, ensure that your virtual environment is set up and that you have installed the necessary libraries. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### imports and setup"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, we want to set up the ipython autoreload extension, which autoreloads imported modules every time we run a cell in this notebook, so that any changes elsewhere in the codebase will be up to date when code in this notebook is executed."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The autoreload extension is already loaded. To reload it, use:\n",
      "  %reload_ext autoreload\n"
     ]
    }
   ],
   "source": [
    "# ipython extension to autoreload imported modules so that any changes will be up to date before running code in this nb\n",
    "%load_ext autoreload \n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.jraph_data import get_lorenz_graph_tuples, print_graph_fts\n",
    "from utils.lorenz import load_lorenz96_2coupled\n",
    "\n",
    "import numpy as np\n",
    "import jax.numpy as jnp\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### CONFIGS"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "TODO move this to top and reference it instead of magic numbers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Configs are useful for keeping track of the parameters you are using for a particular run. Also, some of the parameters need to be reused between the data generation and the model/training setup, so it's convenient to be able to just reference a config file that contains everything in one place. Here is what an example config looks like and how to call attributes: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "metadata": {},
   "outputs": [],
   "source": [
    "import ml_collections\n",
    "\n",
    "config = ml_collections.ConfigDict()\n",
    "\n",
    "# Data params. \n",
    "config.n_samples=20\n",
    "config.input_steps=3\n",
    "config.output_delay=0\n",
    "config.output_steps=2\n",
    "config.timestep_duration=1\n",
    "config.sample_buffer=1\n",
    "config.time_resolution=100\n",
    "config.init_buffer_samples=0\n",
    "config.train_pct=0.7\n",
    "config.val_pct=0.2\n",
    "config.test_pct=0.1\n",
    "config.K=36\n",
    "config.F=8\n",
    "config.c=10\n",
    "config.b=10\n",
    "config.h=1\n",
    "config.seed=42\n",
    "config.normalize=True\n",
    "config.fully_connected_edges=True\n",
    "\n",
    "# Optimizer.\n",
    "config.optimizer = 'adam'\n",
    "config.learning_rate = 1e-3\n",
    "\n",
    "# Training hyperparameters.\n",
    "# config.batch_size = 3\n",
    "config.epochs = 2\n",
    "config.log_every_epochs = 1\n",
    "config.eval_every_epochs = 1\n",
    "config.checkpoint_every_epochs = 1\n",
    "config.max_checkpts_to_keep = None # None means keep all checkpoints\n",
    "\n",
    "# GNN hyperparameters.\n",
    "config.model = 'MLPGraphNetwork'\n",
    "config.n_blocks = 1\n",
    "config.activation = 'relu'\n",
    "config.dropout_rate = 0.1\n",
    "config.skip_connections = False # This was throwing a broadcast error in add_graphs_tuples_nodes when this was set to True\n",
    "config.layer_norm = False # TODO perhaps we want to turn on later\n",
    "config.edge_features = (4, 8) # the last feature size will be the number of features that the graph predicts\n",
    "config.node_features = (32, 2)\n",
    "config.global_features = None\n",
    "config.share_params = False"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then you can simply reference the config's attributes by calling, e.g. \n",
    "`some_param_you_want_to_set = config.param_name`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### DATA"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this section we go over how to run the Lorenz simulation, and how to generate the graph data objects. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Lorenz simulation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.lorenz import run_download_lorenz96_2coupled, load_lorenz96_2coupled"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Normally, we want to save the Lorenz simulations that we run so that we don't have to repeat the ODE integration if we want to reuse the data. \n",
    "\n",
    "The following code runs ODE integration over the coupled 2-layer Lorenz96 model and saves the data to an .npz file. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [],
   "source": [
    "run_download_lorenz96_2coupled(\n",
    "        fname=\"data/demo\", \n",
    "        K=config.K,\n",
    "        F=config.F,\n",
    "        c=config.c,\n",
    "        b=config.b,\n",
    "        h=config.h,\n",
    "        n_steps=100,\n",
    "        resolution=config.time_resolution,\n",
    "        seed=config.seed)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Check that the data folder contains a new .npz file, and that the data_directory.json file now contains a new entry for this simulation run. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, we can load the data: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(100, 72)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[ 8.        ,  8.        ,  8.        , ...,  1.        ,\n",
       "         1.        ,  1.        ],\n",
       "       [ 7.9902934 ,  7.99291748,  7.990142  , ...,  0.9810261 ,\n",
       "         0.98093732,  0.98058919],\n",
       "       [ 7.98172884,  7.99153972,  7.98050267, ...,  0.9642227 ,\n",
       "         0.96387367,  0.96249242],\n",
       "       ...,\n",
       "       [ 4.07543482, -6.49598583, -4.0179002 , ..., -0.12132339,\n",
       "         0.05132519,  0.4683407 ],\n",
       "       [ 3.20270803, -6.75732122, -3.52443301, ..., -0.1154428 ,\n",
       "         0.13212121,  0.34186311],\n",
       "       [ 2.31485714, -6.90357552, -3.07708665, ..., -0.11520219,\n",
       "         0.20385555,  0.2402943 ]])"
      ]
     },
     "execution_count": 117,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t, X = load_lorenz96_2coupled(fname=\"data/demo.npz\")\n",
    "print(X.shape)\n",
    "X"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "You can play around with different ways to visualize this. (There are some existing functions in utils.visualization but they're kinda a mess right now.)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### GraphsTuples"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the pipeline to generate data and train the model, you won't actually need to call the Lorenz simulation directly, since it's wrapped in the calls to get the graph data. \n",
    "\n",
    "We are using the `GraphsTuple` object defined in the `jraph` library. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note here the difference between samples, timesteps, and (simulation) steps. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "TODO insert a visualization of what these are "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.jraph_data import get_lorenz_graph_tuples, print_graph_fts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {},
   "outputs": [],
   "source": [
    "# generate desired dataset with train/val split and subsampled windows\n",
    "graph_tuple_dict = get_lorenz_graph_tuples(\n",
    "    n_samples=config.n_samples,\n",
    "    input_steps=config.input_steps,\n",
    "    output_delay=config.output_delay,\n",
    "    output_steps=config.output_steps,\n",
    "    timestep_duration=config.timestep_duration,\n",
    "    sample_buffer=config.sample_buffer,\n",
    "    time_resolution=config.time_resolution,\n",
    "    init_buffer_samples=config.init_buffer_samples,\n",
    "    train_pct=config.train_pct,\n",
    "    val_pct=config.val_pct,\n",
    "    test_pct=config.test_pct,\n",
    "    K=config.K,\n",
    "    F=config.F,\n",
    "    c=config.c,\n",
    "    b=config.b,\n",
    "    h=config.h,\n",
    "    seed=config.seed,\n",
    "    normalize=config.normalize,\n",
    "    fully_connected_edges=config.fully_connected_edges)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note the additional parameters for `normalization` and `fully_connected_edges`. If `fully_connected_edges` is False, each node will only be connected to its two nearest neighbors and itself (5 total edges per node). There are some theoretical and practical considerations about which to choose (atm I am not sure what is best – it seems like theoretically fully-connected could be better, but that performed way worse in hyperparameter tuning. Also, Keisler's Graph Weather paper only had edges between adjacent nodes (on a sphere), so it seems like in theory we may not actually need fully-connected?)\n",
    "\n",
    "Also, (not doing it here, but) it can be helpful to compare the normalized with non-normalized data to check that the pre-processing step is correct. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "example of how to index the graph_tuple_dict object: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "graph_tuple_dict type: <class 'dict'>\n",
      "graph_tuple_dict keys: dict_keys(['train', 'val', 'test'])\n",
      "graph_tuple_dict value type: <class 'dict'>\n",
      "train dataset keys: dict_keys(['inputs', 'targets'])\n",
      "train dataset value type: <class 'list'>\n",
      "size of train inputs: 14\n",
      "size of train targets: 14\n",
      "size of val inputs: 4\n",
      "size of val targets: 4\n",
      "size of test inputs: 2\n",
      "size of test targets: 2\n",
      "train inputs window type: <class 'list'>\n",
      "train input window size (i.e. input steps per window): 3\n",
      "element type in window: <class 'jraph._src.graph.GraphsTuple'>\n"
     ]
    }
   ],
   "source": [
    "train_dataset = graph_tuple_dict['train']\n",
    "val_dataset = graph_tuple_dict['val']\n",
    "test_dataset = graph_tuple_dict['test']\n",
    "\n",
    "train_inputs = train_dataset['inputs']\n",
    "train_targets = train_dataset['targets']\n",
    "val_inputs = val_dataset['inputs']\n",
    "val_targets = val_dataset['targets']\n",
    "test_inputs = test_dataset['inputs']\n",
    "test_targets = test_dataset['targets']\n",
    "\n",
    "sample_input_window = train_inputs[0]\n",
    "sample_target_window = train_targets[0]\n",
    "sample_graph = sample_input_window[0]\n",
    "\n",
    "print(\"graph_tuple_dict type:\", type(graph_tuple_dict))\n",
    "print(\"graph_tuple_dict keys:\", graph_tuple_dict.keys())\n",
    "print(\"graph_tuple_dict value type:\", type(train_dataset))\n",
    "print(\"train dataset keys:\", train_dataset.keys())\n",
    "print(\"train dataset value type:\", type(train_inputs))\n",
    "\n",
    "print(\"size of train inputs:\", len(train_inputs))\n",
    "print(\"size of train targets:\", len(train_targets))\n",
    "print(\"size of val inputs:\", len(val_inputs))\n",
    "print(\"size of val targets:\", len(val_targets))\n",
    "print(\"size of test inputs:\", len(test_inputs))\n",
    "print(\"size of test targets:\", len(test_targets))\n",
    "\n",
    "print(\"train inputs window type:\", type(sample_input_window))\n",
    "print(\"train input window size (i.e. input steps per window):\", len(sample_input_window))\n",
    "print(\"element type in window:\", type(sample_graph))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is an example of what a single GraphsTuple object looks like (representing the state of the system at a single instant in time). (It's terrible to scroll through, would recommend collapsing). Below are better ways to index through the object."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GraphsTuple(nodes=array([[0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25882027, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ],\n",
      "       [0.25872678, 1.3448845 ]], dtype=float32), edges=Array([[ 0.],\n",
      "       [-1.],\n",
      "       [-2.],\n",
      "       ...,\n",
      "       [ 2.],\n",
      "       [ 1.],\n",
      "       [ 0.]], dtype=float32), receivers=Array([ 0,  1,  2, ..., 33, 34, 35], dtype=int32), senders=Array([ 0,  0,  0, ..., 35, 35, 35], dtype=int32), globals=Array([[1.]], dtype=float32), n_node=Array([36], dtype=int32), n_edge=Array([1296], dtype=int32))\n"
     ]
    }
   ],
   "source": [
    "print(sample_graph)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of nodes: 36\n",
      "Number of edges: 1296\n",
      "Node features shape: (36, 2)\n",
      "Edge features shape: (1296, 1)\n",
      "Global features shape: (1, 1)\n"
     ]
    }
   ],
   "source": [
    "# helper function \n",
    "print_graph_fts(sample_graph)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(36, 2)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25882027, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ],\n",
       "       [0.25872678, 1.3448845 ]], dtype=float32)"
      ]
     },
     "execution_count": 123,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# node features (i.e. 36 nodes x 2 layers)\n",
    "print(graph_tuple_dict['train']['inputs'][0][0].nodes.shape)\n",
    "graph_tuple_dict['train']['inputs'][0][0].nodes "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1296, 1)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Array([[ 0.],\n",
       "       [-1.],\n",
       "       [-2.],\n",
       "       ...,\n",
       "       [ 2.],\n",
       "       [ 1.],\n",
       "       [ 0.]], dtype=float32)"
      ]
     },
     "execution_count": 124,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# edge features \n",
    "print(graph_tuple_dict['train']['inputs'][0][0].edges.shape)\n",
    "graph_tuple_dict['train']['inputs'][0][0].edges"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "understanding edges: in a GraphsTuple, edges are encoded using three attributes: `GraphsTuple.senders`, `GraphsTuple.receivers`, and `GraphsTuple.edges`. \n",
    "\n",
    "The ith element of the `senders` attribute denotes the index of the source node; the ith element of the `receivers` attribute denotes the index of the receiving node. (The `senders` and `receivers` attributes are both 1d arrays.) The ith array of the `edges` attribute (a 2d array) are the encoded features for each node. In this case, each edge only has a single feature, which encodes the distance between the sender and receiver node. \n",
    "\n",
    "If the graph is fully connected, we expect 36\\*36 = 1296 edges. (If not, we default to 5 edges per node, so we would expect 36*5 = 180 edges.)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Side note, gathering node-level data into a time series is kind of a pain because it requires iterating over all GraphsTuples in the desired time frame... "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Side note 2: in jraph, there is an ability to create [batches](https://jraph.readthedocs.io/en/latest/api.html#batching-padding-utilities) of GraphsTuples, which basically means combining multiple GraphsTuples into a single GraphsTuple object. This actually makes gathering time series data easier/quicker; HOWEVER, it does not play nice with jax and is not compilable :( hence why some parts of the codebase do really slow and stupid indexing. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Side note 3: the `global_features` attribute in these graphs is just a dummy variable and is not used during training. I was trying to get rid of the variable but ran into some random bugs so I just left it as a placeholder since I didn't feel like wasting time dealing with it. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### MODELS"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here is an example of how to use a basic model. The process is essentially the same for any model you want to use/define, with the differences primarily being the parameters you pass in when initializing an instance of the model class. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.jraph_models import MLPGraphNetwork\n",
    "import flax.linen as nn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MLPGraphNetwork(\n",
       "    # attributes\n",
       "    n_blocks = 1\n",
       "    share_params = False\n",
       "    dropout_rate = 0.1\n",
       "    skip_connections = False\n",
       "    layer_norm = False\n",
       "    deterministic = True\n",
       "    activation = relu\n",
       "    edge_features = (4, 8)\n",
       "    node_features = (32, 2)\n",
       "    global_features = None\n",
       ")"
      ]
     },
     "execution_count": 126,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "activation_funcs = {\n",
    "    \"relu\": nn.relu,\n",
    "    \"elu\": nn.elu,\n",
    "    \"leaky_relu\": nn.leaky_relu,\n",
    "}\n",
    "activation = activation_funcs[config.activation]\n",
    "\n",
    "model = MLPGraphNetwork(            \n",
    "    n_blocks=config.n_blocks,\n",
    "    share_params=config.share_params,\n",
    "    dropout_rate=config.dropout_rate,\n",
    "    skip_connections=config.skip_connections,\n",
    "    layer_norm=config.layer_norm,\n",
    "    deterministic=True,\n",
    "    activation=activation,\n",
    "    edge_features=config.edge_features,\n",
    "    node_features=config.node_features,\n",
    "    global_features=config.global_features,\n",
    ")\n",
    "model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These models inherent from the flax nn.Module class.\n",
    "\n",
    "As of right now, models take in *lists* of GraphsTuples (i.e. windows of data). However, it currently can only process one timestep of data at a time (it can't process windows of data since there is no recurrent cell or anything to do that) so the model just uses the first GraphsTuple in a window and drops the rest. This is important to keep in mind when trying to call the model on data. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To actually use a model, things rapidly get more complicated. At this point we bring in the `flax` library to help us keep track of model states, params, optimizers, etc. (If you've used pytorch or tensorflow you usually don't have to deal with this because it's taken care of behind the scenes!) \n",
    "\n",
    "Before initialization, a flax/jax model is sort of just nebulous emptiness. The model doesn't even know how big it needs to be - it figures this out when you give it some sample data during initialization. \n",
    "\n",
    "For more details, see the [flax documentation](https://flax.readthedocs.io/en/latest/guides/flax_fundamentals/flax_basics.html#model-parameters-initialization)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {},
   "outputs": [],
   "source": [
    "import jax\n",
    "from flax.training import train_state\n",
    "import optax # library with optimizer classes used with jax"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create and initialize the parameters.\n",
    "rng = jax.random.key(0)\n",
    "rng, init_rng = jax.random.split(rng)\n",
    "params = jax.jit(model.init)(init_rng, sample_input_window)\n",
    "\n",
    "# Create an optimizer (you can create different types of optimizers, e.g. adam, SGD, etc.)\n",
    "tx = optax.adam(learning_rate=config.learning_rate)\n",
    "\n",
    "# Create the training state.\n",
    "state = train_state.TrainState.create(\n",
    "    apply_fn=model.apply, params=params, tx=tx\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A flax train_state is basically a convenient group of a bunch of things we need to call a forward pass, perform an update step for optimization, etc. These include the model `apply` function (i.e. forward pass/prediction function), the actual parameters, and the optimizer object. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we can finally perform a forward pass. Note that this model is still totally untrained, so the outputs are going to be garbage. \n",
    "\n",
    "Since we need both the model apply function and the model's external params, the train_state is set up to allow us to do this more conveniently than if we used the raw jax. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "metadata": {},
   "outputs": [],
   "source": [
    "garbage_output = state.apply_fn(state.params, sample_input_window, rngs={}) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 130,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'list'>\n",
      "1\n",
      "<class 'jraph._src.graph.GraphsTuple'>\n"
     ]
    }
   ],
   "source": [
    "print(type(garbage_output))\n",
    "print(len(garbage_output))\n",
    "print(type(garbage_output[0]))\n",
    "# print(garbage_output[0].nodes) # commented out because it's long"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Remember how the input to the model is a window (even though we aren't currently using the whole window)? Well, we want the output type to match the input type (in case we want to chain multiple MLPBlocks together, like we do in the MLPGraphNetwork class); however, we only output one timestep at a time, hence why we have a list containing a single GraphsTuple object. (If we want to output a whole window of data, we'll have to call the model `apply` function autoregressively.)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### TRAINING"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A single training step consists of:\n",
    "* computing rollout and loss for a single forward pass \n",
    "* updating model params based on gradient"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.jraph_training import create_model, train_step"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note that for training, if we want dropout, we want to set the `deterministic` param for the model to False so that the dropout occurs randomly. (You should turn off `dropout` and `deterministic` for evaluation.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create and initialize the network.\n",
    "rng = jax.random.key(0)\n",
    "rng, init_rng = jax.random.split(rng)\n",
    "init_net = create_model(config, deterministic=True)\n",
    "params = jax.jit(init_net.init)(init_rng, sample_input_window)\n",
    "\n",
    "# Create the optimizer.\n",
    "tx = optax.adam(learning_rate=config.learning_rate)\n",
    "\n",
    "# Create the training state.\n",
    "net = create_model(config, deterministic=False)\n",
    "state = train_state.TrainState.create(\n",
    "    apply_fn=net.apply, params=params, tx=tx\n",
    ")\n",
    "\n",
    "rng, dropout_rng = jax.random.split(rng)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "metadata": {},
   "outputs": [],
   "source": [
    "state, metrics_update, pred_nodes = train_step(\n",
    "    state=state,\n",
    "    n_rollout_steps=config.output_steps,\n",
    "    input_window_graphs=sample_input_window,\n",
    "    target_window_graphs=sample_target_window,\n",
    "    rngs={'dropout': dropout_rng})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 134,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[Array([[ -98.67787  ,  -39.9216   ],\n",
       "        [-120.49137  ,  -50.342407 ],\n",
       "        [-104.10324  ,  -36.979095 ],\n",
       "        [ -91.72358  ,  -45.434982 ],\n",
       "        [-102.39264  ,  -23.706886 ],\n",
       "        [-105.23365  ,  -38.129944 ],\n",
       "        [-107.99459  ,  -34.880634 ],\n",
       "        [ -75.59818  ,   -2.6439273],\n",
       "        [-107.79181  ,  -55.576935 ],\n",
       "        [   0.       ,  -81.21088  ],\n",
       "        [-120.72201  ,  -47.33336  ],\n",
       "        [   0.       ,  -99.84738  ],\n",
       "        [ -61.948593 ,  -91.852005 ],\n",
       "        [   0.       ,  -19.733942 ],\n",
       "        [-114.46859  ,  -39.500988 ],\n",
       "        [-108.012985 ,   -7.471518 ],\n",
       "        [ -91.07159  ,    0.       ],\n",
       "        [ -57.294647 ,    0.       ],\n",
       "        [   0.       ,  -39.664696 ],\n",
       "        [ -82.92593  ,  -11.08629  ],\n",
       "        [-102.70282  ,  -42.68211  ],\n",
       "        [ -93.7846   ,  -39.141167 ],\n",
       "        [-107.27263  ,  -63.723927 ],\n",
       "        [ -72.32507  ,  -66.53941  ],\n",
       "        [ -68.34085  ,  -87.95816  ],\n",
       "        [ -90.03124  ,  -48.92655  ],\n",
       "        [ -91.88272  ,   -0.9934881],\n",
       "        [-106.62928  ,   -7.7114606],\n",
       "        [ -62.93171  ,    2.6884978],\n",
       "        [-104.14404  ,  -30.487356 ],\n",
       "        [   0.       ,  -49.252712 ],\n",
       "        [   0.       ,    0.       ],\n",
       "        [-110.31584  ,  -34.416157 ],\n",
       "        [-108.635956 ,  -60.364822 ],\n",
       "        [-107.265495 ,  -34.46271  ],\n",
       "        [-105.96914  ,  -43.04244  ]], dtype=float32),\n",
       " Array([[ -98.72615  ,  -39.949898 ],\n",
       "        [-120.56735  ,  -50.40268  ],\n",
       "        [-104.12997  ,  -37.02776  ],\n",
       "        [ -91.75906  ,  -45.58336  ],\n",
       "        [-102.4426   ,  -23.787077 ],\n",
       "        [-105.279015 ,  -38.148052 ],\n",
       "        [-108.029785 ,  -34.922615 ],\n",
       "        [ -75.66549  ,   -2.646826 ],\n",
       "        [-107.8706   ,  -55.574516 ],\n",
       "        [   0.       ,  -81.30723  ],\n",
       "        [-120.79002  ,  -47.406075 ],\n",
       "        [   0.       ,  -99.979126 ],\n",
       "        [ -61.978172 ,  -91.92586  ],\n",
       "        [   0.       ,  -19.8112   ],\n",
       "        [-114.52513  ,  -39.540672 ],\n",
       "        [-108.02388  ,   -7.419073 ],\n",
       "        [ -91.14046  ,    0.       ],\n",
       "        [ -57.28967  ,    0.       ],\n",
       "        [   0.       ,  -39.669083 ],\n",
       "        [ -82.92543  ,  -11.086861 ],\n",
       "        [-102.75729  ,  -42.71567  ],\n",
       "        [ -93.85593  ,  -39.120506 ],\n",
       "        [-107.35231  ,  -63.75791  ],\n",
       "        [ -72.35122  ,  -66.66132  ],\n",
       "        [ -68.36242  ,  -88.12676  ],\n",
       "        [ -90.069984 ,  -48.96087  ],\n",
       "        [ -91.902725 ,   -1.01836  ],\n",
       "        [-106.68349  ,   -7.7302594],\n",
       "        [ -62.993374 ,    2.7189777],\n",
       "        [-104.17473  ,  -30.4798   ],\n",
       "        [   0.       ,  -49.24508  ],\n",
       "        [   0.       ,    0.       ],\n",
       "        [-110.32674  ,  -34.450756 ],\n",
       "        [-108.71016  ,  -60.40379  ],\n",
       "        [-107.33441  ,  -34.528248 ],\n",
       "        [-105.91428  ,  -43.127853 ]], dtype=float32)]"
      ]
     },
     "execution_count": 134,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_nodes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The `metrics_update` variable is an object that holds some useful information about metrics from the training step, namely the loss. It can be accessed as follows: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "TrainMetrics(_reduction_counter=_ReductionCounter(value=Array(1, dtype=int32, weak_type=True)), loss=Metric.from_output.<locals>.FromOutput(total=Array(5125.2056, dtype=float32), count=Array(1., dtype=float32)))\n",
      "access loss like this: 5125.2056\n"
     ]
    }
   ],
   "source": [
    "print(metrics_update)\n",
    "print('access loss like this:', metrics_update.loss.total)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### EVALUATION (VALIDATION, TESTING)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The validation step is interwoven throughout the training pipeline; it usually occurs every time you finish a training epoch, or if you want to reduce runtime, every X epochs. \n",
    "\n",
    "The testing dataset should not be used until the **VERY** end of model development, e.g. when you're about to present your results (preparing a poster, presentation, manuscript, etc.) It shouldn't influence your model development at all. That's what the validation dataset is for. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "example of a single evaluation step: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.jraph_training import evaluate_step"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Again, during evaluation, we want to turn OFF the `dropout` and `deterministic` params in the model instantiation, so we make a new model; in our state we want to copy the params from the training state, keep the correct `apply` function from the new evaluatio model, and don't care about optimizers because we're just doing evaluation. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 137,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create the evaluation state, corresponding to a deterministic model.\n",
    "eval_net = create_model(config, deterministic=True) # Note that dropout is deactivated if deterministic is True. \n",
    "eval_state = state.replace(apply_fn=eval_net.apply)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_metrics, pred_nodes = evaluate_step(\n",
    "    state=eval_state,\n",
    "    n_rollout_steps=config.output_steps,\n",
    "    input_window_graphs=sample_input_window,\n",
    "    target_window_graphs=sample_target_window)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We expect the prediction to be different from above because the params should've been updated. Ideally the loss should also go down a bit. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[Array([[-83.19216 , -30.645123],\n",
       "        [-83.19216 , -30.645128],\n",
       "        [-83.19216 , -30.645126],\n",
       "        [-83.192154, -30.64513 ],\n",
       "        [-83.19217 , -30.645134],\n",
       "        [-83.19217 , -30.645134],\n",
       "        [-83.192154, -30.64513 ],\n",
       "        [-83.19216 , -30.645132],\n",
       "        [-83.192154, -30.645132],\n",
       "        [-83.19216 , -30.645126],\n",
       "        [-83.192154, -30.645119],\n",
       "        [-83.192154, -30.645124],\n",
       "        [-83.19216 , -30.645126],\n",
       "        [-83.19217 , -30.645132],\n",
       "        [-83.192154, -30.645134],\n",
       "        [-83.192154, -30.645134],\n",
       "        [-83.19226 , -30.64483 ],\n",
       "        [-83.19215 , -30.645119],\n",
       "        [-83.19213 , -30.645105],\n",
       "        [-83.192116, -30.6451  ],\n",
       "        [-83.19212 , -30.645102],\n",
       "        [-83.19213 , -30.6451  ],\n",
       "        [-83.19212 , -30.645098],\n",
       "        [-83.19212 , -30.645103],\n",
       "        [-83.19213 , -30.645111],\n",
       "        [-83.19212 , -30.645103],\n",
       "        [-83.19212 , -30.645115],\n",
       "        [-83.192116, -30.645102],\n",
       "        [-83.19212 , -30.64511 ],\n",
       "        [-83.19213 , -30.645105],\n",
       "        [-83.19213 , -30.645103],\n",
       "        [-83.19212 , -30.645102],\n",
       "        [-83.19212 , -30.645092],\n",
       "        [-83.19212 , -30.645103],\n",
       "        [-83.192154, -30.64514 ],\n",
       "        [-83.19216 , -30.645124]], dtype=float32),\n",
       " Array([[-83.22854 , -30.676598],\n",
       "        [-83.23088 , -30.672749],\n",
       "        [-83.23175 , -30.679148],\n",
       "        [-83.23164 , -30.679113],\n",
       "        [-83.23158 , -30.678902],\n",
       "        [-83.23159 , -30.67889 ],\n",
       "        [-83.23159 , -30.678907],\n",
       "        [-83.23158 , -30.678896],\n",
       "        [-83.23158 , -30.678902],\n",
       "        [-83.23159 , -30.678904],\n",
       "        [-83.23158 , -30.678896],\n",
       "        [-83.23158 , -30.678894],\n",
       "        [-83.23158 , -30.678892],\n",
       "        [-83.23159 , -30.678917],\n",
       "        [-83.231606, -30.678928],\n",
       "        [-83.23157 , -30.678823],\n",
       "        [-83.23122 , -30.678185],\n",
       "        [-83.22098 , -30.669567],\n",
       "        [-83.22009 , -30.671675],\n",
       "        [-83.220566, -30.671947],\n",
       "        [-83.2206  , -30.671862],\n",
       "        [-83.22059 , -30.671848],\n",
       "        [-83.22059 , -30.671854],\n",
       "        [-83.220604, -30.671858],\n",
       "        [-83.220604, -30.671862],\n",
       "        [-83.2206  , -30.671843],\n",
       "        [-83.2206  , -30.67185 ],\n",
       "        [-83.2206  , -30.671879],\n",
       "        [-83.2206  , -30.671858],\n",
       "        [-83.22059 , -30.671764],\n",
       "        [-83.22058 , -30.671824],\n",
       "        [-83.22062 , -30.67233 ],\n",
       "        [-83.22063 , -30.671892],\n",
       "        [-83.220825, -30.66781 ],\n",
       "        [-83.21877 , -30.67069 ],\n",
       "        [-83.19317 , -30.763945]], dtype=float32)]"
      ]
     },
     "execution_count": 139,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_nodes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Array(3990.5747, dtype=float32)"
      ]
     },
     "execution_count": 140,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "eval_metrics.loss.total"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Yep, as expected, we verify both are true. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The function `evaluate_model` does similar things except it computes loss for all windows in the dataset (not just a single window like what we passed in above) and it does so for multiple modes that you specify via the `splits` param (e.g. `splits = ['val', 'test']`). This is used in the `train_and_evaluate` pipeline and would also come in handy for evaluating over the testing dataset at the very end of model development. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### FULL TRAIN PIPELINE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The function `train_and_evaluate` runs the entire training pipeline, start to end for the number of epochs specified in the config, and logs a lot of stuff. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.jraph_training import train_and_evaluate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set up logging\n",
    "import logging\n",
    "logger = logging.getLogger()\n",
    "logger.setLevel(logging.INFO)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:absl:Obtaining datasets.\n",
      "INFO:absl:Hyperparameters: {'F': 8, 'K': 36, 'activation': 'relu', 'b': 10, 'c': 10, 'checkpoint_every_epochs': 1, 'dropout_rate': 0.1, 'edge_features': (4, 8), 'epochs': 2, 'eval_every_epochs': 1, 'fully_connected_edges': True, 'global_features': None, 'h': 1, 'init_buffer_samples': 0, 'input_steps': 3, 'layer_norm': False, 'learning_rate': 0.001, 'log_every_epochs': 1, 'max_checkpts_to_keep': None, 'model': 'MLPGraphNetwork', 'n_blocks': 1, 'n_samples': 20, 'node_features': (32, 2), 'normalize': True, 'optimizer': 'adam', 'output_delay': 0, 'output_steps': 2, 'sample_buffer': 1, 'seed': 42, 'share_params': False, 'skip_connections': False, 'test_pct': 0.1, 'time_resolution': 100, 'timestep_duration': 1, 'train_pct': 0.7, 'val_pct': 0.2}\n",
      "INFO:absl:Initializing network.\n",
      "INFO:absl:\n",
      "+----------------------------------------+----------+------+----------+-------+\n",
      "| Name                                   | Shape    | Size | Mean     | Std   |\n",
      "+----------------------------------------+----------+------+----------+-------+\n",
      "| params/MLPBlock_0/MLP_0/Dense_0/bias   | (4,)     | 4    | 0.0      | 0.0   |\n",
      "| params/MLPBlock_0/MLP_0/Dense_0/kernel | (6, 4)   | 24   | -0.0819  | 0.36  |\n",
      "| params/MLPBlock_0/MLP_0/Dense_1/bias   | (8,)     | 8    | 0.0      | 0.0   |\n",
      "| params/MLPBlock_0/MLP_0/Dense_1/kernel | (4, 8)   | 32   | -0.0639  | 0.439 |\n",
      "| params/MLPBlock_0/MLP_1/Dense_0/bias   | (32,)    | 32   | 0.0      | 0.0   |\n",
      "| params/MLPBlock_0/MLP_1/Dense_0/kernel | (19, 32) | 608  | -0.00843 | 0.219 |\n",
      "| params/MLPBlock_0/MLP_1/Dense_1/bias   | (2,)     | 2    | 0.0      | 0.0   |\n",
      "| params/MLPBlock_0/MLP_1/Dense_1/kernel | (32, 2)  | 64   | -0.00627 | 0.195 |\n",
      "+----------------------------------------+----------+------+----------+-------+\n",
      "Total: 774\n",
      "INFO:absl:Checkpoint.restore_or_initialize() ...\n",
      "INFO:absl:No checkpoint specified. Restore the latest checkpoint.\n",
      "INFO:absl:Checkpoint None does not exist.\n",
      "INFO:absl:Checkpoint.save() ...\n",
      "INFO:absl:Checkpoint.save() finished after 0.03s.\n",
      "INFO:absl:Checkpoint.restore_or_initialize() finished after 0.04s.\n",
      "INFO:absl:Starting training.\n",
      "ERROR:absl:Could not start profiling: Profile has already been started. Only one profile may be run at a time.\n",
      "Traceback (most recent call last):\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/periodic_actions.py\", line 327, in _start_session\n",
      "    profiler.start(logdir=self._logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/profiler.py\", line 38, in start\n",
      "    jax.profiler.start_trace(logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/jax/_src/profiler.py\", line 111, in start_trace\n",
      "    raise RuntimeError(\"Profile has already been started. \"\n",
      "RuntimeError: Profile has already been started. Only one profile may be run at a time.\n",
      "INFO:absl:[0] train_loss=3391.252197265625\n",
      "INFO:absl:Checkpoint.save() ...\n",
      "INFO:absl:[0] val_loss=1661.358642578125\n",
      "INFO:absl:[0] test_loss=1719.959228515625\n",
      "INFO:absl:Checkpoint.save() finished after 0.02s.\n",
      "INFO:absl:[1] train_loss=999.2669677734375\n",
      "INFO:absl:[1] val_loss=345.2799377441406\n",
      "INFO:absl:Checkpoint.save() ...\n",
      "INFO:absl:[1] test_loss=352.36181640625\n",
      "INFO:absl:Checkpoint.save() finished after 0.02s.\n"
     ]
    }
   ],
   "source": [
    "workdir=\"tests/outputs/demo\"\n",
    "\n",
    "trained_state, train_metrics, eval_metrics_dict = train_and_evaluate(\n",
    "    config=config, workdir=workdir)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now if you go to the workdir folder, you can see the checkpoints and tensorboard objects it created. Also notice in the logs that the loss gets better over the (two) epochs we trained for! "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### HYPERPARAMETER TUNING"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We use the `optuna` library for hyperparameter tuning, which is a flexible framework that can work with any ML library we use. \n",
    "\n",
    "terminology:\n",
    "* objective function: a function that returns the value you want to optimize (for example, an entire training pipeline, which trains a model for X epochs and returns the final validation loss. We want to minimize the loss by the end of training, and we need to tune the hyperparameters to do so.)\n",
    "* Parameter: A variable whose value is to be optimized. can be continuous, discrete, categorical, etc. (e.g. learning rate, number of GNBlocks chained together, choice of activation function, etc.)\n",
    "* Trial: An object which is like a single run of the objective function; in each trial, optuna generates different parameters.\n",
    "* Study: An optimization session, which is a set of trials. parameters vary across trials, and we can choose the best set of hyperparams by the end of the study. \n",
    "\n",
    "This tutorial explains how things work in more depth: https://uvadlc-notebooks.readthedocs.io/en/latest/tutorial_notebooks/guide4/Research_Projects_with_JAX.html#Automatic-hyperparameter-tuning-with-Optuna \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here is an example run: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "metadata": {},
   "outputs": [],
   "source": [
    "import optuna \n",
    "from utils.jraph_training import train_and_evaluate_with_data, create_dataset\n",
    "import os \n",
    "from functools import partial\n",
    "\n",
    "CHECKPOINT_PATH = 'experiments/tuning'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "define the objective function: "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 182,
   "metadata": {},
   "outputs": [],
   "source": [
    "def objective(trial, study_name, datasets):\n",
    "    \"\"\" Defines the objective function to be optimized over, aka the validation loss of a model.\n",
    "    \n",
    "        Args:\n",
    "            trial: object which characterizes the current run \n",
    "            datasets: dictionary of data. we explicitly pass this in so that we don't have to waste runtime regenerating the same dataset over and over. \n",
    "    \"\"\"\n",
    "    # create config \n",
    "    config = ml_collections.ConfigDict()\n",
    "\n",
    "    # Data params. \n",
    "    config.n_samples=20\n",
    "    config.input_steps=3\n",
    "    config.output_delay=0\n",
    "    config.output_steps=2\n",
    "    config.timestep_duration=1\n",
    "    config.sample_buffer=1\n",
    "    config.time_resolution=100\n",
    "    config.init_buffer_samples=0\n",
    "    config.train_pct=0.7\n",
    "    config.val_pct=0.2\n",
    "    config.test_pct=0.1\n",
    "    config.K=36\n",
    "    config.F=8\n",
    "    config.c=10\n",
    "    config.b=10\n",
    "    config.h=1\n",
    "    config.seed=42\n",
    "    config.normalize=True\n",
    "    config.fully_connected_edges=True\n",
    "\n",
    "    config.max_checkpts_to_keep = 2\n",
    "\n",
    "    # Optimizer.\n",
    "    # config.optimizer = \"adam\"\n",
    "    config.optimizer = trial.suggest_categorical(\"optimizer\", [\"adam\", \"sgd\"])\n",
    "    config.learning_rate = trial.suggest_float('learning_rate', 3e-4, 3e-3, \n",
    "                                               log=True)\n",
    "    if config.optimizer == \"sgd\":\n",
    "        config.momentum = trial.suggest_float('momentum', 0, 0.999) # upper bound is inclusive, and we want to exclude a momentum of 1 because that would yield no decay \n",
    "\n",
    "    # Training hyperparameters.\n",
    "    config.batch_size = 1 # variable currently not used\n",
    "    config.epochs = 4\n",
    "    config.log_every_epochs = 1\n",
    "    config.eval_every_epochs = 1\n",
    "    config.checkpoint_every_epochs = 1\n",
    "\n",
    "    # GNN hyperparameters.\n",
    "    config.model = 'MLPGraphNetwork'\n",
    "    config.n_blocks = trial.suggest_int('n_blocks', 1, 3)\n",
    "    config.share_params = False\n",
    "    config.dropout_rate = trial.suggest_float('dropout_rate', 0, 0.2)\n",
    "    config.skip_connections = False # This was throwing a broadcast error in add_graphs_tuples_nodes when this was set to True\n",
    "    config.layer_norm = False # TODO perhaps we want to turn on later\n",
    "    config.activation = trial.suggest_categorical(\n",
    "        'activation', [\"relu\", \"elu\", \"leaky_relu\"])\n",
    "    # config.activation = \"leaky_relu\"\n",
    "    \n",
    "    config.pred_x1 = True\n",
    "    config.pred_x2 = True\n",
    "\n",
    "    if config.pred_x1 and config.pred_x2:\n",
    "        output_layer = 2\n",
    "    else:\n",
    "        output_layer = 1\n",
    "\n",
    "    # choose the hidden layer feature size using powers of 2 \n",
    "    config.edge_features = (\n",
    "        2**trial.suggest_int(\"edge_mlp_1_power\", 1, 3), # range 2 - 8; upper bound is inclusive\n",
    "        2**trial.suggest_int(\"edge_mlp_2_power\", 1, 3), # range 2 - 8\n",
    "    )\n",
    "    config.node_features = (\n",
    "        2**trial.suggest_int(\"node_mlp_1_power\", 1, 6), \n",
    "        # 2**trial.suggest_int(\"node_mlp_2_power\", 1, 6), \n",
    "        output_layer) \n",
    "    # note the last feature size will be the number of features that the graph predicts\n",
    "    config.global_features = None\n",
    "\n",
    "    # generate a workdir \n",
    "    # TODO: check if we actually care about referencing this in the future or if we can just create a temp dir \n",
    "    workdir=os.path.join(CHECKPOINT_PATH, study_name, f\"trial_{trial.number}\")\n",
    "\n",
    "    # run training \n",
    "    state, train_metrics, eval_metrics_dict = train_and_evaluate_with_data(config=config, workdir=workdir, datasets=datasets, trial=trial)\n",
    "    \n",
    "    # retrieve and return val loss (MSE)\n",
    "    return eval_metrics_dict['val'].compute()['loss']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 183,
   "metadata": {},
   "outputs": [],
   "source": [
    "# generate dataset \n",
    "datasets = create_dataset(config)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 184,
   "metadata": {},
   "outputs": [],
   "source": [
    "def prepare_study(study_name):\n",
    "    # get the objective function that reuses the pre-generated datasets \n",
    "    objective_partial = partial(objective, study_name=study_name, \n",
    "                                datasets=datasets)\n",
    "\n",
    "    # run optimization study\n",
    "    db_path = os.path.join(CHECKPOINT_PATH, study_name, \"optuna_hparam_search.db\")\n",
    "    if not os.path.exists(os.path.join(CHECKPOINT_PATH, study_name)):\n",
    "        os.makedirs(os.path.join(CHECKPOINT_PATH, study_name))\n",
    "\n",
    "    study = optuna.create_study(\n",
    "        study_name=study_name,\n",
    "        storage=f'sqlite:///{db_path}', # generates a new db if it doesn't exist\n",
    "        direction='minimize',\n",
    "        pruner=optuna.pruners.MedianPruner(\n",
    "            n_startup_trials=5, \n",
    "            n_warmup_steps=50,\n",
    "            ), \n",
    "        load_if_exists=True, \n",
    "    )\n",
    "    \n",
    "    return study, objective_partial"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create the study. Then run the optimization. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 185,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-02-06 10:58:20,747] A new study created in RDB with name: demo\n"
     ]
    }
   ],
   "source": [
    "# create the study\n",
    "study, objective_partial = prepare_study(study_name=\"demo\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 186,
   "metadata": {},
   "outputs": [],
   "source": [
    "# turn off logging because it'll get annoying \n",
    "logger.setLevel(logging.WARNING)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 187,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "ERROR:absl:Could not start profiling: Profile has already been started. Only one profile may be run at a time.\n",
      "Traceback (most recent call last):\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/periodic_actions.py\", line 327, in _start_session\n",
      "    profiler.start(logdir=self._logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/profiler.py\", line 38, in start\n",
      "    jax.profiler.start_trace(logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/jax/_src/profiler.py\", line 111, in start_trace\n",
      "    raise RuntimeError(\"Profile has already been started. \"\n",
      "RuntimeError: Profile has already been started. Only one profile may be run at a time.\n",
      "[I 2024-02-06 10:58:30,887] Trial 0 finished with value: 502.29071044921875 and parameters: {'optimizer': 'adam', 'learning_rate': 0.0009468917515128164, 'n_blocks': 1, 'dropout_rate': 0.07210241902162905, 'activation': 'relu', 'edge_mlp_1_power': 3, 'edge_mlp_2_power': 1, 'node_mlp_1_power': 2}. Best is trial 0 with value: 502.29071044921875.\n",
      "ERROR:absl:Could not start profiling: Profile has already been started. Only one profile may be run at a time.\n",
      "Traceback (most recent call last):\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/periodic_actions.py\", line 327, in _start_session\n",
      "    profiler.start(logdir=self._logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/profiler.py\", line 38, in start\n",
      "    jax.profiler.start_trace(logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/jax/_src/profiler.py\", line 111, in start_trace\n",
      "    raise RuntimeError(\"Profile has already been started. \"\n",
      "RuntimeError: Profile has already been started. Only one profile may be run at a time.\n",
      "[I 2024-02-06 10:58:47,288] Trial 1 finished with value: 5.622570037841797 and parameters: {'optimizer': 'adam', 'learning_rate': 0.00233126579667957, 'n_blocks': 3, 'dropout_rate': 0.045360734021383325, 'activation': 'relu', 'edge_mlp_1_power': 2, 'edge_mlp_2_power': 2, 'node_mlp_1_power': 1}. Best is trial 1 with value: 5.622570037841797.\n",
      "ERROR:absl:Could not start profiling: Profile has already been started. Only one profile may be run at a time.\n",
      "Traceback (most recent call last):\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/periodic_actions.py\", line 327, in _start_session\n",
      "    profiler.start(logdir=self._logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/profiler.py\", line 38, in start\n",
      "    jax.profiler.start_trace(logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/jax/_src/profiler.py\", line 111, in start_trace\n",
      "    raise RuntimeError(\"Profile has already been started. \"\n",
      "RuntimeError: Profile has already been started. Only one profile may be run at a time.\n",
      "[I 2024-02-06 10:58:57,852] Trial 2 finished with value: 631.892578125 and parameters: {'optimizer': 'adam', 'learning_rate': 0.0007231905147257836, 'n_blocks': 2, 'dropout_rate': 0.020073194874270464, 'activation': 'relu', 'edge_mlp_1_power': 3, 'edge_mlp_2_power': 1, 'node_mlp_1_power': 4}. Best is trial 1 with value: 5.622570037841797.\n",
      "ERROR:absl:Could not start profiling: Profile has already been started. Only one profile may be run at a time.\n",
      "Traceback (most recent call last):\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/periodic_actions.py\", line 327, in _start_session\n",
      "    profiler.start(logdir=self._logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/profiler.py\", line 38, in start\n",
      "    jax.profiler.start_trace(logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/jax/_src/profiler.py\", line 111, in start_trace\n",
      "    raise RuntimeError(\"Profile has already been started. \"\n",
      "RuntimeError: Profile has already been started. Only one profile may be run at a time.\n",
      "[I 2024-02-06 10:59:03,911] Trial 3 finished with value: 11.934476852416992 and parameters: {'optimizer': 'adam', 'learning_rate': 0.002178545756841023, 'n_blocks': 1, 'dropout_rate': 0.10228770459337273, 'activation': 'elu', 'edge_mlp_1_power': 1, 'edge_mlp_2_power': 3, 'node_mlp_1_power': 6}. Best is trial 1 with value: 5.622570037841797.\n",
      "ERROR:absl:Could not start profiling: Profile has already been started. Only one profile may be run at a time.\n",
      "Traceback (most recent call last):\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/periodic_actions.py\", line 327, in _start_session\n",
      "    profiler.start(logdir=self._logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/clu/profiler.py\", line 38, in start\n",
      "    jax.profiler.start_trace(logdir)\n",
      "  File \"/Users/h.lu/Documents/_code/_research lorenz code/lorenzGNN/lorenzvenv/lib/python3.9/site-packages/jax/_src/profiler.py\", line 111, in start_trace\n",
      "    raise RuntimeError(\"Profile has already been started. \"\n",
      "RuntimeError: Profile has already been started. Only one profile may be run at a time.\n",
      "[I 2024-02-06 10:59:09,745] Trial 4 finished with value: 2186.28955078125 and parameters: {'optimizer': 'adam', 'learning_rate': 0.0003321442204132884, 'n_blocks': 1, 'dropout_rate': 0.18057515225952836, 'activation': 'relu', 'edge_mlp_1_power': 3, 'edge_mlp_2_power': 2, 'node_mlp_1_power': 4}. Best is trial 1 with value: 5.622570037841797.\n"
     ]
    }
   ],
   "source": [
    "# run the optimization \n",
    "study.optimize(objective_partial, \n",
    "                n_trials=5-len(study.trials), \n",
    "                n_jobs=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then we can look at the different trials in the study and choose the best one. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 188,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[FrozenTrial(number=0, state=TrialState.COMPLETE, values=[502.29071044921875], datetime_start=datetime.datetime(2024, 2, 6, 10, 58, 23, 308016), datetime_complete=datetime.datetime(2024, 2, 6, 10, 58, 30, 852128), params={'optimizer': 'adam', 'learning_rate': 0.0009468917515128164, 'n_blocks': 1, 'dropout_rate': 0.07210241902162905, 'activation': 'relu', 'edge_mlp_1_power': 3, 'edge_mlp_2_power': 1, 'node_mlp_1_power': 2}, user_attrs={}, system_attrs={}, intermediate_values={0: 1555.1356201171875, 1: 1050.9931640625, 2: 720.1218872070312, 3: 502.29071044921875}, distributions={'optimizer': CategoricalDistribution(choices=('adam', 'sgd')), 'learning_rate': FloatDistribution(high=0.003, log=True, low=0.0003, step=None), 'n_blocks': IntDistribution(high=3, log=False, low=1, step=1), 'dropout_rate': FloatDistribution(high=0.2, log=False, low=0.0, step=None), 'activation': CategoricalDistribution(choices=('relu', 'elu', 'leaky_relu')), 'edge_mlp_1_power': IntDistribution(high=3, log=False, low=1, step=1), 'edge_mlp_2_power': IntDistribution(high=3, log=False, low=1, step=1), 'node_mlp_1_power': IntDistribution(high=6, log=False, low=1, step=1)}, trial_id=1, value=None),\n",
       " FrozenTrial(number=1, state=TrialState.COMPLETE, values=[5.622570037841797], datetime_start=datetime.datetime(2024, 2, 6, 10, 58, 30, 910123), datetime_complete=datetime.datetime(2024, 2, 6, 10, 58, 47, 269558), params={'optimizer': 'adam', 'learning_rate': 0.00233126579667957, 'n_blocks': 3, 'dropout_rate': 0.045360734021383325, 'activation': 'relu', 'edge_mlp_1_power': 2, 'edge_mlp_2_power': 2, 'node_mlp_1_power': 1}, user_attrs={}, system_attrs={}, intermediate_values={0: 14.976932525634766, 1: 5.6012067794799805, 2: 5.5649614334106445, 3: 5.622570037841797}, distributions={'optimizer': CategoricalDistribution(choices=('adam', 'sgd')), 'learning_rate': FloatDistribution(high=0.003, log=True, low=0.0003, step=None), 'n_blocks': IntDistribution(high=3, log=False, low=1, step=1), 'dropout_rate': FloatDistribution(high=0.2, log=False, low=0.0, step=None), 'activation': CategoricalDistribution(choices=('relu', 'elu', 'leaky_relu')), 'edge_mlp_1_power': IntDistribution(high=3, log=False, low=1, step=1), 'edge_mlp_2_power': IntDistribution(high=3, log=False, low=1, step=1), 'node_mlp_1_power': IntDistribution(high=6, log=False, low=1, step=1)}, trial_id=2, value=None),\n",
       " FrozenTrial(number=2, state=TrialState.COMPLETE, values=[631.892578125], datetime_start=datetime.datetime(2024, 2, 6, 10, 58, 47, 302272), datetime_complete=datetime.datetime(2024, 2, 6, 10, 58, 57, 831878), params={'optimizer': 'adam', 'learning_rate': 0.0007231905147257836, 'n_blocks': 2, 'dropout_rate': 0.020073194874270464, 'activation': 'relu', 'edge_mlp_1_power': 3, 'edge_mlp_2_power': 1, 'node_mlp_1_power': 4}, user_attrs={}, system_attrs={}, intermediate_values={0: 29040.623046875, 1: 3601.515869140625, 2: 1434.814697265625, 3: 631.892578125}, distributions={'optimizer': CategoricalDistribution(choices=('adam', 'sgd')), 'learning_rate': FloatDistribution(high=0.003, log=True, low=0.0003, step=None), 'n_blocks': IntDistribution(high=3, log=False, low=1, step=1), 'dropout_rate': FloatDistribution(high=0.2, log=False, low=0.0, step=None), 'activation': CategoricalDistribution(choices=('relu', 'elu', 'leaky_relu')), 'edge_mlp_1_power': IntDistribution(high=3, log=False, low=1, step=1), 'edge_mlp_2_power': IntDistribution(high=3, log=False, low=1, step=1), 'node_mlp_1_power': IntDistribution(high=6, log=False, low=1, step=1)}, trial_id=3, value=None),\n",
       " FrozenTrial(number=3, state=TrialState.COMPLETE, values=[11.934476852416992], datetime_start=datetime.datetime(2024, 2, 6, 10, 58, 57, 866174), datetime_complete=datetime.datetime(2024, 2, 6, 10, 59, 3, 877191), params={'optimizer': 'adam', 'learning_rate': 0.002178545756841023, 'n_blocks': 1, 'dropout_rate': 0.10228770459337273, 'activation': 'elu', 'edge_mlp_1_power': 1, 'edge_mlp_2_power': 3, 'node_mlp_1_power': 6}, user_attrs={}, system_attrs={}, intermediate_values={0: 472.70074462890625, 1: 87.14776611328125, 2: 22.471092224121094, 3: 11.934476852416992}, distributions={'optimizer': CategoricalDistribution(choices=('adam', 'sgd')), 'learning_rate': FloatDistribution(high=0.003, log=True, low=0.0003, step=None), 'n_blocks': IntDistribution(high=3, log=False, low=1, step=1), 'dropout_rate': FloatDistribution(high=0.2, log=False, low=0.0, step=None), 'activation': CategoricalDistribution(choices=('relu', 'elu', 'leaky_relu')), 'edge_mlp_1_power': IntDistribution(high=3, log=False, low=1, step=1), 'edge_mlp_2_power': IntDistribution(high=3, log=False, low=1, step=1), 'node_mlp_1_power': IntDistribution(high=6, log=False, low=1, step=1)}, trial_id=4, value=None),\n",
       " FrozenTrial(number=4, state=TrialState.COMPLETE, values=[2186.28955078125], datetime_start=datetime.datetime(2024, 2, 6, 10, 59, 3, 934065), datetime_complete=datetime.datetime(2024, 2, 6, 10, 59, 9, 725793), params={'optimizer': 'adam', 'learning_rate': 0.0003321442204132884, 'n_blocks': 1, 'dropout_rate': 0.18057515225952836, 'activation': 'relu', 'edge_mlp_1_power': 3, 'edge_mlp_2_power': 2, 'node_mlp_1_power': 4}, user_attrs={}, system_attrs={}, intermediate_values={0: 4319.21533203125, 1: 3418.157958984375, 2: 2723.766357421875, 3: 2186.28955078125}, distributions={'optimizer': CategoricalDistribution(choices=('adam', 'sgd')), 'learning_rate': FloatDistribution(high=0.003, log=True, low=0.0003, step=None), 'n_blocks': IntDistribution(high=3, log=False, low=1, step=1), 'dropout_rate': FloatDistribution(high=0.2, log=False, low=0.0, step=None), 'activation': CategoricalDistribution(choices=('relu', 'elu', 'leaky_relu')), 'edge_mlp_1_power': IntDistribution(high=3, log=False, low=1, step=1), 'edge_mlp_2_power': IntDistribution(high=3, log=False, low=1, step=1), 'node_mlp_1_power': IntDistribution(high=6, log=False, low=1, step=1)}, trial_id=5, value=None)]"
      ]
     },
     "execution_count": 188,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "study.trials"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 189,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.hyperparam_tuning import get_best_trial_config, get_best_trial_workdir\n",
    "from utils.jraph_vis import plot_predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 194,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1500x1000 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_predictions(\n",
    "    config=get_best_trial_config(study=study),\n",
    "    workdir=get_best_trial_workdir(study=study), # for loading checkpoints \n",
    "    plot_ith_rollout_step=0, # 0 indexed # for this study, we have a 4-step rollout \n",
    "    # dataset,\n",
    "    # preds,\n",
    "    # timestep_duration,\n",
    "    # n_rollout_steps,\n",
    "    #  total_steps,\n",
    "    node=0, # 0-indexed \n",
    "    plot_mode=\"val\", # i.e. \"train\"/\"val\"/\"test\"\n",
    "    datasets=datasets,\n",
    "    # plot_days=60,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Closing Remarks"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Library documentation that I had to reference a lot: \n",
    "\n",
    "* [jraph](https://jraph.readthedocs.io/en/latest/index.html)\n",
    "* [jax](https://jax.readthedocs.io/en/latest/index.html)\n",
    "* [flax](https://flax.readthedocs.io/en/latest/)\n",
    "* [optuna](https://optuna.readthedocs.io/en/stable/index.html)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "lorenzvenv",
   "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": "3.9.17"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}