examples/Two_Moons_Starter.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "009b6adf",
   "metadata": {},
   "source": [
    "# Two Moons: Tackling Bimodal Posteriors\n",
    "\n",
    "_Authors: Lars Kühmichel, Marvin Schmitt, Valentin Pratz, Stefan T. Radev_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "d5f88a59",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-10-24T08:36:22.149034Z",
     "start_time": "2024-10-24T08:36:20.807192Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2025-02-11 16:22:30.912532: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.\n",
      "2025-02-11 16:22:30.912881: I external/local_tsl/tsl/cuda/cudart_stub.cc:32] Could not find cuda drivers on your machine, GPU will not be used.\n",
      "2025-02-11 16:22:30.914936: I external/local_tsl/tsl/cuda/cudart_stub.cc:32] Could not find cuda drivers on your machine, GPU will not be used.\n",
      "2025-02-11 16:22:30.939014: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n",
      "To enable the following instructions: AVX2 AVX_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2025-02-11 16:22:31.425109: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT\n",
      "2025-02-11 16:22:31.935372: I external/local_xla/xla/stream_executor/cuda/cuda_executor.cc:998] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355\n",
      "2025-02-11 16:22:31.935638: W tensorflow/core/common_runtime/gpu/gpu_device.cc:2251] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.\n",
      "Skipping registering GPU devices...\n"
     ]
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "\n",
    "import bayesflow as bf"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c63b26ba",
   "metadata": {},
   "source": [
    "## Simulator<a class=\"anchor\" id=\"simulator\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9525ffd7",
   "metadata": {},
   "source": [
    "This example will demonstrate amortized estimation of a somewhat strange Bayesian model, whose posterior evaluated at the origin $x = (0, 0)$ of the \"data\" will resemble two crescent moons. The forward process is a noisy non-linear transformation on a 2D plane:\n",
    "\n",
    "$$\n",
    "\\begin{align}\n",
    "x_1 &= -|\\theta_1 + \\theta_2|/\\sqrt{2} + r \\cos(\\alpha) + 0.25\\\\\n",
    "x_2 &= (-\\theta_1 + \\theta_2)/\\sqrt{2} + r\\sin{\\alpha}\n",
    "\\end{align}\n",
    "$$\n",
    "\n",
    "with $x = (x_1, x_2)$ playing the role of \"observables\" (data to be learned from), $\\alpha \\sim \\text{Uniform}(-\\pi/2, \\pi/2)$, and $r \\sim \\text{Normal}(0.1, 0.01)$ being latent variables creating noise in the data, and $\\theta = (\\theta_1, \\theta_2)$ being the parameters that we will later seek to infer from new $x$. We set their priors to\n",
    "\n",
    "$$\n",
    "\\begin{align}\n",
    "\\theta_1, \\theta_2 \\sim \\text{Uniform}(-1, 1).\n",
    "\\end{align}\n",
    "$$\n",
    "\n",
    "This model is typically used for benchmarking simulation-based inference (SBI) methods (see https://arxiv.org/pdf/2101.04653) and any method for amortized Bayesian inference should be capable of recovering the two moons posterior *without* using a gazillion of simulations. Note, that this is a considerably harder task than modeling the common unconditional two moons data set used often in the context of normalizing flows."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "21bf228e706a010",
   "metadata": {},
   "source": [
    "BayesFlow offers many ways to define your data generating process. Here, we use sequential functions to build a simulator object for online training. Within this composite simulator, each function has access to the outputs of the previous functions. This effectively allows you to define any generative graph."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "f761b142a0e1da66",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-10-24T08:36:22.160542Z",
     "start_time": "2024-10-24T08:36:22.155240Z"
    }
   },
   "outputs": [],
   "source": [
    "def theta_prior():\n",
    "    theta = np.random.uniform(-1, 1, 2)\n",
    "    return dict(theta=theta)\n",
    "\n",
    "def forward_model(theta):\n",
    "    alpha = np.random.uniform(-np.pi / 2, np.pi / 2)\n",
    "    r = np.random.normal(0.1, 0.01)\n",
    "    x1 = -np.abs(theta[0] + theta[1]) / np.sqrt(2) + r * np.cos(alpha) + 0.25\n",
    "    x2 = (-theta[0] + theta[1]) / np.sqrt(2) + r * np.sin(alpha)\n",
    "    return dict(x=np.array([x1, x2]))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "722cb773",
   "metadata": {},
   "source": [
    "Within the composite simulator, every simulator has access to the outputs of the previous simulators in the list. For example, the last simulator `forward_model` has access to the outputs of the three other simulators."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "4b89c861527c13b8",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-10-24T08:36:22.305265Z",
     "start_time": "2024-10-24T08:36:22.301546Z"
    }
   },
   "outputs": [],
   "source": [
    "simulator = bf.make_simulator([theta_prior, forward_model])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f6e1eb5777c59eba",
   "metadata": {},
   "source": [
    "Let's generate some data to see what the simulator does:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "e6218e61d529e357",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-10-24T08:36:22.350483Z",
     "start_time": "2024-10-24T08:36:22.345161Z"
    }
   },
   "outputs": [],
   "source": [
    "# generate 3 random draws from the joint distribution p(r, alpha, theta, x)\n",
    "sample_data = simulator.sample(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "46174ccb0167026c",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-10-24T08:36:22.470435Z",
     "start_time": "2024-10-24T08:36:22.464836Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Type of sample_data:\n",
      "\t <class 'dict'>\n",
      "Keys of sample_data:\n",
      "\t dict_keys(['theta', 'x'])\n",
      "Types of sample_data values:\n",
      "\t {'theta': <class 'numpy.ndarray'>, 'x': <class 'numpy.ndarray'>}\n",
      "Shapes of sample_data values:\n",
      "\t {'theta': (3, 2), 'x': (3, 2)}\n"
     ]
    }
   ],
   "source": [
    "print(\"Type of sample_data:\\n\\t\", type(sample_data))\n",
    "print(\"Keys of sample_data:\\n\\t\", sample_data.keys())\n",
    "print(\"Types of sample_data values:\\n\\t\", {k: type(v) for k, v in sample_data.items()})\n",
    "print(\"Shapes of sample_data values:\\n\\t\", {k: v.shape for k, v in sample_data.items()})"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "17f158bd2d7abf75",
   "metadata": {},
   "source": [
    "BayesFlow also provides this simulator and a collection of others in the `bayesflow.benchmarks` module."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f714c3a178b5a375",
   "metadata": {},
   "source": [
    "## Adapter\n",
    "\n",
    "The next step is to tell BayesFlow how to deal with all the simulated variables. You may also think of this as informing BayesFlow about the data flow, i.e., which variables go into which network and what transformations needs to be performed prior to passing the simulator outputs into the networks. This is done via an adapter layer, which is implemented as a sequence of fixed, pseudo-invertible data transforms.\n",
    "\n",
    "Below, we define the data adapter by specifying the input and output keys and the transformations to be applied. This allows us full control over the data flow."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "5c9c2dc70f53d103",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-10-24T08:36:26.618926Z",
     "start_time": "2024-10-24T08:36:26.614443Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Adapter([0: ToArray -> 1: ConvertDType -> 2: Standardize(exclude=['theta']) -> 3: Rename('theta' -> 'inference_variables') -> 4: Rename('x' -> 'inference_conditions')])"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adapter = (\n",
    "    bf.adapters.Adapter()\n",
    "    \n",
    "    # convert any non-arrays to numpy arrays\n",
    "    .to_array()\n",
    "    \n",
    "    # convert from numpy's default float64 to deep learning friendly float32\n",
    "    .convert_dtype(\"float64\", \"float32\")\n",
    "    \n",
    "    # standardize target variables to zero mean and unit variance \n",
    "    .standardize(exclude=\"theta\")\n",
    "    \n",
    "    # rename the variables to match the required approximator inputs\n",
    "    .rename(\"theta\", \"inference_variables\")\n",
    "    .rename(\"x\", \"inference_conditions\")\n",
    ")\n",
    "adapter"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "254e287b2bccdad",
   "metadata": {},
   "source": [
    "## Dataset\n",
    "\n",
    "For this example, we will sample our training data ahead of time and use offline training with a very small number of epochs. In actual applications, you usually want to train much longer in order to max our performance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "39cb5a1c9824246f",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-09-23T14:39:46.950573Z",
     "start_time": "2024-09-23T14:39:46.948624Z"
    }
   },
   "outputs": [],
   "source": [
    "num_training_batches = 512\n",
    "num_validation_batches = 128\n",
    "batch_size = 64\n",
    "epochs = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "9dee7252ef99affa",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-09-23T14:39:53.268860Z",
     "start_time": "2024-09-23T14:39:46.994697Z"
    }
   },
   "outputs": [],
   "source": [
    "training_data = simulator.sample(num_training_batches * batch_size,)\n",
    "validation_data = simulator.sample(num_validation_batches * batch_size,)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2d4c6eb0",
   "metadata": {},
   "source": [
    "## Training a neural network to approximate all posteriors\n",
    "\n",
    "The next step is to set up the neural network that will approximate the posterior $p(\\theta\\,|\\,x)$.\n",
    "\n",
    "We choose **Flow Matching** [1, 2] as the backbone architecture for this example, as it can deal well with the multimodal nature of the posteriors that some observables imply.\n",
    "\n",
    "* [1] Lipman, Y., Chen, R. T., Ben-Hamu, H., Nickel, M., & Le, M. Flow Matching for Generative Modeling. In *The Eleventh International Conference on Learning Representations*.\n",
    "\n",
    "* [2] Wildberger, J. B., Dax, M., Buchholz, S., Green, S. R., Macke, J. H., & Schölkopf, B. Flow Matching for Scalable Simulation-Based Inference. In *Thirty-seventh Conference on Neural Information Processing Systems*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "09206e6f",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-09-23T14:39:53.339590Z",
     "start_time": "2024-09-23T14:39:53.319852Z"
    }
   },
   "outputs": [],
   "source": [
    "flow_matching = bf.networks.FlowMatching(\n",
    "    subnet=\"mlp\", \n",
    "    subnet_kwargs={\"dropout\": 0.0, \"widths\": (256,)*6} # override default dropout = 0.05 and widths = (256,)*5\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "851e522f",
   "metadata": {},
   "source": [
    "This inference network is just a general Flow Matching backbone, not yet adapted to the specific inference task at hand (i.e., posterior appproximation). To achieve this adaptation, we combine the network with our data adapter, which together form an `approximator`. In this case, we need a `ContinuousApproximator` since the target we want to approximate is the posterior of the *continuous* parameter vector $\\theta$."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "76722c33",
   "metadata": {},
   "source": [
    "### Basic Workflow\n",
    "We can hide many of the traditional deep learning steps (e.g., specifying a learning rate and an optimizer) within a `Workflow` object. This object just wraps everything together and includes some nice utility functions for training and *in silico* validation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "96ca6ffa",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-09-23T14:39:53.371691Z",
     "start_time": "2024-09-23T14:39:53.369375Z"
    }
   },
   "outputs": [],
   "source": [
    "flow_matching_workflow = bf.BasicWorkflow(\n",
    "    simulator=simulator,\n",
    "    adapter=adapter,\n",
    "    inference_network=flow_matching,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "708b1303",
   "metadata": {},
   "source": [
    "### Training\n",
    "\n",
    "We are ready to train our deep posterior approximator on the two moons example. We use the utility function `fit_offline`, which wraps the approximator's super flexible `fit` method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "0f496bda",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-09-23T14:42:36.067393Z",
     "start_time": "2024-09-23T14:39:53.513436Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:bayesflow:Fitting on dataset instance of OfflineDataset.\n",
      "INFO:bayesflow:Building on a test batch.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 3ms/step - loss: 1.5975 - loss/inference_loss: 1.5975 - val_loss: 0.3615 - val_loss/inference_loss: 0.3615\n",
      "Epoch 2/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3963 - loss/inference_loss: 0.3963 - val_loss: 0.3259 - val_loss/inference_loss: 0.3259\n",
      "Epoch 3/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3793 - loss/inference_loss: 0.3793 - val_loss: 0.3774 - val_loss/inference_loss: 0.3774\n",
      "Epoch 4/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3734 - loss/inference_loss: 0.3734 - val_loss: 0.2816 - val_loss/inference_loss: 0.2816\n",
      "Epoch 5/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3625 - loss/inference_loss: 0.3625 - val_loss: 0.2620 - val_loss/inference_loss: 0.2620\n",
      "Epoch 6/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3600 - loss/inference_loss: 0.3600 - val_loss: 0.2465 - val_loss/inference_loss: 0.2465\n",
      "Epoch 7/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3646 - loss/inference_loss: 0.3646 - val_loss: 0.3418 - val_loss/inference_loss: 0.3418\n",
      "Epoch 8/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3593 - loss/inference_loss: 0.3593 - val_loss: 0.3830 - val_loss/inference_loss: 0.3830\n",
      "Epoch 9/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3493 - loss/inference_loss: 0.3493 - val_loss: 0.4498 - val_loss/inference_loss: 0.4498\n",
      "Epoch 10/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3492 - loss/inference_loss: 0.3492 - val_loss: 0.3947 - val_loss/inference_loss: 0.3947\n",
      "Epoch 11/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3459 - loss/inference_loss: 0.3459 - val_loss: 0.2970 - val_loss/inference_loss: 0.2970\n",
      "Epoch 12/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3457 - loss/inference_loss: 0.3457 - val_loss: 0.2243 - val_loss/inference_loss: 0.2243\n",
      "Epoch 13/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3451 - loss/inference_loss: 0.3451 - val_loss: 0.4153 - val_loss/inference_loss: 0.4153\n",
      "Epoch 14/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3433 - loss/inference_loss: 0.3433 - val_loss: 0.3919 - val_loss/inference_loss: 0.3919\n",
      "Epoch 15/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3373 - loss/inference_loss: 0.3373 - val_loss: 0.3400 - val_loss/inference_loss: 0.3400\n",
      "Epoch 16/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3440 - loss/inference_loss: 0.3440 - val_loss: 0.1885 - val_loss/inference_loss: 0.1885\n",
      "Epoch 17/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3429 - loss/inference_loss: 0.3429 - val_loss: 0.3297 - val_loss/inference_loss: 0.3297\n",
      "Epoch 18/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3389 - loss/inference_loss: 0.3389 - val_loss: 0.3932 - val_loss/inference_loss: 0.3932\n",
      "Epoch 19/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3330 - loss/inference_loss: 0.3330 - val_loss: 0.2640 - val_loss/inference_loss: 0.2640\n",
      "Epoch 20/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3272 - loss/inference_loss: 0.3272 - val_loss: 0.3262 - val_loss/inference_loss: 0.3262\n",
      "Epoch 21/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3301 - loss/inference_loss: 0.3301 - val_loss: 0.4301 - val_loss/inference_loss: 0.4301\n",
      "Epoch 22/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3329 - loss/inference_loss: 0.3329 - val_loss: 0.4407 - val_loss/inference_loss: 0.4407\n",
      "Epoch 23/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3384 - loss/inference_loss: 0.3384 - val_loss: 0.2786 - val_loss/inference_loss: 0.2786\n",
      "Epoch 24/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3283 - loss/inference_loss: 0.3283 - val_loss: 0.3840 - val_loss/inference_loss: 0.3840\n",
      "Epoch 25/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3235 - loss/inference_loss: 0.3235 - val_loss: 0.3168 - val_loss/inference_loss: 0.3168\n",
      "Epoch 26/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3227 - loss/inference_loss: 0.3227 - val_loss: 0.2289 - val_loss/inference_loss: 0.2289\n",
      "Epoch 27/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3367 - loss/inference_loss: 0.3367 - val_loss: 0.2283 - val_loss/inference_loss: 0.2283\n",
      "Epoch 28/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3331 - loss/inference_loss: 0.3331 - val_loss: 0.3331 - val_loss/inference_loss: 0.3331\n",
      "Epoch 29/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3281 - loss/inference_loss: 0.3281 - val_loss: 0.1447 - val_loss/inference_loss: 0.1447\n",
      "Epoch 30/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3228 - loss/inference_loss: 0.3228 - val_loss: 0.2868 - val_loss/inference_loss: 0.2868\n",
      "Epoch 31/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3226 - loss/inference_loss: 0.3226 - val_loss: 0.2819 - val_loss/inference_loss: 0.2819\n",
      "Epoch 32/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3167 - loss/inference_loss: 0.3167 - val_loss: 0.3676 - val_loss/inference_loss: 0.3676\n",
      "Epoch 33/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3264 - loss/inference_loss: 0.3264 - val_loss: 0.2303 - val_loss/inference_loss: 0.2303\n",
      "Epoch 34/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3141 - loss/inference_loss: 0.3141 - val_loss: 0.2125 - val_loss/inference_loss: 0.2125\n",
      "Epoch 35/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3096 - loss/inference_loss: 0.3096 - val_loss: 0.2754 - val_loss/inference_loss: 0.2754\n",
      "Epoch 36/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3187 - loss/inference_loss: 0.3187 - val_loss: 0.3006 - val_loss/inference_loss: 0.3006\n",
      "Epoch 37/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3157 - loss/inference_loss: 0.3157 - val_loss: 0.3113 - val_loss/inference_loss: 0.3113\n",
      "Epoch 38/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3212 - loss/inference_loss: 0.3212 - val_loss: 0.4190 - val_loss/inference_loss: 0.4190\n",
      "Epoch 39/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3162 - loss/inference_loss: 0.3162 - val_loss: 0.3351 - val_loss/inference_loss: 0.3351\n",
      "Epoch 40/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3159 - loss/inference_loss: 0.3159 - val_loss: 0.3813 - val_loss/inference_loss: 0.3813\n",
      "Epoch 41/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3134 - loss/inference_loss: 0.3134 - val_loss: 0.4158 - val_loss/inference_loss: 0.4158\n",
      "Epoch 42/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3142 - loss/inference_loss: 0.3142 - val_loss: 0.5772 - val_loss/inference_loss: 0.5772\n",
      "Epoch 43/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3136 - loss/inference_loss: 0.3136 - val_loss: 0.3419 - val_loss/inference_loss: 0.3419\n",
      "Epoch 44/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 3ms/step - loss: 0.3104 - loss/inference_loss: 0.3104 - val_loss: 0.5761 - val_loss/inference_loss: 0.5761\n",
      "Epoch 45/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3166 - loss/inference_loss: 0.3166 - val_loss: 0.2678 - val_loss/inference_loss: 0.2678\n",
      "Epoch 46/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3197 - loss/inference_loss: 0.3197 - val_loss: 0.3576 - val_loss/inference_loss: 0.3576\n",
      "Epoch 47/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3165 - loss/inference_loss: 0.3165 - val_loss: 0.3637 - val_loss/inference_loss: 0.3637\n",
      "Epoch 48/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3145 - loss/inference_loss: 0.3145 - val_loss: 0.3348 - val_loss/inference_loss: 0.3348\n",
      "Epoch 49/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 4ms/step - loss: 0.3104 - loss/inference_loss: 0.3104 - val_loss: 0.3068 - val_loss/inference_loss: 0.3068\n",
      "Epoch 50/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 4ms/step - loss: 0.3090 - loss/inference_loss: 0.3090 - val_loss: 0.4407 - val_loss/inference_loss: 0.4407\n"
     ]
    }
   ],
   "source": [
    "history = flow_matching_workflow.fit_offline(\n",
    "    training_data, \n",
    "    epochs=epochs, \n",
    "    batch_size=batch_size, \n",
    "    validation_data=validation_data\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "66248a2f",
   "metadata": {},
   "source": [
    "## Swapping Inference Networks <a class=\"anchor\" id=\"swapping\"></a>\n",
    "\n",
    "Using BayesFlow, it is easy to switch to a different backbone architecture for the inference network. For instance, the code below demonstrates the use of a **Consistency Model**, which can allow for faster sampling during inference."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "80a97232",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2024-09-23T14:42:38.639240Z",
     "start_time": "2024-09-23T14:42:38.637439Z"
    }
   },
   "source": [
    "### Consistency Models: Background\n",
    "\n",
    "Consistency Models (CM; [1]) leverage the nice properties of score-based diffusion to enable few-step sampling. Score-based diffusion initially relied on a stochastic differential equation (SDE) for sampling, but there is also a ordinary (non-stochastic) differential equation (ODE)that has the same _marginal_ distribution at each time step $t$ [2]. This means that even though SDE and ODE produce different paths from the noise distribution to the target distribution, the resulting distributions when looking at many paths at time $t$ is the same. The ODE is also called Probability Flow ODE.\n",
    "\n",
    "CMs leverage the fact that there is no randomness in the ODE formulation. That means, if you start at a certain point in the latent space, you will always take the same path and end up at the same point in the target $\\theta$-space. The same is true for every point on the path: if you integrate to get to time $t=0$, you will end up at the same point as well. In short: for each path, there is exactly one corresponding point in latent space (at $t=T$) and one corresponding point in data space (at $t=0$). \n",
    "\n",
    "The goal of CMs is the following: each point at a time point $t$ belongs to exactly one path, and we want to predict where this path will end up at $t=0$. The function that does this is called the _consistency function_ $f$. If we have the correct function for all $t\\in(0,T]$, we can just sample from the latent distribution ($t=T$) and use $f$ to directly map to the corresponding point at $t=0$, which is in the target distribution. So for sampling from the target distribution, we avoid any integration and only need one evaluation of the consistency function. In practice, the one-step sampling does not work very well. Instead, we leverage a multi-step sampling method where we call $f$ multiple times. Please check out the [1] for more background on this sampling procedure.\n",
    "\n",
    "When reading the above, you might wonder why we also learn the mapping to $t=0$ of all intermediate time steps $t\\in[0, T]$, and not only for $t=T$. The main answer is that for efficient training, we do not want to actually compute the two associated points explicitly. Doing so would require to do a precise integration at training time, which is often not feasible as it is too computationally costly. Learning all time steps opens up the possibility for a different training approach where we can avoid this. The details of this become a bit more complicated, and we advise you to take a look at [1] if you are interested in a more thorough and mathematical discussion. Below, we will give a rough description of the underlying concepts.\n",
    "\n",
    "\n",
    "**Training** First, we know that at $t=0$, it holds that $f(\\theta,t=0)=\\theta$, as $\\theta$ is part of the path that ends at $\\theta$. This _boundary condition_ serves as an \"anchor\" for our training, this is the information that the network knows at the start of the training procedure (we encode it with a time-dependent skip-connection, so the network is forced to be the identity function at $t=0$). For training, we now somehow have to propagate this information to the rest of the part. The basic idea for this is simple. We just take a point $\\theta_1$ closer to the data distribution (smaller time $t_1$) and integrate for a small time step $dt$ to a point $\\theta_2$ on the same path that is closer to the latent distribution (larger time $t_2=t_1+dt$). As we know that for $t=0$ our network provides the correct output for our path, we want to propagate the information from smaller times to larger times. Our training goal is to move the output of $f(\\theta_2, t=t_2)$ towards the output of $f(\\theta_1, t=t_1)$. How to choose $\\theta_1$, $t_1$ and $dt$ is an empirical question, see the [1] for some thoughts on what works well.\n",
    "\n",
    "**Distilling inference** In the case of _distillation_, we start with a trained score-based diffusion model. We can use it to integrate the Probability Flow ODE to get from $\\theta_1$ to $\\theta_2$. If we do not have such a model, it seems as if we were stuck. We do not know which points lie on the same path, so we do not know which outputs to make similar. Fortunately, it turns out that there is an _unbiased approximator_ that, if averaged over many samples (check out the paper for the exact description), will also give us the correct score. If we use this approximator instead of the score model, and use only a single Euler step to move along the path, we get an algorithm similar to the one described for distillation. It is called Consistency Training (CT) and allows us to train a consistency model using only _samples_ from the data distribution. The algorithm for this was improved a lot in [3], and we have incorporated those improvements into our implementation.\n",
    "\n",
    "**Improving consistency training** We have made several approximations to get to a standalone *consistency training* algorithm. As a consequence, the introduced hyperparameters and their choice unfortunately becomes somewhat unintuitive. We have to rely on empirical observations and heuristics to see what works. This was done in [4], we encourage you to use the values provided there as starting points. If you happen to find hyperparameters that work significantly better, please let us know (e.g., by opening an issue or sending an email). This will help others to find the correct region in the hyperparameter space.\n",
    "\n",
    "[1] Song, Y., Dhariwal, P., Chen, M., & Sutskever, I. (2023). Consistency Models. _arXiv preprint_. [https://doi.org/10.48550/arXiv.2303.01469](https://doi.org/10.48550/arXiv.2303.01469)\n",
    "\n",
    "[2] Song, Y., Sohl-Dickstein, J., Kingma, D. P., Kumar, A., Ermon, S., & Poole, B. (2021). Score-Based Generative Modeling through Stochastic Differential Equations. In _International Conference on Learning Representations_. [https://openreview.net/forum?id=PxTIG12RRHS](https://openreview.net/forum?id=PxTIG12RRHS)\n",
    "\n",
    "[3] Song, Y., & Dhariwal, P. (2023). Improved Techniques for Training Consistency Models. _arXiv preprint_. [https://doi.org/10.48550/arXiv.2310.14189](https://doi.org/10.48550/arXiv.2310.14189)\n",
    "\n",
    "[4] Schmitt, M., Pratz, V., Köthe, U., Bürkner, P.-C., & Radev, S. T. (2024). Consistency Models for Scalable and Fast Simulation-Based Inference. _arXiv preprint_. [https://doi.org/10.48550/arXiv.2312.05440](https://doi.org/10.48550/arXiv.2312.05440)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "24b69117",
   "metadata": {},
   "source": [
    "### Consistency Models: Specification"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3016f08f",
   "metadata": {},
   "source": [
    "We can now go ahead and define our new inference network backbone. Apart from the usual parameters like learning rate and batch size, CMs come with a number of different hyperparameters. Unfortunately, they can heavily interact, so they can be hard to tune. The main hyperparameters are:\n",
    "\n",
    "- Maximum time `max_time`: This also serves as the standard deviation of the latent distribution. You can experiment with this, values from 10-200 seem to work well. In any case, it should be larger than the standard deviation of the target distribution.\n",
    "- Minimum/maximum number of discretization steps during training `s0`/`s1`: The effect of those is hard to grasp. 10 works well for `s0`. Intuitively, increasing `s1` along with the number of epochs should lead to better result, but in practice we sometimes observe a breakdown for high values of `s1`. This seems to be problem-dependent, so just try it out.\n",
    "- `sigma2` modifies the time-dependency of the skip connection. Its effect on the training is unclear, we recommend leaving it at 1.0 or setting it to the approximate variance of the target distribution.\n",
    "- Smallest time value `eps` ($t=\\epsilon$ is used instead of $t=0$ for numerical reasons): No large effect in our experiments, as long as it is kept small enough. Probably not worth tuning.\n",
    "\n",
    "You may find that different hyperparameter values work better for your tasks.\n",
    "\n",
    "A short note on dropout: in our experiments, dropout usually lead to worse performance, so generally we recommend setting the droput rate to $0.0$. Consistency training takes advantage of a noisy estimator of the score, so probably the training is already sufficiently noisy and extra dropout for regularization is not necessary."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "d53a41b8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Compute the empirical variance of the draws from the prior θ ~ p(θ)\n",
    "consistency_model = bf.networks.ConsistencyModel(\n",
    "    subnet=\"mlp\",\n",
    "    subnet_kwargs={\"dropout\": 0.0, \"widths\": (256,)*6},\n",
    "    total_steps=num_training_batches*epochs,\n",
    "    max_time=10, # this probably needs to be tuned for a novel application\n",
    "    sigma2=1.0,  # the data adapter standardizes our parameters, so set to 1.0\n",
    ")\n",
    "\n",
    "# Workflow for consistency model\n",
    "consistency_model_workflow = bf.BasicWorkflow(\n",
    "    simulator=simulator,\n",
    "    adapter=adapter,\n",
    "    inference_network=consistency_model,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9fbcca16",
   "metadata": {},
   "source": [
    "### Consistency Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "c3c1a812",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:bayesflow:Fitting on dataset instance of OfflineDataset.\n",
      "INFO:bayesflow:Building on a test batch.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 4ms/step - loss: 0.3549 - loss/inference_loss: 0.3549 - val_loss: 0.2793 - val_loss/inference_loss: 0.2793\n",
      "Epoch 2/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.3020 - loss/inference_loss: 0.3020 - val_loss: 0.3505 - val_loss/inference_loss: 0.3505\n",
      "Epoch 3/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2831 - loss/inference_loss: 0.2831 - val_loss: 0.2429 - val_loss/inference_loss: 0.2429\n",
      "Epoch 4/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2813 - loss/inference_loss: 0.2813 - val_loss: 0.3600 - val_loss/inference_loss: 0.3600\n",
      "Epoch 5/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2732 - loss/inference_loss: 0.2732 - val_loss: 0.2537 - val_loss/inference_loss: 0.2537\n",
      "Epoch 6/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2722 - loss/inference_loss: 0.2722 - val_loss: 0.2904 - val_loss/inference_loss: 0.2904\n",
      "Epoch 7/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2618 - loss/inference_loss: 0.2618 - val_loss: 0.1984 - val_loss/inference_loss: 0.1984\n",
      "Epoch 8/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2663 - loss/inference_loss: 0.2663 - val_loss: 0.1680 - val_loss/inference_loss: 0.1680\n",
      "Epoch 9/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2599 - loss/inference_loss: 0.2599 - val_loss: 0.2595 - val_loss/inference_loss: 0.2595\n",
      "Epoch 10/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2567 - loss/inference_loss: 0.2567 - val_loss: 0.2612 - val_loss/inference_loss: 0.2612\n",
      "Epoch 11/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2530 - loss/inference_loss: 0.2530 - val_loss: 0.2694 - val_loss/inference_loss: 0.2694\n",
      "Epoch 12/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2596 - loss/inference_loss: 0.2596 - val_loss: 0.3073 - val_loss/inference_loss: 0.3073\n",
      "Epoch 13/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2566 - loss/inference_loss: 0.2566 - val_loss: 0.1798 - val_loss/inference_loss: 0.1798\n",
      "Epoch 14/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2589 - loss/inference_loss: 0.2589 - val_loss: 0.2743 - val_loss/inference_loss: 0.2743\n",
      "Epoch 15/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2493 - loss/inference_loss: 0.2493 - val_loss: 0.2189 - val_loss/inference_loss: 0.2189\n",
      "Epoch 16/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2649 - loss/inference_loss: 0.2649 - val_loss: 0.2154 - val_loss/inference_loss: 0.2154\n",
      "Epoch 17/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2609 - loss/inference_loss: 0.2609 - val_loss: 0.2758 - val_loss/inference_loss: 0.2758\n",
      "Epoch 18/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2526 - loss/inference_loss: 0.2526 - val_loss: 0.1542 - val_loss/inference_loss: 0.1542\n",
      "Epoch 19/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2510 - loss/inference_loss: 0.2510 - val_loss: 0.1860 - val_loss/inference_loss: 0.1860\n",
      "Epoch 20/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2559 - loss/inference_loss: 0.2559 - val_loss: 0.2213 - val_loss/inference_loss: 0.2213\n",
      "Epoch 21/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2524 - loss/inference_loss: 0.2524 - val_loss: 0.2497 - val_loss/inference_loss: 0.2497\n",
      "Epoch 22/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2477 - loss/inference_loss: 0.2477 - val_loss: 0.2030 - val_loss/inference_loss: 0.2030\n",
      "Epoch 23/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2447 - loss/inference_loss: 0.2447 - val_loss: 0.2862 - val_loss/inference_loss: 0.2862\n",
      "Epoch 24/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2451 - loss/inference_loss: 0.2451 - val_loss: 0.3859 - val_loss/inference_loss: 0.3859\n",
      "Epoch 25/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2429 - loss/inference_loss: 0.2429 - val_loss: 0.2310 - val_loss/inference_loss: 0.2310\n",
      "Epoch 26/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2437 - loss/inference_loss: 0.2437 - val_loss: 0.2236 - val_loss/inference_loss: 0.2236\n",
      "Epoch 27/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2432 - loss/inference_loss: 0.2432 - val_loss: 0.3466 - val_loss/inference_loss: 0.3466\n",
      "Epoch 28/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2353 - loss/inference_loss: 0.2353 - val_loss: 0.2234 - val_loss/inference_loss: 0.2234\n",
      "Epoch 29/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2351 - loss/inference_loss: 0.2351 - val_loss: 0.1637 - val_loss/inference_loss: 0.1637\n",
      "Epoch 30/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2364 - loss/inference_loss: 0.2364 - val_loss: 0.2324 - val_loss/inference_loss: 0.2324\n",
      "Epoch 31/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2310 - loss/inference_loss: 0.2310 - val_loss: 0.1853 - val_loss/inference_loss: 0.1853\n",
      "Epoch 32/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2287 - loss/inference_loss: 0.2287 - val_loss: 0.1234 - val_loss/inference_loss: 0.1234\n",
      "Epoch 33/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2159 - loss/inference_loss: 0.2159 - val_loss: 0.1985 - val_loss/inference_loss: 0.1985\n",
      "Epoch 34/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2228 - loss/inference_loss: 0.2228 - val_loss: 0.4063 - val_loss/inference_loss: 0.4063\n",
      "Epoch 35/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2155 - loss/inference_loss: 0.2155 - val_loss: 0.2233 - val_loss/inference_loss: 0.2233\n",
      "Epoch 36/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2257 - loss/inference_loss: 0.2257 - val_loss: 0.1208 - val_loss/inference_loss: 0.1208\n",
      "Epoch 37/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2231 - loss/inference_loss: 0.2231 - val_loss: 0.0776 - val_loss/inference_loss: 0.0776\n",
      "Epoch 38/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2193 - loss/inference_loss: 0.2193 - val_loss: 0.2310 - val_loss/inference_loss: 0.2310\n",
      "Epoch 39/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2204 - loss/inference_loss: 0.2204 - val_loss: 0.1733 - val_loss/inference_loss: 0.1733\n",
      "Epoch 40/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2219 - loss/inference_loss: 0.2219 - val_loss: 0.1291 - val_loss/inference_loss: 0.1291\n",
      "Epoch 41/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2127 - loss/inference_loss: 0.2127 - val_loss: 0.1073 - val_loss/inference_loss: 0.1073\n",
      "Epoch 42/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2257 - loss/inference_loss: 0.2257 - val_loss: 0.2174 - val_loss/inference_loss: 0.2174\n",
      "Epoch 43/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2178 - loss/inference_loss: 0.2178 - val_loss: 0.2001 - val_loss/inference_loss: 0.2001\n",
      "Epoch 44/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2150 - loss/inference_loss: 0.2150 - val_loss: 0.2282 - val_loss/inference_loss: 0.2282\n",
      "Epoch 45/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2130 - loss/inference_loss: 0.2130 - val_loss: 0.1956 - val_loss/inference_loss: 0.1956\n",
      "Epoch 46/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2046 - loss/inference_loss: 0.2046 - val_loss: 0.1937 - val_loss/inference_loss: 0.1937\n",
      "Epoch 47/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2082 - loss/inference_loss: 0.2082 - val_loss: 0.2303 - val_loss/inference_loss: 0.2303\n",
      "Epoch 48/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2067 - loss/inference_loss: 0.2067 - val_loss: 0.1300 - val_loss/inference_loss: 0.1300\n",
      "Epoch 49/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 4ms/step - loss: 0.2201 - loss/inference_loss: 0.2201 - val_loss: 0.1432 - val_loss/inference_loss: 0.1432\n",
      "Epoch 50/50\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 3ms/step - loss: 0.2097 - loss/inference_loss: 0.2097 - val_loss: 0.1791 - val_loss/inference_loss: 0.1791\n"
     ]
    }
   ],
   "source": [
    "history = consistency_model_workflow.fit_offline(\n",
    "    training_data, \n",
    "    epochs=epochs, \n",
    "    batch_size=batch_size, \n",
    "    validation_data=validation_data\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a94a43f6",
   "metadata": {},
   "source": [
    "## Good 'ol Coupling Flows\n",
    "\n",
    "Of course, BayesFlow also supports established coupling flow models with a variety of parameters, including the timeless *affine* and *spline* flows."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "27b83a8f",
   "metadata": {},
   "outputs": [],
   "source": [
    "affine_flow = bf.networks.CouplingFlow(subnet=\"mlp\")\n",
    "\n",
    "spline_flow = bf.networks.CouplingFlow(subnet=\"mlp\", transform=\"spline\", depth=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "73b7a64d",
   "metadata": {},
   "outputs": [],
   "source": [
    "epochs = 30 # coupling flows need less epochs than free-form methods"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "e634dc50",
   "metadata": {},
   "outputs": [],
   "source": [
    "affine_flow_workflow = bf.BasicWorkflow(\n",
    "    simulator=simulator,\n",
    "    adapter=adapter,\n",
    "    inference_network=affine_flow,\n",
    ")\n",
    "\n",
    "\n",
    "spline_flow_workflow = bf.BasicWorkflow(\n",
    "    simulator=simulator,\n",
    "    adapter=adapter,\n",
    "    inference_network=spline_flow,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8aecf471",
   "metadata": {},
   "source": [
    "### Coupling Flow Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "f52e8e49",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:bayesflow:Fitting on dataset instance of OfflineDataset.\n",
      "INFO:bayesflow:Building on a test batch.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 6ms/step - loss: -1.2724 - loss/inference_loss: -1.2724 - val_loss: -2.2869 - val_loss/inference_loss: -2.2869\n",
      "Epoch 2/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -2.4998 - loss/inference_loss: -2.4998 - val_loss: -2.1505 - val_loss/inference_loss: -2.1505\n",
      "Epoch 3/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -2.6645 - loss/inference_loss: -2.6645 - val_loss: -2.6893 - val_loss/inference_loss: -2.6893\n",
      "Epoch 4/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 4ms/step - loss: -2.7494 - loss/inference_loss: -2.7494 - val_loss: -2.4783 - val_loss/inference_loss: -2.4783\n",
      "Epoch 5/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -2.7852 - loss/inference_loss: -2.7852 - val_loss: -2.7409 - val_loss/inference_loss: -2.7409\n",
      "Epoch 6/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 4ms/step - loss: -2.8575 - loss/inference_loss: -2.8575 - val_loss: -2.6789 - val_loss/inference_loss: -2.6789\n",
      "Epoch 7/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 4ms/step - loss: -2.9127 - loss/inference_loss: -2.9127 - val_loss: -3.0435 - val_loss/inference_loss: -3.0435\n",
      "Epoch 8/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -2.9346 - loss/inference_loss: -2.9346 - val_loss: -2.9628 - val_loss/inference_loss: -2.9628\n",
      "Epoch 9/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -2.9371 - loss/inference_loss: -2.9371 - val_loss: -2.3899 - val_loss/inference_loss: -2.3899\n",
      "Epoch 10/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 4ms/step - loss: -2.9999 - loss/inference_loss: -2.9999 - val_loss: -2.9758 - val_loss/inference_loss: -2.9758\n",
      "Epoch 11/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.0083 - loss/inference_loss: -3.0083 - val_loss: -2.5975 - val_loss/inference_loss: -2.5975\n",
      "Epoch 12/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.0434 - loss/inference_loss: -3.0434 - val_loss: -3.0387 - val_loss/inference_loss: -3.0387\n",
      "Epoch 13/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.0920 - loss/inference_loss: -3.0920 - val_loss: -2.3253 - val_loss/inference_loss: -2.3253\n",
      "Epoch 14/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.1255 - loss/inference_loss: -3.1255 - val_loss: -3.1556 - val_loss/inference_loss: -3.1556\n",
      "Epoch 15/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.1654 - loss/inference_loss: -3.1654 - val_loss: -2.8726 - val_loss/inference_loss: -2.8726\n",
      "Epoch 16/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.1779 - loss/inference_loss: -3.1779 - val_loss: -3.1917 - val_loss/inference_loss: -3.1917\n",
      "Epoch 17/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.2124 - loss/inference_loss: -3.2124 - val_loss: -2.1857 - val_loss/inference_loss: -2.1857\n",
      "Epoch 18/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - loss: -3.2409 - loss/inference_loss: -3.2409 - val_loss: -2.9640 - val_loss/inference_loss: -2.9640\n",
      "Epoch 19/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.2689 - loss/inference_loss: -3.2689 - val_loss: -2.7462 - val_loss/inference_loss: -2.7462\n",
      "Epoch 20/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.2759 - loss/inference_loss: -3.2759 - val_loss: -2.9262 - val_loss/inference_loss: -2.9262\n",
      "Epoch 21/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - loss: -3.3185 - loss/inference_loss: -3.3185 - val_loss: -2.1308 - val_loss/inference_loss: -2.1308\n",
      "Epoch 22/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.3447 - loss/inference_loss: -3.3447 - val_loss: -2.6838 - val_loss/inference_loss: -2.6838\n",
      "Epoch 23/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.3607 - loss/inference_loss: -3.3607 - val_loss: -3.4517 - val_loss/inference_loss: -3.4517\n",
      "Epoch 24/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - loss: -3.3540 - loss/inference_loss: -3.3540 - val_loss: -3.0357 - val_loss/inference_loss: -3.0357\n",
      "Epoch 25/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.3831 - loss/inference_loss: -3.3831 - val_loss: -3.3768 - val_loss/inference_loss: -3.3768\n",
      "Epoch 26/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - loss: -3.4183 - loss/inference_loss: -3.4183 - val_loss: -3.1671 - val_loss/inference_loss: -3.1671\n",
      "Epoch 27/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - loss: -3.4194 - loss/inference_loss: -3.4194 - val_loss: 9.2951 - val_loss/inference_loss: 9.2951\n",
      "Epoch 28/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.3851 - loss/inference_loss: -3.3851 - val_loss: -2.9896 - val_loss/inference_loss: -2.9896\n",
      "Epoch 29/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 5ms/step - loss: -3.4251 - loss/inference_loss: -3.4251 - val_loss: -3.3175 - val_loss/inference_loss: -3.3175\n",
      "Epoch 30/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - loss: -3.4083 - loss/inference_loss: -3.4083 - val_loss: -3.2599 - val_loss/inference_loss: -3.2599\n"
     ]
    }
   ],
   "source": [
    "history = affine_flow_workflow.fit_offline(\n",
    "    training_data, \n",
    "    epochs=epochs,\n",
    "    batch_size=batch_size,\n",
    "    validation_data=validation_data\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "afa9839f",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:bayesflow:Fitting on dataset instance of OfflineDataset.\n",
      "INFO:bayesflow:Building on a test batch.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m21s\u001b[0m 11ms/step - loss: -1.0211 - loss/inference_loss: -1.0211 - val_loss: -1.5653 - val_loss/inference_loss: -1.5653\n",
      "Epoch 2/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -2.2318 - loss/inference_loss: -2.2318 - val_loss: -2.4503 - val_loss/inference_loss: -2.4503\n",
      "Epoch 3/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -2.5659 - loss/inference_loss: -2.5659 - val_loss: -2.6406 - val_loss/inference_loss: -2.6406\n",
      "Epoch 4/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -2.7533 - loss/inference_loss: -2.7533 - val_loss: -2.6080 - val_loss/inference_loss: -2.6080\n",
      "Epoch 5/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -2.8607 - loss/inference_loss: -2.8607 - val_loss: -2.8657 - val_loss/inference_loss: -2.8657\n",
      "Epoch 6/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -2.9422 - loss/inference_loss: -2.9422 - val_loss: -2.3686 - val_loss/inference_loss: -2.3686\n",
      "Epoch 7/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -2.9989 - loss/inference_loss: -2.9989 - val_loss: -2.9271 - val_loss/inference_loss: -2.9271\n",
      "Epoch 8/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.0638 - loss/inference_loss: -3.0638 - val_loss: -3.0360 - val_loss/inference_loss: -3.0360\n",
      "Epoch 9/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.0553 - loss/inference_loss: -3.0553 - val_loss: -3.2254 - val_loss/inference_loss: -3.2254\n",
      "Epoch 10/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.1095 - loss/inference_loss: -3.1095 - val_loss: -3.0538 - val_loss/inference_loss: -3.0538\n",
      "Epoch 11/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.2038 - loss/inference_loss: -3.2038 - val_loss: -3.1451 - val_loss/inference_loss: -3.1451\n",
      "Epoch 12/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.2396 - loss/inference_loss: -3.2396 - val_loss: -3.2923 - val_loss/inference_loss: -3.2923\n",
      "Epoch 13/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.2695 - loss/inference_loss: -3.2695 - val_loss: -2.7734 - val_loss/inference_loss: -2.7734\n",
      "Epoch 14/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.2778 - loss/inference_loss: -3.2778 - val_loss: -3.3034 - val_loss/inference_loss: -3.3034\n",
      "Epoch 15/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.3281 - loss/inference_loss: -3.3281 - val_loss: -2.5565 - val_loss/inference_loss: -2.5565\n",
      "Epoch 16/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.3416 - loss/inference_loss: -3.3416 - val_loss: -3.1074 - val_loss/inference_loss: -3.1074\n",
      "Epoch 17/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.3612 - loss/inference_loss: -3.3612 - val_loss: -3.4145 - val_loss/inference_loss: -3.4145\n",
      "Epoch 18/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.3806 - loss/inference_loss: -3.3806 - val_loss: -3.3677 - val_loss/inference_loss: -3.3677\n",
      "Epoch 19/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.4344 - loss/inference_loss: -3.4344 - val_loss: -2.7881 - val_loss/inference_loss: -2.7881\n",
      "Epoch 20/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.4364 - loss/inference_loss: -3.4364 - val_loss: -3.2963 - val_loss/inference_loss: -3.2963\n",
      "Epoch 21/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.4700 - loss/inference_loss: -3.4700 - val_loss: -2.8206 - val_loss/inference_loss: -2.8206\n",
      "Epoch 22/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.4920 - loss/inference_loss: -3.4920 - val_loss: -3.0479 - val_loss/inference_loss: -3.0479\n",
      "Epoch 23/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.5153 - loss/inference_loss: -3.5153 - val_loss: -3.0690 - val_loss/inference_loss: -3.0690\n",
      "Epoch 24/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.5309 - loss/inference_loss: -3.5309 - val_loss: -2.9115 - val_loss/inference_loss: -2.9115\n",
      "Epoch 25/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.5516 - loss/inference_loss: -3.5516 - val_loss: -3.0176 - val_loss/inference_loss: -3.0176\n",
      "Epoch 26/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.5659 - loss/inference_loss: -3.5659 - val_loss: -3.0676 - val_loss/inference_loss: -3.0676\n",
      "Epoch 27/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.5792 - loss/inference_loss: -3.5792 - val_loss: -3.3448 - val_loss/inference_loss: -3.3448\n",
      "Epoch 28/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.5791 - loss/inference_loss: -3.5791 - val_loss: -3.1989 - val_loss/inference_loss: -3.1989\n",
      "Epoch 29/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.5712 - loss/inference_loss: -3.5712 - val_loss: -3.0113 - val_loss/inference_loss: -3.0113\n",
      "Epoch 30/30\n",
      "\u001b[1m512/512\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m4s\u001b[0m 8ms/step - loss: -3.4733 - loss/inference_loss: -3.4733 - val_loss: -3.4334 - val_loss/inference_loss: -3.4334\n"
     ]
    }
   ],
   "source": [
    "history = spline_flow_workflow.fit_offline(\n",
    "    training_data,\n",
    "    epochs=epochs,\n",
    "    batch_size=batch_size,\n",
    "    validation_data=validation_data\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f6ffbb96",
   "metadata": {},
   "source": [
    "## Validation"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c0ca73c8",
   "metadata": {},
   "source": [
    "### Two Moons Posterior\n",
    "\n",
    "The two moons posterior at point $x = (0, 0)$ should resemble two crescent shapes. Below, we plot the corresponding posterior samples and posterior density. \n",
    "\n",
    "These results suggest that these generative networks can approximate the true posterior well. You can achieve an even better fit if you use online training, more epochs, or better optimizer hyperparameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "073bcd0b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1500x600 with 4 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Set the number of posterior draws you want to get\n",
    "num_samples = 3000\n",
    "\n",
    "# Obtain samples from amortized posterior\n",
    "conditions = {\"x\": np.array([[0.0, 0.0]]).astype(\"float32\")}\n",
    "\n",
    "# Prepare figure\n",
    "f, axes = plt.subplots(1, 4, figsize=(15, 6))\n",
    "\n",
    "# Obtain samples from the approximators (can also use the workflows' methods)\n",
    "nets = [\n",
    "    flow_matching_workflow.approximator, \n",
    "    consistency_model_workflow.approximator,\n",
    "    affine_flow_workflow.approximator,\n",
    "    spline_flow_workflow.approximator\n",
    "]\n",
    "names = [\"Flow Matching\", \"Consistency Model\", \"Affine Coupling Flow\", \"Spline Coupling Flow\"]\n",
    "colors = [\"#153c7a\", \"#7a1515\", \"#157a2d\", \"#7a6f15\"]\n",
    "\n",
    "for ax, net, name, color in zip(axes, nets, names, colors):\n",
    "\n",
    "    # Obtain samples\n",
    "    samples = net.sample(conditions=conditions, num_datasets=1, num_samples=num_samples)[\"theta\"]\n",
    "\n",
    "    # Plot samples\n",
    "    ax.scatter(samples[0, :, 0], samples[0, :, 1], color=color, alpha=0.75, s=0.5)\n",
    "    sns.despine(ax=ax)\n",
    "    ax.set_title(f\"{name}\", fontsize=16)\n",
    "    ax.grid(alpha=0.3)\n",
    "    ax.set_aspect(\"equal\", adjustable=\"box\")\n",
    "    ax.set_xlim([-0.5, 0.5])\n",
    "    ax.set_ylim([-0.5, 0.5])\n",
    "    ax.set_xlabel(r\"$\\theta_1$\", fontsize=15)\n",
    "    ax.set_ylabel(r\"$\\theta_2$\", fontsize=15)\n",
    "\n",
    "f.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "895b03ae",
   "metadata": {},
   "source": [
    "The posterior looks as we have expected in this case. However, in general, we do not know how the posterior is supposed to look like for any specific dataset. As such, we need diagnostics that validate the correctness of the inferred posterior. One such diagnostic is simulation-based calibration(SBC), which we can apply for free due to amortization. For more details on SBC and diagnostic plots, see:\n",
    "\n",
    "1. Talts, S., Betancourt, M., Simpson, D., Vehtari, A., & Gelman, A. (2018). Validating Bayesian inference algorithms with simulation-based calibration. *arXiv preprint*.\n",
    "2. Säilynoja, T., Bürkner, P. C., & Vehtari, A. (2022). Graphical test for discrete uniformity and its applications in goodness-of-fit evaluation and multiple sample comparison. *Statistics and Computing*.\n",
    "3. The practical SBC interpretation guide by Martin Modrák: https://hyunjimoon.github.io/SBC/articles/rank_visualizations.html\n",
    "\n",
    "Check out the next tutorial for a detailed walkthrough of the workflow's functionality."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "886bce5c",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.11"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": true,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": true,
   "toc_position": {
    "height": "calc(100% - 180px)",
    "left": "10px",
    "top": "150px",
    "width": "165px"
   },
   "toc_section_display": true,
   "toc_window_display": true
  },
  "widgets": {
   "application/vnd.jupyter.widget-state+json": {
    "state": {},
    "version_major": 2,
    "version_minor": 0
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}