{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# How to use CISS-VAE\n",
    "\n",
    "## Overview\n",
    "\n",
    "The **Clustering-Informed Shared-Structure Variational Autoencoder (CISS-VAE)** is a flexible deep learning model for missing data imputation that accommodates all three types of missing data mechanisms: Missing Completely At Random (MCAR), Missing At Random (MAR), and Missing Not At Random (MNAR). While it is particularly well-suited to MNAR, scenarios where missingness patterns carry informative signals, CISS-VAE also functions effectively under MAR assumptions.\n",
    "\n",
    "![**Example CISS-VAE for Imputation Workflow**](CISSVAEModelDiagram.png)\n",
    "\n",
    "<details style=\"background: #deeff7ff; border: #2A373D; color: black; border-radius: 10px\"><summary><span style=\"color: #2A373D\"><b>Click Here for More Information</b></span></summary>\n",
    " <div style=\"padding: 10px 10px 10px 10px\">\n",
    "A key feature of CISS-VAE is the use of unsupervised clustering to capture distinct patterns of missingness. Alongside cluster-specific representations, the method leverages shared encoder and decoder layers. This allows for knowledge transfer across clusters and enhances parameter stability, which is especially important when some clusters have small sample sizes. In situations where the data do not naturally partition into meaningful clusters, the model defaults to a pooled representation, preventing unnecessary complications from cluster-specific components. <br>  <br>\n",
    " \n",
    "Additionally, CISS-VAE incorporates an **iterative learning procedure**, with a validation-based convergence criterion recommended to avoid overfitting. This procedure significantly improves imputation accuracy compared to traditional Variational Autoencoder training approaches in the presence of missing values. Overall, CISS-VAE adapts across a range of missing data mechanisms, leveraging clustering only when it offers clear benefits, and delivering robust, accurate imputations under varying conditions of missingness.\n",
    "    </div>\n",
    "</details>  \n",
    "<br>\n",
    "There are two ways to run the CISS-VAE process. If you know what model\n",
    "parameters you want to use, you can use the {py:func}`ciss_vae.training.run_cissvae.run_cissvae` function to\n",
    "run the model once for the given set of parameters. If you want to tune\n",
    "the model instead, you can use {py:func}`ciss_vae.training.autotune.autotune`.\n",
    "\n",
    "The R package associated with this model can be found at <a href=\"https://ciss-vae.github.io/rCISS-VAE/\">rCISS-VAE</a>.\n",
    "\n",
    "## Installation\n",
    "\n",
    "The CISS-VAE package is currently available for python, with an R\n",
    "package to be released soon. It can be installed from either\n",
    "[github](https://github.com/CISS-VAE/CISS-VAE-python) or PyPI.\n",
    "\n",
    "``` bash\n",
    "# From PyPI \n",
    "pip install ciss-vae\n",
    "\n",
    "```\n",
    "\n",
    "``` bash\n",
    "# From GitHub (latest development version)\n",
    "pip install git+https://github.com/CISS-VAE/CISS-VAE-python.git\n",
    "```\n",
    "\n",
    "<div>\n",
    "\n",
    "> **Note**\n",
    ">\n",
    "> If you want run_cissvae to handle clustering, please install the\n",
    "> clustering dependencies scikit-learn, leidenalg, python-igraph with pip.\n",
    ">\n",
    "> ``` bash\n",
    "> pip install scikit-learn leidenalg python-igraph\n",
    ">\n",
    "> OR\n",
    ">\n",
    "> pip install ciss-vae[clustering]\n",
    "> ```\n",
    "\n",
    "</div>\n",
    "\n",
    "# Running the CISS-VAE Model\n",
    "\n",
    "You can use your own dataset or load the example dataset included with this package. If your data has binary or categorical variables, please see the [binary data vignette](https://ciss-vae.readthedocs.io/en/latest/binary_vignette.html).\n",
    "\n",
    "\n",
    "To load the sample dataset:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ciss_vae.data import load_example_dataset\n",
    "df_missing, df_complete, clusters = load_example_dataset()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "If you already know what parameters you want for your model (or do not want to use the {py:func}`ciss_vae.training.autotune.autotune` function), you can use the {py:func}`ciss_vae.training.run_cissvae.run_cissvae` function to perform the imputation.\n",
    "\n",
    "The input dataset should be one of the following:\n",
    "\n",
    "    - A Pandas DataFrame  \n",
    "\n",
    "    - A NumPy array  \n",
    "\n",
    "    - A PyTorch tensor  \n",
    "\n",
    "Missing values should be represented using np.nan or None.\n",
    "\n",
    "### Assigning cluster labels\n",
    "There are three options for assigning cluster labels to the data:      \n",
    "\n",
    "    1. Manually assign cluster labels and provide them to the function via the `clusters` argument. \n",
    "\n",
    "          \n",
    "    2. Let `run_cissvae()` determine clusters based on patterns of missingness in the data by setting `clusters = None`.  \n",
    "        - To use Kmeans clustering, set n_clusters to the desired number of clusters.    \n",
    "        - To use Leiden Clustering clustering, leave `n_clusters = None`.     \n",
    "  \n",
    "\n",
    "    3. To cluster on proportion of missingness, leave `clusters = None` and provide a missingness proportion matrix with the `missingness_proportion_matrix` argument.  \n",
    "        - To use Kmeans clustering, set n_clusters to the desired number of clusters.     \n",
    "        - To use Leiden Clustering clustering, leave `n_clusters = None`.   \n",
    "\n",
    "<details style=\"background: #deeff7ff; border: #2A373D; color: black; border-radius: 10px\"><summary><span style=\"color: #2A373D\"><b>Click Here for More Information on the Missingness Proportion Matrix</b></span></summary>\n",
    " <div style=\"padding: 10px 10px 10px 10px\"><p>For the missingness proportion matrix, (either a `pandas.DataFrame` or `numpy.ndarray`) the rows should correspond to samples, the columns correspond to features, and the values are proportion of missingness of each feature for each sample. For features with multiple timepoints (like biomarker data collected at multiple visits), you may choose to have one column per feature and let the value be the overall proportion of missingness for that feature across all timepoints. See the <a href=\"https://ciss-vae.readthedocs.io/en/latest/missingness_prop_vignette.html\">clustering on missingness proportion tutorial</a> for more details.</p></div>\n",
    "    </details>    \n",
    "      \n",
    "\n",
    "To run the CISSVAE model with default parameters:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cluster dataset:\n",
      " ClusterDataset(n_samples=8000, n_features=30, n_clusters=4)\n",
      "  • Original missing: 61800 / 200000 (30.90%)\n",
      "  • Validation held-out: 13783 (9.97% of non-missing)\n",
      "  • .data shape:     (8000, 30)\n",
      "  • .masks shape:    (8000, 30)\n",
      "  • .val_data shape: (8000, 30)\n",
      "  • Validation units: {'Y11': {'kind': 'continuous', 'cols': ['Y11']}, 'Y12': {'kind': 'continuous', 'cols': ['Y12']}, 'Y13': {'kind': 'continuous', 'cols': ['Y13']}, 'Y14': {'kind': 'continuous', 'cols': ['Y14']}, 'Y15': {'kind': 'continuous', 'cols': ['Y15']}, 'Y21': {'kind': 'continuous', 'cols': ['Y21']}, 'Y22': {'kind': 'continuous', 'cols': ['Y22']}, 'Y23': {'kind': 'continuous', 'cols': ['Y23']}, 'Y24': {'kind': 'continuous', 'cols': ['Y24']}, 'Y25': {'kind': 'continuous', 'cols': ['Y25']}, 'Y31': {'kind': 'continuous', 'cols': ['Y31']}, 'Y32': {'kind': 'continuous', 'cols': ['Y32']}, 'Y33': {'kind': 'continuous', 'cols': ['Y33']}, 'Y34': {'kind': 'continuous', 'cols': ['Y34']}, 'Y35': {'kind': 'continuous', 'cols': ['Y35']}, 'Y41': {'kind': 'continuous', 'cols': ['Y41']}, 'Y42': {'kind': 'continuous', 'cols': ['Y42']}, 'Y43': {'kind': 'continuous', 'cols': ['Y43']}, 'Y44': {'kind': 'continuous', 'cols': ['Y44']}, 'Y45': {'kind': 'continuous', 'cols': ['Y45']}, 'Y51': {'kind': 'continuous', 'cols': ['Y51']}, 'Y52': {'kind': 'continuous', 'cols': ['Y52']}, 'Y53': {'kind': 'continuous', 'cols': ['Y53']}, 'Y54': {'kind': 'continuous', 'cols': ['Y54']}, 'Y55': {'kind': 'continuous', 'cols': ['Y55']}}\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1600x800 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "from ciss_vae.training.run_cissvae import run_cissvae\n",
    "from ciss_vae.data import load_example_dataset\n",
    "\n",
    "# optional, display vae architecture\n",
    "from ciss_vae.utils.helpers import plot_vae_architecture\n",
    "\n",
    "data, _, clusters = load_example_dataset()\n",
    "\n",
    "imputed_data, vae = run_cissvae(data = data,\n",
    "## Dataset params\n",
    "    val_proportion = 0.1, ## Fraction of non-missing data held out for validation. Can input a list here if you want different proportions for different clusters. \n",
    "    replacement_value = 0.0, \n",
    "    columns_ignore = data.columns[:5], ## columns to ignore when selecting validation dataset (and clustering if you do not provide clusters). For example, demographic columns with no missingness.\n",
    "    print_dataset = True, \n",
    "\n",
    "## Cluster params\n",
    "    clusters = clusters, ## Where your cluster list goes. If none, will do clustering for you  \n",
    "    n_clusters = None, ## If you want run_cissvae to do clustering and you know how many clusters your data should have, enter that number here\n",
    "    # -- Params for Leiden Clustering --\n",
    "    k_neighbors = 15,\n",
    "    leiden_resolution = 0.5, ## Lower resolution = fewer clusters and bigger clusters, higher resolution = more, smaller clusters\n",
    "    leiden_objective = \"CPM\", \n",
    "    # -- End Params for Leiden Clustering --\n",
    "    seed = 42,\n",
    "    missingness_proportion_matrix = None,\n",
    "    \n",
    "## VAE model params\n",
    "    hidden_dims = [150, 120, 60], ## Dimensions of hidden layers, in order. One number per layer. \n",
    "    latent_dim = 15, ## Dimensions of latent embedding\n",
    "    layer_order_enc = [\"unshared\", \"unshared\", \"unshared\"], ## order of shared vs unshared layers for encode (can use u or s instead of unshared, shared)\n",
    "    layer_order_dec=[\"shared\", \"shared\",  \"shared\"],  ## order of shared vs unshared layers for decode\n",
    "    latent_shared=False, \n",
    "    output_shared=False, \n",
    "    batch_size = 4000, ## batch size for data loader\n",
    "    return_model = True, ## if true, outputs imputed dataset and model, otherwise just outputs imputed dataset. Set to true to return model for `plot_vae_architecture`\n",
    "\n",
    "## Initial Training params\n",
    "    epochs = 1000, ## default \n",
    "    initial_lr = 0.01, ## default\n",
    "    decay_factor = 0.999, ## default, factor learning rate is multiplied by after each epoch, prevents overfitting\n",
    "    beta= 0.001, ## default\n",
    "    device = None, ## If none, will use gpu if available, cpu if not. See torch.devices for info \n",
    "\n",
    "## Impute-refit loop params\n",
    "    max_loops = 100, ## max number of refit loops\n",
    "    patience = 2, ## number of loops to check after best_dataset updated. Can increase to avoid local extrema\n",
    "    epochs_per_loop = None, ## If none, same as epochs\n",
    "    initial_lr_refit = None, ## If none, picks up from end of initial training\n",
    "    decay_factor_refit = None, ## If none, same as decay_factor\n",
    "    beta_refit = None, ## if none, same as beta\n",
    "\n",
    "## Other params\n",
    "    verbose = False, \n",
    "    return_silhouettes = False, ## if true, will return silhouettes from clustering. If run_cissvae did not perform clustering, will return \"None\"\n",
    "    return_history = False, ## if true, will return training MSE history as pandas dataframe\n",
    "    return_clusters = False, ## if true, will return the cluster labels. Helpful if run_cissvae performs clustering\n",
    ")\n",
    "\n",
    "## OPTIONAL - PLOT VAE ARCHITECTURE\n",
    "plot_vae_architecture(model = vae,\n",
    "                        title = None, ## Set title of plot\n",
    "                        ## Colors below are default\n",
    "                        color_shared = \"skyblue\", \n",
    "                        color_unshared =\"lightcoral\",\n",
    "                        color_latent = \"gold\", # xx fix\n",
    "                        color_input = \"lightgreen\",\n",
    "                        color_output = \"lightgreen\",\n",
    "                        figsize=(16, 8),\n",
    "                        return_fig = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Use the `return_clusters` parameter to get the cluster labels.\n",
    "Use the `return_history` parameter to get a dataframe with the validation imputation loss history.\n",
    "\n",
    "### Avoiding imputation of certain entries\n",
    "In some cases, not all missing entries in a dataset are viable for imputation. For example, biomarker values after time of death would not necessarily be reasonable to impute and therefore should be ignored during the impute-refit training loop. To set certain data entries as un-imputable, create an 'imputable' matrix of the same size as the dataset, with 1 for entries that are non-missing or viable for imputation and 0 for entries that are missing and non-viable for imputation. \n",
    "\n",
    "This matrix (or pandas.Dataframe) can then be passed to the `run_cissvae()` function. Make sure that the 'imputable' matrix has the same column labels as the original data set to use the `cols_ignore` option to ignore certain columns during imputation. \n",
    "\n",
    "For more information, see the [imputable matrix vignette](https://ciss-vae.readthedocs.io/en/latest/dni_vignette.html).\n",
    "\n",
    "# Hyperparameter Tuning with Optuna\n",
    "\n",
    "The {py:func}`ciss_vae.training.autotune.autotune` function lets you tune the model's hyperparameters with\n",
    "optuna to get the best possible model.\n",
    "\n",
    "## Dataset Preparation\n",
    "\n",
    "Your dataset should be one of the following:\n",
    "\n",
    "    - A Pandas DataFrame  \n",
    "\n",
    "    - A NumPy array  \n",
    "\n",
    "    - A PyTorch tensor  \n",
    "\n",
    "Missing values should be represented using np.nan or None.\n",
    "\n",
    "Once the dataset is loaded, the first step is to identify patterns of\n",
    "missingness using clustering.\n",
    "\n",
    "## Clustering on missingness pattern\n",
    "\n",
    "Before fitting the model, the dataset should clustered based on its\n",
    "missingness pattern (i.e., which variables are missing in each\n",
    "observation).\n",
    "\n",
    "The built-in function can perfrom either leiden clustering or Kmeans clustering:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ciss_vae.utils.clustering import cluster_on_missing\n",
    "data, _, clusters = load_example_dataset()\n",
    "clusters, _ = cluster_on_missing(\n",
    "    data, \n",
    "    cols_ignore=data.columns[:5], \n",
    "    n_clusters=None, \n",
    "    k_neighbors=500, ## use higher k for fewer clusters generally\n",
    "    use_snn=True,\n",
    "    leiden_resolution=0.005, ## higher resolution -> more clusters\n",
    "    leiden_objective=\"CPM\",\n",
    "    seed=42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "This function uses Leiden Clustering clustering to detect structure in binary\n",
    "missingness masks, and will automatically determine the number of\n",
    "clusters if not specified. If n_clusters is specified, uses KMeans.\n",
    "\n",
    "**Options:**  \n",
    "- cols_ignore: list of columns to exclude when computing the missingness\n",
    "pattern. Ex: identifiers\n",
    "- n_clusters: set this to use K-Means instead of nonparametric\n",
    "    clustering.  \n",
    "\n",
    "**To cluster on proportion of missingness, see <a href=\"https://ciss-vae.readthedocs.io/en/latest/missingness_prop_vignette.html\">tutorial</a> for more details.**\n",
    "\n",
    "## Creating a `ClusterDataset` object\n",
    "\n",
    "After obtaining cluster labels, construct a {py:class}`ciss_vae.classes.cluster_dataset.ClusterDataset`. This is the object that is fed into the autotune function."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ciss_vae.classes.cluster_dataset import ClusterDataset\n",
    "\n",
    "dataset = ClusterDataset(data = data,\n",
    "cluster_labels = clusters,\n",
    "val_proportion = 0.1, ## 10% non-missing data is default. \n",
    "replacement_value = 0, ## value to replace all missing data with before running model. Could be set to 0 or random\n",
    "columns_ignore = data.columns[:5] ## Tells ClusterDataset not to hold out entries of these columns for validation\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "## Create a SearchSpace object:\n",
    "\n",
    "In the SearchSpace object, define the search space for each\n",
    "hyperparameter. Each of the parameters in {py:class}`ciss_vae.training.autotune.SearchSpace` can be set as\n",
    "either tunable or non-tunable.\n",
    "\n",
    "Types of parameters:  \n",
    "- (min, max, step) -\\> creates a range   \n",
    "- \\[a, b, c\\] -\\> select value from list\n",
    "- x -\\> set param as non-tunable    \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ciss_vae.training.autotune import SearchSpace, autotune\n",
    "## These are the default parameters. Please note these parameters may not be best for all datasets depending on size and complexity.\n",
    "\n",
    "searchspace = SearchSpace(\n",
    "                 num_hidden_layers=(1, 4), ## Set number of hidden layers\n",
    "                 hidden_dims=[64, 512], ## Allowable dimensions of hidden layers\n",
    "                 latent_dim=[10, 100],\n",
    "                 latent_shared=[True, False],\n",
    "                 output_shared=[True,False],\n",
    "                 lr=(1e-4, 1e-3),\n",
    "                 decay_factor=(0.9, 0.999),\n",
    "                 beta=0.01,\n",
    "                 num_epochs=1000,\n",
    "                 batch_size=64,\n",
    "                 num_shared_encode=[0, 1, 3],\n",
    "                 num_shared_decode=[0, 1, 3],\n",
    "                 encoder_shared_placement = [\"at_end\", \"at_start\", \"alternating\", \"random\"], ## where should the shared layers be placed in the encoder\n",
    "                 decoder_shared_placement = [\"at_end\", \"at_start\", \"alternating\", \"random\"], ## where should the shared layers be placed in the decoder\n",
    "                 refit_patience=2,\n",
    "                 refit_loops=100,\n",
    "                 epochs_per_loop = 1000,\n",
    "                 reset_lr_refit = [True, False])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving and Loading SearchSpace Objects\n",
    "\n",
    "SearchSpace objects can be saved as json files for future use. To save a SearchSpace object:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "searchspace.save(\"saved_ss.json\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To load a SearchSpace object:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SearchSpace(\n",
      "  num_hidden_layers: [1, 4]  [TUNABLE]\n",
      "  hidden_dims: [64, 512]  [TUNABLE]\n",
      "  latent_dim: [10, 100]  [TUNABLE]\n",
      "  latent_shared: [True, False]  [TUNABLE]\n",
      "  output_shared: [True, False]  [TUNABLE]\n",
      "  lr: [0.0001, 0.001]  [TUNABLE]\n",
      "  decay_factor: [0.9, 0.999]  [TUNABLE]\n",
      "  weight_decay: 0.001  [FIXED]\n",
      "  beta: 0.01  [FIXED]\n",
      "  num_epochs: 1000  [FIXED]\n",
      "  batch_size: 64  [FIXED]\n",
      "  num_shared_encode: [0, 1, 3]  [TUNABLE]\n",
      "  num_shared_decode: [0, 1, 3]  [TUNABLE]\n",
      "  encoder_shared_placement: ['at_end', 'at_start', 'alternating', 'random']  [TUNABLE]\n",
      "  decoder_shared_placement: ['at_end', 'at_start', 'alternating', 'random']  [TUNABLE]\n",
      "  refit_patience: 2  [FIXED]\n",
      "  refit_loops: 100  [FIXED]\n",
      "  epochs_per_loop: 1000  [FIXED]\n",
      "  reset_lr_refit: [True, False]  [TUNABLE]\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "loaded_ss = SearchSpace().load(\"saved_ss.json\")\n",
    "\n",
    "print(loaded_ss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Run the `autotune` function:\n",
    "\n",
    "Once the search space is set, the autotune function can be run.\n",
    "\n",
    "There are a few options for running the autotune function, depending on your goals.   \n",
    "    1. Default:      \n",
    "        - Tune on a random sample of parameters from the SearchSpace obbject. This is the traditional autotune behavior.   \n",
    "        - For this behavior, set `constant_layer_size=False` and `evaluate_all_orders=False`.  \n",
    "    2. Tune with constant layer size:  \n",
    "        - All layers will be the same size. The size will be one selected from searchspace. \n",
    "        (ex: if `searchspace.num_hidden_layers` = [64, 512], all layers will be 64 or all layers will be 512),  \n",
    "        - For this behavior, set `constant_layer_size=False`  \n",
    "    3. Tune for all permutations of shared layer placement:  \n",
    "        - Will tune all possible layer orders/placements. Use `max_exhaustive_orders` to set a cap on the number of permutations to try.  \n",
    "        - For this behavior, set `evaluate_all_orders=True`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2026-04-14 14:26:10,482] A new study created in memory with name: vae_autotune\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[Warning] CUDA requested but not available. Falling back to CPU.\n",
      "Starting Optuna optimization with 20 trials...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2026-04-14 20:26:06,038] Trial 0 finished with value: {'Total Imputation Error': 2.6448564529418945} and parameters: {'num_hidden_layers': 2, 'hidden_dim_0': 64, 'hidden_dim_1': 64, 'latent_dim': 10, 'latent_shared': True, 'output_shared': True, 'lr': 0.0009330606024425672, 'decay_factor': 0.9816830436117746, 'num_shared_encode': 0, 'num_shared_decode': 1, 'encoder_shared_placement': 'at_start', 'decoder_shared_placement': 'alternating', 'reset_lr_refit': False}. Best is trial 0 with value: 2.6448564529418945.\n",
      "[I 2026-04-15 01:47:11,118] Trial 1 finished with value: {'Total Imputation Error': 8.764293670654297} and parameters: {'num_hidden_layers': 1, 'hidden_dim_0': 64, 'latent_dim': 100, 'latent_shared': True, 'output_shared': True, 'lr': 0.0004833180632488466, 'decay_factor': 0.9423050781656934, 'num_shared_encode': 1, 'num_shared_decode': 0, 'encoder_shared_placement': 'alternating', 'decoder_shared_placement': 'at_end', 'reset_lr_refit': False}. Best is trial 0 with value: 2.6448564529418945.\n"
     ]
    }
   ],
   "source": [
    "best_imputed_df,  best_model, study, results_df = autotune(\n",
    "    search_space = searchspace,\n",
    "    train_dataset = dataset,                   # ClusterDataset object\n",
    "    save_model_path=None,\n",
    "    save_search_space_path=None,\n",
    "    n_trials=20,\n",
    "    study_name=\"vae_autotune\",                 # Default study name\n",
    "    device_preference=\"cuda\",\n",
    "    show_progress=False,                       # Show progress bar for training\n",
    "    optuna_dashboard_db=None,                  # If using optuna dashboard set db location here\n",
    "    load_if_exists=True,                       # If using optuna dashboard, if study by 'study_name' already exists, will load that study\n",
    "    seed = 42,                                 # Sets seed for random order of shared/unshared layers\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "## (optional) Using Optuna Dashboard\n",
    "\n",
    "You can use [optuna\n",
    "dashboard](https://optuna-dashboard.readthedocs.io/en/stable/getting-started.html)\n",
    "to visualize the importance of your tuning parameters. If you use VSCode\n",
    "or [Positron](https://positron.posit.co/download.html), there is an\n",
    "extension for viewing optuna dashboards in your development environment.\n",
    "\n",
    "![Screenshot of Optuna Dashboard](optuna_dash_v1.png)\n",
    "\n",
    "The optuna database file can be opened via commandline as well. [tutorial](https://optuna-dashboard.readthedocs.io/en/stable/getting-started.html#command-line-interface)\n",
    "\n",
    "\n",
    "To use optuna dashboard, set your database url in the autotune function. You\n",
    "can have multiple autotune 'studies' in the same database."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "best_imputed_df,  best_model, study, results_df = autotune(\n",
    "    search_space = searchspace,\n",
    "    train_dataset = dataset,                   # 'ClusterDataset' object\n",
    "    save_model_path=None,\n",
    "    save_search_space_path=None,\n",
    "    n_trials=20,\n",
    "    study_name=\"vae_autotune\",                 # Default study name\n",
    "    device_preference=\"cuda\",\n",
    "    show_progress=False,                       # Show progress bar for training\n",
    "    optuna_dashboard_db=\"sqlite:///db.sqlite3\",   # If using optuna dashboard set db location here, otherwise set to None\n",
    "    load_if_exists=True,                       # Continues previous study by study_name if one exists. If false, will give error if study_name already exists in the set dashboard\n",
    "    seed = 42,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "# Saving and loading models\n",
    "\n",
    "## Saving\n",
    "\n",
    "If you want to save your model and load it later, there are two options.\n",
    "\n",
    "To save the model weights after training:\n",
    "\n",
    "``` python\n",
    "## assuming your trained model is called 'model'\n",
    "import torch\n",
    "\n",
    "torch.save(model.state_dict(), \"trained_vae.pt\")\n",
    "```\n",
    "\n",
    "If you want to save the entire model:\n",
    "\n",
    "``` python\n",
    "torch.save(model, \"trained_vae_full.pt\")\n",
    "```\n",
    "\n",
    "## Loading a Model\n",
    "\n",
    "To reload a model saved by the `autotune()` function or by using `torch.save(model)`:\n",
    "\n",
    "``` python\n",
    "import torch\n",
    "\n",
    "model = torch.load('trained_vae_full.pt', map_location = device)\n",
    "\n",
    "```\n",
    "Here, device would be whatever device you want to use with the model (eg 'cpu', 'cuda')\n",
    "\n",
    "To reload the model for imputation or further training from a `state_dict()`:   \n",
    "1. Re-create the model architecture with the same settings used during training  \n",
    "2. Load the saved weights  \n",
    "\n",
    "``` python\n",
    "from ciss_vae.classes.vae import CISSVAE\n",
    "\n",
    "# 1. Define the architecture (must match the saved model!)\n",
    "model = CISSVAE(\n",
    "    input_dim=...,\n",
    "    hidden_dims=[...],\n",
    "    layer_order_enc=[...],\n",
    "    layer_order_dec=[...],\n",
    "    latent_shared=...,\n",
    "    num_clusters=...,\n",
    "    latent_dim=...,\n",
    "    output_shared=...\n",
    ")\n",
    "model.load_state_dict(torch.load(\"trained_vae.pt\"))\n",
    "\n",
    "\n",
    "\n",
    "## optional to get imputed dataset. \n",
    "from ciss_vae.utils.helpers import get_imputed_df\n",
    "from torch.utils.data import DataLoader\n",
    "\n",
    "## assuming dataset is a ClusterDataset\n",
    "data_loader =  DataLoader(dataset, batch_size=4000)\n",
    "\n",
    "imputed_df = get_imputed_df(model, data_loader)\n",
    "```\n"
   ]
  }
 ],
 "metadata": {
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}