predict_neighbor_features.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "class NeighborPredictor(nn.Module):\n",
    "    def __init__(self, input_channels, output_features, num_layers, hidden_dim):\n",
    "        super(NeighborPredictor, self).__init__()\n",
    "\n",
    "        self.fc_layers = nn.ModuleList([])\n",
    "\n",
    "        for i in range(num_layers-1):\n",
    "            in_feats = input_channels if i==0 else hidden_dim\n",
    "            self.fc_layers.append(nn.Linear(in_feats, hidden_dim)) \n",
    "        \n",
    "        self.output_fc = nn.Linear(hidden_dim, output_features)\n",
    "    \n",
    "    def forward(self, x):\n",
    "        for layer in self.fc_layers:\n",
    "            x = F.relu(layer(x))\n",
    "        x = F.sigmoid(self.output_fc(x))\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/iailab42/kanea0/my-env/lib/python3.10/site-packages/torch_geometric/data/dataset.py:238: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.\n",
      "  if osp.exists(f) and torch.load(f) != _repr(self.pre_transform):\n",
      "/home/iailab42/kanea0/my-env/lib/python3.10/site-packages/torch_geometric/data/dataset.py:246: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.\n",
      "  if osp.exists(f) and torch.load(f) != _repr(self.pre_filter):\n",
      "/home/iailab42/kanea0/my-env/lib/python3.10/site-packages/torch_geometric/io/fs.py:215: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.\n",
      "  return torch.load(f, map_location)\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "import torch.nn.functional as F\n",
    "\n",
    "import torch_geometric.transforms as T\n",
    "from torch_geometric import datasets,utils\n",
    "\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "##from modules.heteroGNN import HeteroGNN\n",
    "from utils import graph_polluters\n",
    "from tqdm import tqdm, trange\n",
    "\n",
    "# We initialize conference node features with a single one-vector as feature:\n",
    "dataset = datasets.DBLP('./data/dblp', transform=T.Constant(node_types='conference'))\n",
    "data = dataset[0]\n",
    "data = graph_polluters.remove_edges(data, 0.5, [('author','to','paper')], [('paper','to','author')])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "edge_type = ('author', 'to', 'paper')\n",
    "from_type, to_type = edge_type[0], edge_type[2]\n",
    "to_features = []\n",
    "\n",
    "degrees = utils.degree(data[edge_type].edge_index[0], num_nodes=data[from_type].x.shape[0])\n",
    "mask = degrees != 0\n",
    "\n",
    "for i in range(data[from_type].x.shape[0]):\n",
    "    if degrees[i] == 0:\n",
    "        continue\n",
    "    _, edge_index, _, _ = utils.k_hop_subgraph(i, 1, data[edge_type].edge_index, directed=True, flow='target_to_source')\n",
    "    to_indices = edge_index[0,:]\n",
    "    to_features.append(torch.mean(data[to_type].x[to_indices], dim=0))\n",
    "to_features = torch.stack(to_features)#[mask]\n",
    "from_features = data[from_type].x[mask]\n",
    "max_dim = max(from_features.shape[1], to_features.shape[1]) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[0., 0., 0.,  ..., 0., 0., 0.],\n",
       "        [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "        [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "        ...,\n",
       "        [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "        [0., 0., 0.,  ..., 0., 0., 0.],\n",
       "        [0., 0., 0.,  ..., 0., 0., 0.]])"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "to_features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 50/50 [00:01<00:00, 32.53it/s]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7fb9e972bfd0>]"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "epochs = 50\n",
    "outputs = []\n",
    "losses = []\n",
    "val_losses = []\n",
    "# Validation using MSE Loss function\n",
    "loss_function = torch.nn.MSELoss()\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "from_features = from_features.to(device)\n",
    "to_features = to_features.to(device)\n",
    "\n",
    "size = from_features.shape[0]\n",
    "\n",
    "from_features_train = from_features[:size//2,:]\n",
    "to_features_train = to_features[:size//2,:]\n",
    "from_features_val = from_features[size//2:,:]\n",
    "to_features_val = to_features[size//2:,:]\n",
    "\n",
    "model = NeighborPredictor(from_features.shape[1], to_features.shape[1], 3, max_dim).to(device)\n",
    "\n",
    "with torch.no_grad():  # Initialize lazy modules.\n",
    "    out = model(from_features)\n",
    "\n",
    "optimizer = torch.optim.Adam(model.parameters(),\n",
    "                             weight_decay = 1e-8)\n",
    "\n",
    "for epoch in trange(epochs):\n",
    "    prediction = model(from_features_train)\n",
    "    loss = loss_function(prediction, to_features_train)\n",
    "    \n",
    "    optimizer.zero_grad()\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "    \n",
    "    # Storing the losses in a list for plotting\n",
    "    losses.append(loss.item())\n",
    "    \n",
    "    with torch.no_grad():\n",
    "        prediction = model(from_features_val)\n",
    "        loss = loss_function(prediction, to_features_val)\n",
    "        val_losses.append(loss.item())\n",
    "\n",
    " \n",
    "# plot chart\n",
    "plt.xlabel('Iterations')\n",
    "plt.ylabel('Loss')\n",
    "plt.plot(val_losses)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([3141, 4231])"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([3141, 4231])"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "to_features.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(0.2501, device='cuda:0')"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "loss_function(out, to_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n",
      "tensor(84)\n"
     ]
    }
   ],
   "source": [
    "#connect low-degree nodes to node with features most similar to prediction\n",
    "degrees = utils.degree(data[edge_type].edge_index[0], num_nodes=data[from_type].x.shape[0])\n",
    "low_degree_mask = degrees == 0\n",
    "num_to_nodes = data[to_type].x.shape[0]\n",
    "low_degree_nodes = data[from_type].x[low_degree_mask]\n",
    "predictions = model(low_degree_nodes.to(device)).cpu()\n",
    "for node,prediction,from_index in zip(low_degree_nodes,predictions,low_degree_mask.nonzero()):\n",
    "    preds_extended = torch.ones((num_to_nodes,1)) @ prediction.unsqueeze(0)\n",
    "    similarities = torch.cosine_similarity(preds_extended,data[to_type].x)\n",
    "    print(torch.argmax(similarities, dim=0))\n",
    "    \n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
       "        0., 0., 0., 0., 0., 0., 1., 0., 0., 0.])"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data[from_type].x[low_degree_mask][0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([1.9606e-18, 2.3999e-18, 1.4657e-18,  ..., 3.9491e-18, 1.9398e-18,\n",
       "        2.2068e-18], device='cuda:0', grad_fn=<SelectBackward0>)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model(data[from_type].x[low_degree_mask].to(device))[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[   6],\n",
       "        [  15],\n",
       "        [  21],\n",
       "        [  33],\n",
       "        [  34],\n",
       "        [  37],\n",
       "        [  46],\n",
       "        [  48],\n",
       "        [  51],\n",
       "        [  62],\n",
       "        [  66],\n",
       "        [  69],\n",
       "        [ 108],\n",
       "        [ 109],\n",
       "        [ 115],\n",
       "        [ 116],\n",
       "        [ 122],\n",
       "        [ 131],\n",
       "        [ 133],\n",
       "        [ 138],\n",
       "        [ 159],\n",
       "        [ 162],\n",
       "        [ 166],\n",
       "        [ 170],\n",
       "        [ 173],\n",
       "        [ 177],\n",
       "        [ 185],\n",
       "        [ 201],\n",
       "        [ 208],\n",
       "        [ 211],\n",
       "        [ 213],\n",
       "        [ 216],\n",
       "        [ 223],\n",
       "        [ 228],\n",
       "        [ 237],\n",
       "        [ 261],\n",
       "        [ 272],\n",
       "        [ 275],\n",
       "        [ 279],\n",
       "        [ 283],\n",
       "        [ 291],\n",
       "        [ 292],\n",
       "        [ 302],\n",
       "        [ 311],\n",
       "        [ 312],\n",
       "        [ 316],\n",
       "        [ 323],\n",
       "        [ 328],\n",
       "        [ 330],\n",
       "        [ 343],\n",
       "        [ 349],\n",
       "        [ 376],\n",
       "        [ 426],\n",
       "        [ 434],\n",
       "        [ 446],\n",
       "        [ 448],\n",
       "        [ 462],\n",
       "        [ 475],\n",
       "        [ 482],\n",
       "        [ 491],\n",
       "        [ 500],\n",
       "        [ 507],\n",
       "        [ 509],\n",
       "        [ 517],\n",
       "        [ 523],\n",
       "        [ 526],\n",
       "        [ 531],\n",
       "        [ 538],\n",
       "        [ 586],\n",
       "        [ 589],\n",
       "        [ 593],\n",
       "        [ 611],\n",
       "        [ 621],\n",
       "        [ 622],\n",
       "        [ 627],\n",
       "        [ 635],\n",
       "        [ 653],\n",
       "        [ 672],\n",
       "        [ 685],\n",
       "        [ 694],\n",
       "        [ 696],\n",
       "        [ 701],\n",
       "        [ 703],\n",
       "        [ 724],\n",
       "        [ 751],\n",
       "        [ 752],\n",
       "        [ 765],\n",
       "        [ 770],\n",
       "        [ 777],\n",
       "        [ 795],\n",
       "        [ 806],\n",
       "        [ 825],\n",
       "        [ 831],\n",
       "        [ 835],\n",
       "        [ 842],\n",
       "        [ 848],\n",
       "        [ 860],\n",
       "        [ 864],\n",
       "        [ 866],\n",
       "        [ 878],\n",
       "        [ 881],\n",
       "        [ 886],\n",
       "        [ 887],\n",
       "        [ 892],\n",
       "        [ 899],\n",
       "        [ 903],\n",
       "        [ 909],\n",
       "        [ 914],\n",
       "        [ 916],\n",
       "        [ 918],\n",
       "        [ 924],\n",
       "        [ 927],\n",
       "        [ 935],\n",
       "        [ 936],\n",
       "        [ 947],\n",
       "        [ 948],\n",
       "        [ 953],\n",
       "        [ 956],\n",
       "        [ 957],\n",
       "        [ 969],\n",
       "        [ 972],\n",
       "        [ 978],\n",
       "        [ 983],\n",
       "        [ 987],\n",
       "        [ 999],\n",
       "        [1019],\n",
       "        [1026],\n",
       "        [1041],\n",
       "        [1045],\n",
       "        [1049],\n",
       "        [1050],\n",
       "        [1057],\n",
       "        [1059],\n",
       "        [1064],\n",
       "        [1066],\n",
       "        [1067],\n",
       "        [1074],\n",
       "        [1080],\n",
       "        [1085],\n",
       "        [1094],\n",
       "        [1095],\n",
       "        [1100],\n",
       "        [1114],\n",
       "        [1126],\n",
       "        [1130],\n",
       "        [1134],\n",
       "        [1146],\n",
       "        [1164],\n",
       "        [1170],\n",
       "        [1182],\n",
       "        [1188],\n",
       "        [1195],\n",
       "        [1200],\n",
       "        [1201],\n",
       "        [1205],\n",
       "        [1213],\n",
       "        [1217],\n",
       "        [1222],\n",
       "        [1224],\n",
       "        [1234],\n",
       "        [1236],\n",
       "        [1237],\n",
       "        [1240],\n",
       "        [1241],\n",
       "        [1244],\n",
       "        [1245],\n",
       "        [1253],\n",
       "        [1260],\n",
       "        [1261],\n",
       "        [1263],\n",
       "        [1266],\n",
       "        [1268],\n",
       "        [1272],\n",
       "        [1275],\n",
       "        [1293],\n",
       "        [1295],\n",
       "        [1298],\n",
       "        [1300],\n",
       "        [1307],\n",
       "        [1315],\n",
       "        [1318],\n",
       "        [1322],\n",
       "        [1327],\n",
       "        [1329],\n",
       "        [1333],\n",
       "        [1334],\n",
       "        [1337],\n",
       "        [1339],\n",
       "        [1348],\n",
       "        [1356],\n",
       "        [1381],\n",
       "        [1382],\n",
       "        [1397],\n",
       "        [1430],\n",
       "        [1434],\n",
       "        [1436],\n",
       "        [1438],\n",
       "        [1444],\n",
       "        [1457],\n",
       "        [1466],\n",
       "        [1469],\n",
       "        [1471],\n",
       "        [1472],\n",
       "        [1478],\n",
       "        [1486],\n",
       "        [1489],\n",
       "        [1492],\n",
       "        [1505],\n",
       "        [1506],\n",
       "        [1518],\n",
       "        [1519],\n",
       "        [1522],\n",
       "        [1524],\n",
       "        [1526],\n",
       "        [1527],\n",
       "        [1528],\n",
       "        [1541],\n",
       "        [1557],\n",
       "        [1558],\n",
       "        [1580],\n",
       "        [1583],\n",
       "        [1584],\n",
       "        [1587],\n",
       "        [1592],\n",
       "        [1598],\n",
       "        [1600],\n",
       "        [1602],\n",
       "        [1616],\n",
       "        [1619],\n",
       "        [1622],\n",
       "        [1626],\n",
       "        [1630],\n",
       "        [1649],\n",
       "        [1656],\n",
       "        [1658],\n",
       "        [1667],\n",
       "        [1671],\n",
       "        [1675],\n",
       "        [1678],\n",
       "        [1682],\n",
       "        [1683],\n",
       "        [1688],\n",
       "        [1689],\n",
       "        [1691],\n",
       "        [1692],\n",
       "        [1693],\n",
       "        [1695],\n",
       "        [1699],\n",
       "        [1702],\n",
       "        [1703],\n",
       "        [1704],\n",
       "        [1716],\n",
       "        [1737],\n",
       "        [1744],\n",
       "        [1747],\n",
       "        [1750],\n",
       "        [1751],\n",
       "        [1758],\n",
       "        [1759],\n",
       "        [1765],\n",
       "        [1766],\n",
       "        [1775],\n",
       "        [1780],\n",
       "        [1784],\n",
       "        [1790],\n",
       "        [1795],\n",
       "        [1797],\n",
       "        [1798],\n",
       "        [1805],\n",
       "        [1810],\n",
       "        [1819],\n",
       "        [1827],\n",
       "        [1838],\n",
       "        [1840],\n",
       "        [1847],\n",
       "        [1849],\n",
       "        [1851],\n",
       "        [1852],\n",
       "        [1853],\n",
       "        [1856],\n",
       "        [1860],\n",
       "        [1862],\n",
       "        [1866],\n",
       "        [1867],\n",
       "        [1868],\n",
       "        [1876],\n",
       "        [1885],\n",
       "        [1889],\n",
       "        [1897],\n",
       "        [1898],\n",
       "        [1903],\n",
       "        [1904],\n",
       "        [1907],\n",
       "        [1916],\n",
       "        [1920],\n",
       "        [1927],\n",
       "        [1931],\n",
       "        [1934],\n",
       "        [1936],\n",
       "        [1937],\n",
       "        [1938],\n",
       "        [1940],\n",
       "        [1952],\n",
       "        [1953],\n",
       "        [1955],\n",
       "        [1964],\n",
       "        [1966],\n",
       "        [1967],\n",
       "        [1980],\n",
       "        [1989],\n",
       "        [1998],\n",
       "        [2001],\n",
       "        [2006],\n",
       "        [2012],\n",
       "        [2019],\n",
       "        [2024],\n",
       "        [2030],\n",
       "        [2034],\n",
       "        [2036],\n",
       "        [2090],\n",
       "        [2094],\n",
       "        [2102],\n",
       "        [2104],\n",
       "        [2111],\n",
       "        [2120],\n",
       "        [2126],\n",
       "        [2131],\n",
       "        [2137],\n",
       "        [2138],\n",
       "        [2141],\n",
       "        [2150],\n",
       "        [2158],\n",
       "        [2162],\n",
       "        [2169],\n",
       "        [2178],\n",
       "        [2182],\n",
       "        [2184],\n",
       "        [2193],\n",
       "        [2197],\n",
       "        [2199],\n",
       "        [2203],\n",
       "        [2208],\n",
       "        [2210],\n",
       "        [2212],\n",
       "        [2215],\n",
       "        [2219],\n",
       "        [2221],\n",
       "        [2230],\n",
       "        [2232],\n",
       "        [2247],\n",
       "        [2263],\n",
       "        [2266],\n",
       "        [2276],\n",
       "        [2282],\n",
       "        [2290],\n",
       "        [2291],\n",
       "        [2301],\n",
       "        [2311],\n",
       "        [2312],\n",
       "        [2324],\n",
       "        [2343],\n",
       "        [2362],\n",
       "        [2366],\n",
       "        [2371],\n",
       "        [2389],\n",
       "        [2398],\n",
       "        [2405],\n",
       "        [2429],\n",
       "        [2445],\n",
       "        [2450],\n",
       "        [2451],\n",
       "        [2462],\n",
       "        [2469],\n",
       "        [2471],\n",
       "        [2481],\n",
       "        [2482],\n",
       "        [2484],\n",
       "        [2488],\n",
       "        [2493],\n",
       "        [2498],\n",
       "        [2502],\n",
       "        [2504],\n",
       "        [2509],\n",
       "        [2516],\n",
       "        [2517],\n",
       "        [2525],\n",
       "        [2529],\n",
       "        [2538],\n",
       "        [2544],\n",
       "        [2554],\n",
       "        [2567],\n",
       "        [2573],\n",
       "        [2576],\n",
       "        [2577],\n",
       "        [2578],\n",
       "        [2580],\n",
       "        [2581],\n",
       "        [2591],\n",
       "        [2592],\n",
       "        [2593],\n",
       "        [2601],\n",
       "        [2604],\n",
       "        [2606],\n",
       "        [2608],\n",
       "        [2612],\n",
       "        [2616],\n",
       "        [2618],\n",
       "        [2619],\n",
       "        [2626],\n",
       "        [2630],\n",
       "        [2632],\n",
       "        [2633],\n",
       "        [2634],\n",
       "        [2641],\n",
       "        [2651],\n",
       "        [2658],\n",
       "        [2682],\n",
       "        [2689],\n",
       "        [2692],\n",
       "        [2693],\n",
       "        [2695],\n",
       "        [2700],\n",
       "        [2702],\n",
       "        [2704],\n",
       "        [2711],\n",
       "        [2716],\n",
       "        [2717],\n",
       "        [2719],\n",
       "        [2733],\n",
       "        [2743],\n",
       "        [2747],\n",
       "        [2750],\n",
       "        [2753],\n",
       "        [2758],\n",
       "        [2759],\n",
       "        [2763],\n",
       "        [2766],\n",
       "        [2782],\n",
       "        [2784],\n",
       "        [2788],\n",
       "        [2789],\n",
       "        [2793],\n",
       "        [2795],\n",
       "        [2801],\n",
       "        [2803],\n",
       "        [2812],\n",
       "        [2817],\n",
       "        [2824],\n",
       "        [2829],\n",
       "        [2830],\n",
       "        [2840],\n",
       "        [2843],\n",
       "        [2846],\n",
       "        [2848],\n",
       "        [2849],\n",
       "        [2858],\n",
       "        [2859],\n",
       "        [2862],\n",
       "        [2863],\n",
       "        [2877],\n",
       "        [2878],\n",
       "        [2881],\n",
       "        [2888],\n",
       "        [2893],\n",
       "        [2896],\n",
       "        [2900],\n",
       "        [2907],\n",
       "        [2914],\n",
       "        [2918],\n",
       "        [2919],\n",
       "        [2928],\n",
       "        [2936],\n",
       "        [2938],\n",
       "        [2941],\n",
       "        [2953],\n",
       "        [2958],\n",
       "        [2959],\n",
       "        [2961],\n",
       "        [2964],\n",
       "        [2965],\n",
       "        [2967],\n",
       "        [2975],\n",
       "        [2976],\n",
       "        [2980],\n",
       "        [2983],\n",
       "        [2986],\n",
       "        [2987],\n",
       "        [2990],\n",
       "        [2991],\n",
       "        [2994],\n",
       "        [2996],\n",
       "        [2999],\n",
       "        [3002],\n",
       "        [3005],\n",
       "        [3009],\n",
       "        [3010],\n",
       "        [3011],\n",
       "        [3016],\n",
       "        [3017],\n",
       "        [3019],\n",
       "        [3021],\n",
       "        [3026],\n",
       "        [3027],\n",
       "        [3029],\n",
       "        [3031],\n",
       "        [3033],\n",
       "        [3034],\n",
       "        [3037],\n",
       "        [3044],\n",
       "        [3046],\n",
       "        [3049],\n",
       "        [3050],\n",
       "        [3058],\n",
       "        [3059],\n",
       "        [3067],\n",
       "        [3068],\n",
       "        [3071],\n",
       "        [3072],\n",
       "        [3073],\n",
       "        [3077],\n",
       "        [3079],\n",
       "        [3080],\n",
       "        [3082],\n",
       "        [3085],\n",
       "        [3087],\n",
       "        [3089],\n",
       "        [3094],\n",
       "        [3096],\n",
       "        [3097],\n",
       "        [3100],\n",
       "        [3101],\n",
       "        [3102],\n",
       "        [3103],\n",
       "        [3105],\n",
       "        [3107],\n",
       "        [3108],\n",
       "        [3109],\n",
       "        [3110],\n",
       "        [3112],\n",
       "        [3116],\n",
       "        [3117],\n",
       "        [3119],\n",
       "        [3130],\n",
       "        [3137],\n",
       "        [3139],\n",
       "        [3145],\n",
       "        [3149],\n",
       "        [3151],\n",
       "        [3153],\n",
       "        [3155],\n",
       "        [3156],\n",
       "        [3157],\n",
       "        [3158],\n",
       "        [3159],\n",
       "        [3160],\n",
       "        [3168],\n",
       "        [3169],\n",
       "        [3172],\n",
       "        [3173],\n",
       "        [3175],\n",
       "        [3178],\n",
       "        [3180],\n",
       "        [3187],\n",
       "        [3188],\n",
       "        [3193],\n",
       "        [3195],\n",
       "        [3200],\n",
       "        [3201],\n",
       "        [3211],\n",
       "        [3214],\n",
       "        [3216],\n",
       "        [3223],\n",
       "        [3227],\n",
       "        [3231],\n",
       "        [3232],\n",
       "        [3244],\n",
       "        [3260],\n",
       "        [3262],\n",
       "        [3263],\n",
       "        [3264],\n",
       "        [3271],\n",
       "        [3273],\n",
       "        [3277],\n",
       "        [3285],\n",
       "        [3287],\n",
       "        [3291],\n",
       "        [3292],\n",
       "        [3295],\n",
       "        [3297],\n",
       "        [3299],\n",
       "        [3300],\n",
       "        [3303],\n",
       "        [3305],\n",
       "        [3306],\n",
       "        [3307],\n",
       "        [3309],\n",
       "        [3311],\n",
       "        [3315],\n",
       "        [3323],\n",
       "        [3324],\n",
       "        [3325],\n",
       "        [3333],\n",
       "        [3334],\n",
       "        [3335],\n",
       "        [3338],\n",
       "        [3340],\n",
       "        [3341],\n",
       "        [3342],\n",
       "        [3343],\n",
       "        [3346],\n",
       "        [3350],\n",
       "        [3351],\n",
       "        [3354],\n",
       "        [3356],\n",
       "        [3357],\n",
       "        [3358],\n",
       "        [3362],\n",
       "        [3366],\n",
       "        [3370],\n",
       "        [3372],\n",
       "        [3376],\n",
       "        [3377],\n",
       "        [3378],\n",
       "        [3385],\n",
       "        [3386],\n",
       "        [3389],\n",
       "        [3392],\n",
       "        [3394],\n",
       "        [3395],\n",
       "        [3396],\n",
       "        [3397],\n",
       "        [3398],\n",
       "        [3400],\n",
       "        [3401],\n",
       "        [3403],\n",
       "        [3406],\n",
       "        [3407],\n",
       "        [3409],\n",
       "        [3414],\n",
       "        [3419],\n",
       "        [3421],\n",
       "        [3423],\n",
       "        [3426],\n",
       "        [3429],\n",
       "        [3430],\n",
       "        [3431],\n",
       "        [3432],\n",
       "        [3436],\n",
       "        [3437],\n",
       "        [3438],\n",
       "        [3441],\n",
       "        [3444],\n",
       "        [3446],\n",
       "        [3447],\n",
       "        [3448],\n",
       "        [3449],\n",
       "        [3451],\n",
       "        [3452],\n",
       "        [3453],\n",
       "        [3454],\n",
       "        [3456],\n",
       "        [3459],\n",
       "        [3462],\n",
       "        [3465],\n",
       "        [3466],\n",
       "        [3467],\n",
       "        [3469],\n",
       "        [3474],\n",
       "        [3475],\n",
       "        [3477],\n",
       "        [3486],\n",
       "        [3489],\n",
       "        [3493],\n",
       "        [3497],\n",
       "        [3498],\n",
       "        [3500],\n",
       "        [3505],\n",
       "        [3507],\n",
       "        [3508],\n",
       "        [3517],\n",
       "        [3519],\n",
       "        [3520],\n",
       "        [3522],\n",
       "        [3532],\n",
       "        [3534],\n",
       "        [3535],\n",
       "        [3536],\n",
       "        [3538],\n",
       "        [3539],\n",
       "        [3540],\n",
       "        [3541],\n",
       "        [3542],\n",
       "        [3548],\n",
       "        [3550],\n",
       "        [3552],\n",
       "        [3553],\n",
       "        [3556],\n",
       "        [3557],\n",
       "        [3561],\n",
       "        [3564],\n",
       "        [3566],\n",
       "        [3569],\n",
       "        [3571],\n",
       "        [3574],\n",
       "        [3575],\n",
       "        [3577],\n",
       "        [3578],\n",
       "        [3579],\n",
       "        [3581],\n",
       "        [3582],\n",
       "        [3583],\n",
       "        [3585],\n",
       "        [3586],\n",
       "        [3588],\n",
       "        [3590],\n",
       "        [3597],\n",
       "        [3601],\n",
       "        [3602],\n",
       "        [3610],\n",
       "        [3611],\n",
       "        [3613],\n",
       "        [3614],\n",
       "        [3617],\n",
       "        [3618],\n",
       "        [3620],\n",
       "        [3623],\n",
       "        [3626],\n",
       "        [3627],\n",
       "        [3628],\n",
       "        [3630],\n",
       "        [3631],\n",
       "        [3633],\n",
       "        [3634],\n",
       "        [3635],\n",
       "        [3636],\n",
       "        [3646],\n",
       "        [3647],\n",
       "        [3651],\n",
       "        [3656],\n",
       "        [3657],\n",
       "        [3659],\n",
       "        [3664],\n",
       "        [3665],\n",
       "        [3666],\n",
       "        [3667],\n",
       "        [3670],\n",
       "        [3671],\n",
       "        [3673],\n",
       "        [3674],\n",
       "        [3677],\n",
       "        [3678],\n",
       "        [3679],\n",
       "        [3681],\n",
       "        [3686],\n",
       "        [3690],\n",
       "        [3692],\n",
       "        [3695],\n",
       "        [3700],\n",
       "        [3703],\n",
       "        [3704],\n",
       "        [3705],\n",
       "        [3709],\n",
       "        [3710],\n",
       "        [3711],\n",
       "        [3714],\n",
       "        [3717],\n",
       "        [3718],\n",
       "        [3720],\n",
       "        [3721],\n",
       "        [3722],\n",
       "        [3723],\n",
       "        [3724],\n",
       "        [3728],\n",
       "        [3729],\n",
       "        [3730],\n",
       "        [3733],\n",
       "        [3735],\n",
       "        [3738],\n",
       "        [3740],\n",
       "        [3747],\n",
       "        [3751],\n",
       "        [3756],\n",
       "        [3759],\n",
       "        [3761],\n",
       "        [3762],\n",
       "        [3763],\n",
       "        [3764],\n",
       "        [3765],\n",
       "        [3767],\n",
       "        [3769],\n",
       "        [3771],\n",
       "        [3772],\n",
       "        [3773],\n",
       "        [3774],\n",
       "        [3775],\n",
       "        [3777],\n",
       "        [3780],\n",
       "        [3783],\n",
       "        [3785],\n",
       "        [3787],\n",
       "        [3789],\n",
       "        [3793],\n",
       "        [3796],\n",
       "        [3797],\n",
       "        [3798],\n",
       "        [3799],\n",
       "        [3801],\n",
       "        [3802],\n",
       "        [3803],\n",
       "        [3805],\n",
       "        [3808],\n",
       "        [3809],\n",
       "        [3811],\n",
       "        [3812],\n",
       "        [3816],\n",
       "        [3817],\n",
       "        [3819],\n",
       "        [3820],\n",
       "        [3821],\n",
       "        [3825],\n",
       "        [3828],\n",
       "        [3830],\n",
       "        [3832],\n",
       "        [3834],\n",
       "        [3835],\n",
       "        [3838],\n",
       "        [3839],\n",
       "        [3842],\n",
       "        [3845],\n",
       "        [3846],\n",
       "        [3850],\n",
       "        [3855],\n",
       "        [3857],\n",
       "        [3858],\n",
       "        [3859],\n",
       "        [3860],\n",
       "        [3861],\n",
       "        [3865],\n",
       "        [3868],\n",
       "        [3869],\n",
       "        [3872],\n",
       "        [3876],\n",
       "        [3879],\n",
       "        [3880],\n",
       "        [3881],\n",
       "        [3882],\n",
       "        [3883],\n",
       "        [3885],\n",
       "        [3886],\n",
       "        [3888],\n",
       "        [3891],\n",
       "        [3893],\n",
       "        [3899],\n",
       "        [3901],\n",
       "        [3904],\n",
       "        [3905],\n",
       "        [3906],\n",
       "        [3917],\n",
       "        [3919],\n",
       "        [3920],\n",
       "        [3921],\n",
       "        [3922],\n",
       "        [3927],\n",
       "        [3929],\n",
       "        [3931],\n",
       "        [3932],\n",
       "        [3934],\n",
       "        [3936],\n",
       "        [3938],\n",
       "        [3944],\n",
       "        [3945],\n",
       "        [3955],\n",
       "        [3956],\n",
       "        [3957],\n",
       "        [3959],\n",
       "        [3963],\n",
       "        [3964],\n",
       "        [3966],\n",
       "        [3972],\n",
       "        [3973],\n",
       "        [3974],\n",
       "        [3976],\n",
       "        [3977],\n",
       "        [3979],\n",
       "        [3980],\n",
       "        [3981],\n",
       "        [3984],\n",
       "        [3986],\n",
       "        [3987],\n",
       "        [3988],\n",
       "        [3989],\n",
       "        [3996],\n",
       "        [4000],\n",
       "        [4004],\n",
       "        [4005],\n",
       "        [4008],\n",
       "        [4010],\n",
       "        [4012],\n",
       "        [4015],\n",
       "        [4016],\n",
       "        [4019],\n",
       "        [4024],\n",
       "        [4026],\n",
       "        [4027],\n",
       "        [4029],\n",
       "        [4030],\n",
       "        [4031],\n",
       "        [4033],\n",
       "        [4035],\n",
       "        [4042],\n",
       "        [4043],\n",
       "        [4044],\n",
       "        [4045],\n",
       "        [4046],\n",
       "        [4047],\n",
       "        [4048],\n",
       "        [4050]])"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "low_degree_mask.nonzero()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.mask_in"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "a = torch.randn(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([ 2.1277,  0.5740, -0.6464, -1.6797,  0.4067])"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "a"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "b = torch.ones((5)) @ a"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([])"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "b.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "a = torch.eye(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "b = torch.as_tensor([[0., 0., 0.], [1, 1., 0.], [1., 0., 0.]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "c = torch.cosine_similarity(a,b)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor(1)"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.argmax(c, dim=0)"
   ]
  }
 ],
 "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": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}