Trying to get S4 running (no success)

Dockerfile

@@ -1,5 +1,7 @@
-FROM pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime
 
+FROM pytorch/pytorch:1.13.1-cuda11.6-cudnn8-devel
+#FROM getkeops/keops-full:2.1-geomloss0.2.5-cuda11.8-pytorch2.0.0-python3.10
+# FROM pytorch/pytorch:2.1.0-cuda11.8-cudnn8-devel
 RUN apt-get update
 RUN apt-get install -y git
 

requirements.txt

@@ -11,4 +11,6 @@ clearml
 properscoring
 nbconvert
 torchinfo
-tabulate
+tabulate
+einops
+opt_einsum

src/models/tsdiff_s4/backbones.py

@@ -0,0 +1,172 @@
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+# SPDX-License-Identifier: Apache-2.0
+import math
+
+import torch
+from torch import nn
+
+from src.models.tsdiff_s4.s4 import S4
+
+
+class SinusoidalPositionEmbeddings(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        self.dim = dim
+
+    def forward(self, time):
+        device = time.device
+        half_dim = self.dim // 2
+        embeddings = math.log(10000) / (half_dim - 1)
+        embeddings = torch.exp(
+            torch.arange(half_dim, device=device) * -embeddings
+        )
+        embeddings = time[:, None] * embeddings[None, :]
+        embeddings = torch.cat((embeddings.sin(), embeddings.cos()), dim=-1)
+        return embeddings
+
+
+class S4Layer(nn.Module):
+    def __init__(
+        self,
+        d_model,
+        dropout=0.0,
+    ):
+        super().__init__()
+        self.layer = S4(
+            d_model=d_model,
+            d_state=128,
+            bidirectional=True,
+            dropout=dropout,
+            transposed=True,
+            postact=None,
+        )
+        self.norm = nn.LayerNorm(d_model)
+        self.dropout = (
+            nn.Dropout1d(dropout) if dropout > 0.0 else nn.Identity()
+        )
+
+    def forward(self, x):
+        """
+        Input x is shape (B, d_input, L)
+        """
+        z = x
+        # Prenorm
+        z = self.norm(z.transpose(-1, -2)).transpose(-1, -2)
+        # Apply layer: we ignore the state input and output for training
+        z, _ = self.layer(z)
+        # Dropout on the output of the layer
+        z = self.dropout(z)
+        # Residual connection
+        x = z + x
+        return x, None
+
+    def default_state(self, *args, **kwargs):
+        return self.layer.default_state(*args, **kwargs)
+
+    def step(self, x, state, **kwargs):
+        z = x
+        # Prenorm
+        z = self.norm(z.transpose(-1, -2)).transpose(-1, -2)
+        # Apply layer
+        z, state = self.layer.step(z, state, **kwargs)
+        # Residual connection
+        x = z + x
+        return x, state
+
+
+class S4Block(nn.Module):
+    def __init__(self, d_model, dropout=0.0, expand=2, num_features=0):
+        super().__init__()
+        self.s4block = S4Layer(d_model, dropout=dropout)
+
+        self.time_linear = nn.Linear(d_model, d_model)
+        self.tanh = nn.Tanh()
+        self.sigm = nn.Sigmoid()
+        self.out_linear1 = nn.Conv1d(
+            in_channels=d_model, out_channels=d_model, kernel_size=1
+        )
+        self.out_linear2 = nn.Conv1d(
+            in_channels=d_model, out_channels=d_model, kernel_size=1
+        )
+        self.feature_encoder = nn.Conv1d(num_features, d_model, kernel_size=1)
+
+    def forward(self, x, t, features=None):
+        t = self.time_linear(t)[:, None, :].repeat(1, x.shape[2], 1)
+        t = t.transpose(-1, -2)
+        out, _ = self.s4block(x + t)
+        if features is not None:
+            out = out + self.feature_encoder(features)
+        out = self.tanh(out) * self.sigm(out)
+        out1 = self.out_linear1(out)
+        out2 = self.out_linear2(out)
+        return out1 + x, out2
+
+
+def Conv1dKaiming(in_channels, out_channels, kernel_size):
+    layer = nn.Conv1d(in_channels, out_channels, kernel_size)
+    nn.init.kaiming_normal_(layer.weight)
+    return layer
+
+
+class BackboneModel(nn.Module):
+    def __init__(
+        self,
+        input_dim,
+        hidden_dim,
+        output_dim,
+        step_emb,
+        num_residual_blocks,
+        num_features,
+        residual_block="s4",
+        dropout=0.0,
+        init_skip=True,
+    ):
+        super().__init__()
+        if residual_block == "s4":
+            residual_block = S4Block
+        else:
+            raise ValueError(f"Unknown residual block {residual_block}")
+        self.input_init = nn.Sequential(
+            nn.Linear(input_dim, hidden_dim),
+            nn.ReLU(),
+        )
+        self.time_init = nn.Sequential(
+            nn.Linear(step_emb, hidden_dim),
+            nn.SiLU(),
+            nn.Linear(hidden_dim, hidden_dim),
+            nn.SiLU(),
+        )
+        self.out_linear = nn.Sequential(
+            nn.Linear(hidden_dim, hidden_dim),
+            nn.ReLU(),
+            nn.Linear(hidden_dim, output_dim),
+        )
+        residual_blocks = []
+        for i in range(num_residual_blocks):
+            residual_blocks.append(
+                residual_block(
+                    hidden_dim, num_features=num_features, dropout=dropout
+                )
+            )
+        self.residual_blocks = nn.ModuleList(residual_blocks)
+        self.step_embedding = SinusoidalPositionEmbeddings(step_emb)
+        self.init_skip = init_skip
+
+    def forward(self, input, t, features=None):
+        x = self.input_init(input)  # B, L ,C
+        step_emb = self.step_embedding(t)
+        t = self.time_init(step_emb)
+        x = x.transpose(-1, -2)
+        if features is not None:
+            features = features.transpose(-1, -2)
+        skips = []
+        for layer in self.residual_blocks:
+            x, skip = layer(x, t, features)
+            skips.append(skip)
+
+        skip = torch.stack(skips).sum(0)
+        skip = skip.transpose(-1, -2)
+        out = self.out_linear(skip)
+        if self.init_skip:
+            out = out + input
+        return out

src/notebooks/diffusion-training.ipynb

@@ -2,20 +2,41 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 1,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/opt/conda/lib/python3.10/site-packages/tqdm/auto.py:22: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+      "  from .autonotebook import tqdm as notebook_tqdm\n"
+     ]
+    }
+   ],
    "source": [
     "import sys\n",
     "sys.path.append('../..')\n",
-    "import torch"
+    "import torch\n",
+    "\n",
+    "%load_ext autoreload\n",
+    "%autoreload 2"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": 3,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n"
+     ]
+    }
+   ],
    "source": [
     "from src.data import DataProcessor, DataConfig\n",
     "from src.trainers.quantile_trainer import AutoRegressiveQuantileTrainer, NonAutoRegressiveQuantileRegression\n",
@@ -44,7 +65,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -62,8 +83,8 @@
     "\n",
     "data_config.NOMINAL_NET_POSITION = True\n",
     "\n",
-    "data_processor = DataProcessor(data_config, path=\"../../\")\n",
+    "data_processor = DataProcessor(data_config, path=\"../../\", lstm=True)\n",
-    "data_processor.set_batch_size(1024)\n",
+    "data_processor.set_batch_size(128)\n",
     "data_processor.set_full_day_skip(True)"
    ]
   },
@@ -222,6 +243,165 @@
     "sample_diffusion(new_model, 1, inputs)"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Trying out BackboneModel using S4 state space model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[KeOps] Compiling cuda jit compiler engine ... \n",
+      "[KeOps] Warning : There were warnings or errors compiling formula :\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libstdc++.so: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010001\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libstdc++.so: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010002\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010001\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010002\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010001\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010002\n",
+      "\n",
+      "OK\n",
+      "[pyKeOps] Compiling nvrtc binder for python ... \n",
+      "[KeOps] Warning : There were warnings or errors compiling formula :\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libstdc++.so: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010001\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libstdc++.so: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010002\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010001\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010002\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010001\n",
+      "/usr/bin/ld: warning: /opt/conda/lib/libgcc_s.so.1: unsupported GNU_PROPERTY_TYPE (5) type: 0xc0010002\n",
+      "\n",
+      "OK\n"
+     ]
+    }
+   ],
+   "source": [
+    "import sys\n",
+    "sys.path.append('../..')\n",
+    "import torch\n",
+    "\n",
+    "%load_ext autoreload\n",
+    "%autoreload 2\n",
+    "from src.models.tsdiff_s4.backbones import BackboneModel\n",
+    "from src.trainers.diffusion_trainer import DiffusionTrainer\n",
+    "\n",
+    "backbone = BackboneModel(\n",
+    "    input_dim=1,\n",
+    "    hidden_dim=512,\n",
+    "    output_dim=1,\n",
+    "    step_emb=128,\n",
+    "    num_residual_blocks=3,\n",
+    "    num_features=2\n",
+    ")\n",
+    "backbone = backbone.to(\"cuda\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[KeOps] Generating code for formula Sum_Reduction(ComplexMult(Real2Complex(1/ComplexSquareAbs(ComplexMult(Var(1,2,0)-Var(2,2,1),Var(1,2,0)-Conj(Var(2,2,1))))),ComplexMult(Var(1,2,0)*ComplexReal(Var(0,2,1))-Real2Complex(Sum(Var(0,2,1)*Var(2,2,1))),Conj(ComplexMult(Var(1,2,0)-Var(2,2,1),Var(1,2,0)-Conj(Var(2,2,1)))))),0) ... "
+     ]
+    },
+    {
+     "ename": "",
+     "evalue": "",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[1;31mThe Kernel crashed while executing code in the the current cell or a previous cell. Please review the code in the cell(s) to identify a possible cause of the failure. Click <a href='https://aka.ms/vscodeJupyterKernelCrash'>here</a> for more info. View Jupyter <a href='command:jupyter.viewOutput'>log</a> for further details."
+     ]
+    }
+   ],
+   "source": [
+    "# now lets find out what the input shape of the featues and input must be\n",
+    "\n",
+    "# input: (B, L, C)\n",
+    "# features: (B, L, F)\n",
+    "# time: (B, 1)\n",
+    "\n",
+    "# output: (B, L, C)? \n",
+    "\n",
+    "input = torch.randn(2, 96, 1).to(\"cuda\")\n",
+    "features = torch.randn(2, 96, 2).to(\"cuda\")\n",
+    "times = torch.randn(2).to(\"cuda\")\n",
+    "\n",
+    "backbone(input, times, features).shape"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "KeyError",
+     "evalue": "'nvrtc'",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[3], line 13\u001b[0m\n\u001b[1;32m     10\u001b[0m times \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39mtensor([\u001b[38;5;241m0\u001b[39m]\u001b[38;5;241m*\u001b[39m\u001b[38;5;241m2\u001b[39m)\u001b[38;5;241m.\u001b[39mto(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcuda\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m     11\u001b[0m features \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39mrandn(\u001b[38;5;241m2\u001b[39m, \u001b[38;5;241m96\u001b[39m, \u001b[38;5;241m2\u001b[39m)\u001b[38;5;241m.\u001b[39mto(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcuda\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m---> 13\u001b[0m \u001b[43mmodel\u001b[49m\u001b[43m(\u001b[49m\u001b[43minputs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtimes\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mfeatures\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241m.\u001b[39mshape\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1501\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1496\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1497\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1498\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1499\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1500\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1501\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1502\u001b[0m \u001b[38;5;66;03m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1503\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[38;5;241m=\u001b[39m [], []\n",
+      "File \u001b[0;32m/workspaces/Thesis/src/notebooks/../../src/models/tsdiff_s4/backbones.py:164\u001b[0m, in \u001b[0;36mBackboneModel.forward\u001b[0;34m(self, input, t, features)\u001b[0m\n\u001b[1;32m    162\u001b[0m skips \u001b[38;5;241m=\u001b[39m []\n\u001b[1;32m    163\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m layer \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mresidual_blocks:\n\u001b[0;32m--> 164\u001b[0m     x, skip \u001b[38;5;241m=\u001b[39m \u001b[43mlayer\u001b[49m\u001b[43m(\u001b[49m\u001b[43mx\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mt\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mfeatures\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    165\u001b[0m     skips\u001b[38;5;241m.\u001b[39mappend(skip)\n\u001b[1;32m    167\u001b[0m skip \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39mstack(skips)\u001b[38;5;241m.\u001b[39msum(\u001b[38;5;241m0\u001b[39m)\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1501\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1496\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1497\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1498\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1499\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1500\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1501\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1502\u001b[0m \u001b[38;5;66;03m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1503\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[38;5;241m=\u001b[39m [], []\n",
+      "File \u001b[0;32m/workspaces/Thesis/src/notebooks/../../src/models/tsdiff_s4/backbones.py:96\u001b[0m, in \u001b[0;36mS4Block.forward\u001b[0;34m(self, x, t, features)\u001b[0m\n\u001b[1;32m     94\u001b[0m t \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtime_linear(t)[:, \u001b[38;5;28;01mNone\u001b[39;00m, :]\u001b[38;5;241m.\u001b[39mrepeat(\u001b[38;5;241m1\u001b[39m, x\u001b[38;5;241m.\u001b[39mshape[\u001b[38;5;241m2\u001b[39m], \u001b[38;5;241m1\u001b[39m)\n\u001b[1;32m     95\u001b[0m t \u001b[38;5;241m=\u001b[39m t\u001b[38;5;241m.\u001b[39mtranspose(\u001b[38;5;241m-\u001b[39m\u001b[38;5;241m1\u001b[39m, \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m2\u001b[39m)\n\u001b[0;32m---> 96\u001b[0m out, _ \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43ms4block\u001b[49m\u001b[43m(\u001b[49m\u001b[43mx\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m+\u001b[39;49m\u001b[43m \u001b[49m\u001b[43mt\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m     97\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m features \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m     98\u001b[0m     out \u001b[38;5;241m=\u001b[39m out \u001b[38;5;241m+\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mfeature_encoder(features)\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1501\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1496\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1497\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1498\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1499\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1500\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1501\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1502\u001b[0m \u001b[38;5;66;03m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1503\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[38;5;241m=\u001b[39m [], []\n",
+      "File \u001b[0;32m/workspaces/Thesis/src/notebooks/../../src/models/tsdiff_s4/backbones.py:56\u001b[0m, in \u001b[0;36mS4Layer.forward\u001b[0;34m(self, x)\u001b[0m\n\u001b[1;32m     54\u001b[0m z \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mnorm(z\u001b[38;5;241m.\u001b[39mtranspose(\u001b[38;5;241m-\u001b[39m\u001b[38;5;241m1\u001b[39m, \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m2\u001b[39m))\u001b[38;5;241m.\u001b[39mtranspose(\u001b[38;5;241m-\u001b[39m\u001b[38;5;241m1\u001b[39m, \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m2\u001b[39m)\n\u001b[1;32m     55\u001b[0m \u001b[38;5;66;03m# Apply layer: we ignore the state input and output for training\u001b[39;00m\n\u001b[0;32m---> 56\u001b[0m z, _ \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mlayer\u001b[49m\u001b[43m(\u001b[49m\u001b[43mz\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m     57\u001b[0m \u001b[38;5;66;03m# Dropout on the output of the layer\u001b[39;00m\n\u001b[1;32m     58\u001b[0m z \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdropout(z)\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1501\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1496\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1497\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1498\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1499\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1500\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1501\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1502\u001b[0m \u001b[38;5;66;03m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1503\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[38;5;241m=\u001b[39m [], []\n",
+      "File \u001b[0;32m/workspaces/Thesis/src/notebooks/../../src/models/tsdiff_s4/s4.py:1761\u001b[0m, in \u001b[0;36mS4.forward\u001b[0;34m(self, u, state, rate, lengths, **kwargs)\u001b[0m\n\u001b[1;32m   1759\u001b[0m \u001b[38;5;66;03m# Compute SS Kernel\u001b[39;00m\n\u001b[1;32m   1760\u001b[0m L_kernel \u001b[38;5;241m=\u001b[39m L \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mL \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;28;01melse\u001b[39;00m \u001b[38;5;28mmin\u001b[39m(L, \u001b[38;5;28mround\u001b[39m(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mL \u001b[38;5;241m/\u001b[39m rate))\n\u001b[0;32m-> 1761\u001b[0m k, k_state \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mkernel\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m   1762\u001b[0m \u001b[43m    \u001b[49m\u001b[43mL\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mL_kernel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mrate\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mrate\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mstate\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mstate\u001b[49m\n\u001b[1;32m   1763\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m  \u001b[38;5;66;03m# (C H L) (B C H L)\u001b[39;00m\n\u001b[1;32m   1765\u001b[0m \u001b[38;5;66;03m# Convolution\u001b[39;00m\n\u001b[1;32m   1766\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mbidirectional:\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1501\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1496\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1497\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1498\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1499\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1500\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1501\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1502\u001b[0m \u001b[38;5;66;03m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1503\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[38;5;241m=\u001b[39m [], []\n",
+      "File \u001b[0;32m/workspaces/Thesis/src/notebooks/../../src/models/tsdiff_s4/s4.py:1549\u001b[0m, in \u001b[0;36mSSKernel.forward\u001b[0;34m(self, state, L, rate)\u001b[0m\n\u001b[1;32m   1548\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, state\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mNone\u001b[39;00m, L\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mNone\u001b[39;00m, rate\u001b[38;5;241m=\u001b[39m\u001b[38;5;241m1.0\u001b[39m):\n\u001b[0;32m-> 1549\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mkernel\u001b[49m\u001b[43m(\u001b[49m\u001b[43mstate\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mstate\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mL\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mL\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mrate\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mrate\u001b[49m\u001b[43m)\u001b[49m\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1501\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1496\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1497\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1498\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1499\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1500\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1501\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1502\u001b[0m \u001b[38;5;66;03m# Do not call functions when jit is used\u001b[39;00m\n\u001b[1;32m   1503\u001b[0m full_backward_hooks, non_full_backward_hooks \u001b[38;5;241m=\u001b[39m [], []\n",
+      "File \u001b[0;32m/workspaces/Thesis/src/notebooks/../../src/models/tsdiff_s4/s4.py:925\u001b[0m, in \u001b[0;36mSSKernelNPLR.forward\u001b[0;34m(self, state, rate, L)\u001b[0m\n\u001b[1;32m    923\u001b[0m     r \u001b[38;5;241m=\u001b[39m cauchy_mult(v, z, w, symmetric\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mTrue\u001b[39;00m)\n\u001b[1;32m    924\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m has_pykeops:\n\u001b[0;32m--> 925\u001b[0m     r \u001b[38;5;241m=\u001b[39m \u001b[43mcauchy_conj\u001b[49m\u001b[43m(\u001b[49m\u001b[43mv\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mz\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mw\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    926\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m    927\u001b[0m     r \u001b[38;5;241m=\u001b[39m cauchy_naive(v, z, w)\n",
+      "File \u001b[0;32m/workspaces/Thesis/src/notebooks/../../src/models/tsdiff_s4/s4.py:89\u001b[0m, in \u001b[0;36mcauchy_conj\u001b[0;34m(v, z, w)\u001b[0m\n\u001b[1;32m     86\u001b[0m z \u001b[38;5;241m=\u001b[39m _c2r(z)\n\u001b[1;32m     87\u001b[0m w \u001b[38;5;241m=\u001b[39m _c2r(w)\n\u001b[0;32m---> 89\u001b[0m r \u001b[38;5;241m=\u001b[39m \u001b[38;5;241m2\u001b[39m \u001b[38;5;241m*\u001b[39m \u001b[43mcauchy_mult\u001b[49m\u001b[43m(\u001b[49m\u001b[43mv\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mz\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mw\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mbackend\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mGPU\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m)\u001b[49m\n\u001b[1;32m     90\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m _r2c(r)\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/pykeops/torch/generic/generic_red.py:688\u001b[0m, in \u001b[0;36mGenred.__call__\u001b[0;34m(self, backend, device_id, ranges, out, *args)\u001b[0m\n\u001b[1;32m    686\u001b[0m params\u001b[38;5;241m.\u001b[39mny \u001b[38;5;241m=\u001b[39m ny\n\u001b[1;32m    687\u001b[0m params\u001b[38;5;241m.\u001b[39mout \u001b[38;5;241m=\u001b[39m out\n\u001b[0;32m--> 688\u001b[0m out \u001b[38;5;241m=\u001b[39m \u001b[43mGenredAutograd_fun\u001b[49m\u001b[43m(\u001b[49m\u001b[43mparams\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    690\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m postprocess(out, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtorch\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mreduction_op, nout, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mopt_arg, dtype)\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/pykeops/torch/generic/generic_red.py:384\u001b[0m, in \u001b[0;36mGenredAutograd_fun\u001b[0;34m(*inputs)\u001b[0m\n\u001b[1;32m    383\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mGenredAutograd_fun\u001b[39m(\u001b[38;5;241m*\u001b[39minputs):\n\u001b[0;32m--> 384\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mGenredAutograd\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mapply\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43minputs\u001b[49m\u001b[43m)\u001b[49m[\u001b[38;5;241m0\u001b[39m]\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/autograd/function.py:506\u001b[0m, in \u001b[0;36mFunction.apply\u001b[0;34m(cls, *args, **kwargs)\u001b[0m\n\u001b[1;32m    503\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m torch\u001b[38;5;241m.\u001b[39m_C\u001b[38;5;241m.\u001b[39m_are_functorch_transforms_active():\n\u001b[1;32m    504\u001b[0m     \u001b[38;5;66;03m# See NOTE: [functorch vjp and autograd interaction]\u001b[39;00m\n\u001b[1;32m    505\u001b[0m     args \u001b[38;5;241m=\u001b[39m _functorch\u001b[38;5;241m.\u001b[39mutils\u001b[38;5;241m.\u001b[39munwrap_dead_wrappers(args)\n\u001b[0;32m--> 506\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mapply\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m    508\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mcls\u001b[39m\u001b[38;5;241m.\u001b[39msetup_context \u001b[38;5;241m==\u001b[39m _SingleLevelFunction\u001b[38;5;241m.\u001b[39msetup_context:\n\u001b[1;32m    509\u001b[0m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mRuntimeError\u001b[39;00m(\n\u001b[1;32m    510\u001b[0m         \u001b[38;5;124m'\u001b[39m\u001b[38;5;124mIn order to use an autograd.Function with functorch transforms \u001b[39m\u001b[38;5;124m'\u001b[39m\n\u001b[1;32m    511\u001b[0m         \u001b[38;5;124m'\u001b[39m\u001b[38;5;124m(vmap, grad, jvp, jacrev, ...), it must override the setup_context \u001b[39m\u001b[38;5;124m'\u001b[39m\n\u001b[1;32m    512\u001b[0m         \u001b[38;5;124m'\u001b[39m\u001b[38;5;124mstaticmethod. For more details, please see \u001b[39m\u001b[38;5;124m'\u001b[39m\n\u001b[1;32m    513\u001b[0m         \u001b[38;5;124m'\u001b[39m\u001b[38;5;124mhttps://pytorch.org/docs/master/notes/extending.func.html\u001b[39m\u001b[38;5;124m'\u001b[39m)\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/pykeops/torch/generic/generic_red.py:295\u001b[0m, in \u001b[0;36mGenredAutograd.forward\u001b[0;34m(*inputs)\u001b[0m\n\u001b[1;32m    293\u001b[0m \u001b[38;5;129m@staticmethod\u001b[39m\n\u001b[1;32m    294\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mforward\u001b[39m(\u001b[38;5;241m*\u001b[39minputs):\n\u001b[0;32m--> 295\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mGenredAutograd_base\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_forward\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43minputs\u001b[49m\u001b[43m)\u001b[49m\n",
+      "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/pykeops/torch/generic/generic_red.py:91\u001b[0m, in \u001b[0;36mGenredAutograd_base._forward\u001b[0;34m(params, *args)\u001b[0m\n\u001b[1;32m     85\u001b[0m device_id, device_args \u001b[38;5;241m=\u001b[39m set_device(\n\u001b[1;32m     86\u001b[0m     tagCPUGPU, tagHostDevice, params\u001b[38;5;241m.\u001b[39mdevice_id_request, \u001b[38;5;241m*\u001b[39margs\n\u001b[1;32m     87\u001b[0m )\n\u001b[1;32m     89\u001b[0m \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;21;01mpykeops\u001b[39;00m\u001b[38;5;21;01m.\u001b[39;00m\u001b[38;5;21;01mcommon\u001b[39;00m\u001b[38;5;21;01m.\u001b[39;00m\u001b[38;5;21;01mkeops_io\u001b[39;00m \u001b[38;5;28;01mimport\u001b[39;00m keops_binder\n\u001b[0;32m---> 91\u001b[0m myconv \u001b[38;5;241m=\u001b[39m \u001b[43mkeops_binder\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mnvrtc\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mif\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mtagCPUGPU\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01melse\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mcpp\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m]\u001b[49m(\n\u001b[1;32m     92\u001b[0m     tagCPUGPU,\n\u001b[1;32m     93\u001b[0m     tag1D2D,\n\u001b[1;32m     94\u001b[0m     tagHostDevice,\n\u001b[1;32m     95\u001b[0m     use_ranges,\n\u001b[1;32m     96\u001b[0m     device_id,\n\u001b[1;32m     97\u001b[0m     params\u001b[38;5;241m.\u001b[39mformula,\n\u001b[1;32m     98\u001b[0m     params\u001b[38;5;241m.\u001b[39maliases,\n\u001b[1;32m     99\u001b[0m     \u001b[38;5;28mlen\u001b[39m(args),\n\u001b[1;32m    100\u001b[0m     params\u001b[38;5;241m.\u001b[39mdtype,\n\u001b[1;32m    101\u001b[0m     \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtorch\u001b[39m\u001b[38;5;124m\"\u001b[39m,\n\u001b[1;32m    102\u001b[0m     params\u001b[38;5;241m.\u001b[39moptional_flags,\n\u001b[1;32m    103\u001b[0m )\u001b[38;5;241m.\u001b[39mimport_module()\n\u001b[1;32m    105\u001b[0m \u001b[38;5;66;03m# N.B.: KeOps C++ expects contiguous data arrays\u001b[39;00m\n\u001b[1;32m    106\u001b[0m test_contig \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mall\u001b[39m(arg\u001b[38;5;241m.\u001b[39mis_contiguous() \u001b[38;5;28;01mfor\u001b[39;00m arg \u001b[38;5;129;01min\u001b[39;00m args)\n",
+      "\u001b[0;31mKeyError\u001b[0m: 'nvrtc'"
+     ]
+    }
+   ],
+   "source": [
+    "# inputDim = data_processor.get_input_size()\n",
+    "learningRate = 0.0001\n",
+    "epochs=150\n",
+    "\n",
+    "#### Model ####\n",
+    "model = BackboneModel(1, 512, output_dim=1, step_emb=64, num_residual_blocks=4, num_features=2)\n",
+    "model.to(\"cuda\")\n",
+    "\n",
+    "inputs = torch.randn(2, 96, 1).to(\"cuda\")\n",
+    "times = torch.tensor([0]*2).to(\"cuda\")\n",
+    "features = torch.randn(2, 96, 2).to(\"cuda\")\n",
+    "\n",
+    "model(inputs, times, features).shape\n",
+    "\n",
+    "#### Trainer ####\n",
+    "# trainer = DiffusionTrainer(model, data_processor, \"cuda\")\n",
+    "# trainer.train(epochs, learningRate, None)"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -246,7 +426,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.11"
+   "version": "3.10.8"
   }
  },
  "nbformat": 4,

src/trainers/diffusion_trainer.py

@@ -132,6 +132,8 @@ class DiffusionTrainer:
 
                 t = self.sample_timesteps(time_series.shape[0]).to(self.device)
                 x_t, noise = self.noise_time_series(time_series, t)
+                x_t = x_t.unsqueeze(-1)
+                print(x_t.shape, t.shape, base_pattern.shape)
                 predicted_noise = self.model(x_t, t, base_pattern)
                 loss = criterion(predicted_noise, noise)
 

test.py

@@ -0,0 +1,2 @@
+import pykeops
+pykeops.test_numpy_bindings()