{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "6c6e33cb-72f9-42fa-936a-33b5fe338d15", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "torch.Size([1, 1024, 128]) torch.Size([1, 1024, 128])\n", "Causality check: gradients should not flow \"from future to past\"\n", "tensor(-4.1268e-09) tensor(0.0844)\n" ] } ], "source": [ "# %load standalone_hyena.py\n", "\"\"\"\n", "Simplified standalone version of Hyena: https://arxiv.org/abs/2302.10866, designed for quick experimentation.\n", "A complete version is available under `src.models.sequence.hyena`.\n", "\"\"\"\n", "\n", "import math\n", "import torch\n", "import torch.nn as nn\n", "import torch.nn.functional as F\n", "from einops import rearrange\n", "\n", "\n", "def fftconv(u, k, D):\n", " seqlen = u.shape[-1]\n", " fft_size = 2 * seqlen\n", " \n", " k_f = torch.fft.rfft(k, n=fft_size) / fft_size\n", " u_f = torch.fft.rfft(u.to(dtype=k.dtype), n=fft_size)\n", " \n", " if len(u.shape) > 3: k_f = k_f.unsqueeze(1)\n", " y = torch.fft.irfft(u_f * k_f, n=fft_size, norm='forward')[..., :seqlen]\n", "\n", " out = y + u * D.unsqueeze(-1)\n", " return out.to(dtype=u.dtype)\n", "\n", "\n", "@torch.jit.script \n", "def mul_sum(q, y):\n", " return (q * y).sum(dim=1)\n", "\n", "class OptimModule(nn.Module):\n", " \"\"\" Interface for Module that allows registering buffers/parameters with configurable optimizer hyperparameters \"\"\"\n", "\n", " def register(self, name, tensor, lr=None, wd=0.0):\n", " \"\"\"Register a tensor with a configurable learning rate and 0 weight decay\"\"\"\n", "\n", " if lr == 0.0:\n", " self.register_buffer(name, tensor)\n", " else:\n", " self.register_parameter(name, nn.Parameter(tensor))\n", "\n", " optim = {}\n", " if lr is not None: optim[\"lr\"] = lr\n", " if wd is not None: optim[\"weight_decay\"] = wd\n", " setattr(getattr(self, name), \"_optim\", optim)\n", " \n", "\n", "class Sin(nn.Module):\n", " def __init__(self, dim, w=10, train_freq=True):\n", " super().__init__()\n", " self.freq = nn.Parameter(w * torch.ones(1, dim)) if train_freq else w * torch.ones(1, dim)\n", "\n", " def forward(self, x):\n", " return torch.sin(self.freq * x)\n", " \n", " \n", "class PositionalEmbedding(OptimModule):\n", " def __init__(self, emb_dim: int, seq_len: int, lr_pos_emb: float=1e-5, **kwargs): \n", " \"\"\"Complex exponential positional embeddings for Hyena filters.\"\"\" \n", " super().__init__()\n", " \n", " self.seq_len = seq_len\n", " # The time embedding fed to the filteres is normalized so that t_f = 1\n", " t = torch.linspace(0, 1, self.seq_len)[None, :, None] # 1, L, 1\n", " \n", " if emb_dim > 1:\n", " bands = (emb_dim - 1) // 2 \n", " # To compute the right embeddings we use the \"proper\" linspace \n", " t_rescaled = torch.linspace(0, seq_len - 1, seq_len)[None, :, None]\n", " w = 2 * math.pi * t_rescaled / seq_len # 1, L, 1 \n", " \n", " f = torch.linspace(1e-4, bands - 1, bands)[None, None] \n", " z = torch.exp(-1j * f * w)\n", " z = torch.cat([t, z.real, z.imag], dim=-1)\n", " self.register(\"z\", z, lr=lr_pos_emb) \n", " self.register(\"t\", t, lr=0.0)\n", " \n", " def forward(self, L):\n", " return self.z[:, :L], self.t[:, :L]\n", " \n", "\n", "class ExponentialModulation(OptimModule):\n", " def __init__(\n", " self,\n", " d_model,\n", " fast_decay_pct=0.3,\n", " slow_decay_pct=1.5,\n", " target=1e-2,\n", " modulation_lr=0.0,\n", " modulate: bool=True,\n", " shift: float = 0.0,\n", " **kwargs\n", " ):\n", " super().__init__()\n", " self.modulate = modulate\n", " self.shift = shift\n", " max_decay = math.log(target) / fast_decay_pct\n", " min_decay = math.log(target) / slow_decay_pct\n", " deltas = torch.linspace(min_decay, max_decay, d_model)[None, None]\n", " self.register(\"deltas\", deltas, lr=modulation_lr)\n", " \n", " def forward(self, t, x):\n", " if self.modulate:\n", " decay = torch.exp(-t * self.deltas.abs()) \n", " x = x * (decay + self.shift)\n", " return x \n", "\n", "\n", "class HyenaFilter(OptimModule):\n", " def __init__(\n", " self, \n", " d_model,\n", " emb_dim=3, # dim of input to MLP, augments with positional encoding\n", " order=16, # width of the implicit MLP \n", " fused_fft_conv=False,\n", " seq_len=1024, \n", " lr=1e-3, \n", " lr_pos_emb=1e-5,\n", " dropout=0.0, \n", " w=1, # frequency of periodic activations \n", " wd=0, # weight decay of kernel parameters \n", " bias=True,\n", " num_inner_mlps=2,\n", " normalized=False,\n", " **kwargs\n", " ):\n", " \"\"\"\n", " Implicit long filter with modulation.\n", " \n", " Args:\n", " d_model: number of channels in the input\n", " emb_dim: dimension of the positional encoding (`emb_dim` - 1) // 2 is the number of bands\n", " order: width of the FFN\n", " num_inner_mlps: number of inner linear layers inside filter MLP\n", " \"\"\"\n", " super().__init__()\n", " self.d_model = d_model\n", " self.use_bias = bias\n", " self.fused_fft_conv = fused_fft_conv\n", " self.bias = nn.Parameter(torch.randn(self.d_model))\n", " self.dropout = nn.Dropout(dropout)\n", " \n", " act = Sin(dim=order, w=w)\n", " self.emb_dim = emb_dim\n", " assert emb_dim % 2 != 0 and emb_dim >= 3, \"emb_dim must be odd and greater or equal to 3 (time, sine and cosine)\"\n", " self.seq_len = seq_len\n", " \n", " self.pos_emb = PositionalEmbedding(emb_dim, seq_len, lr_pos_emb)\n", " \n", " self.implicit_filter = nn.Sequential(\n", " nn.Linear(emb_dim, order),\n", " act,\n", " )\n", " for i in range(num_inner_mlps):\n", " self.implicit_filter.append(nn.Linear(order, order))\n", " self.implicit_filter.append(act)\n", "\n", " self.implicit_filter.append(nn.Linear(order, d_model, bias=False))\n", " \n", " self.modulation = ExponentialModulation(d_model, **kwargs)\n", " \n", " self.normalized = normalized\n", " for c in self.implicit_filter.children():\n", " for name, v in c.state_dict().items(): \n", " optim = {\"weight_decay\": wd, \"lr\": lr}\n", " setattr(getattr(c, name), \"_optim\", optim)\n", "\n", " def filter(self, L, *args, **kwargs):\n", " z, t = self.pos_emb(L)\n", " h = self.implicit_filter(z)\n", " h = self.modulation(t, h)\n", " return h\n", "\n", " def forward(self, x, L, k=None, bias=None, *args, **kwargs):\n", " if k is None: k = self.filter(L)\n", " \n", " # Ensure compatibility with filters that return a tuple \n", " k = k[0] if type(k) is tuple else k \n", "\n", " y = fftconv(x, k, bias)\n", " return y\n", " \n", " \n", "class HyenaOperator(nn.Module):\n", " def __init__(\n", " self,\n", " d_model,\n", " l_max,\n", " order=2, \n", " filter_order=64,\n", " dropout=0.0, \n", " filter_dropout=0.0, \n", " **filter_args,\n", " ):\n", " r\"\"\"\n", " Hyena operator described in the paper https://arxiv.org/pdf/2302.10866.pdf\n", " \n", " Args:\n", " d_model (int): Dimension of the input and output embeddings (width of the layer)\n", " l_max: (int): Maximum input sequence length. Defaults to None\n", " order: (int): Depth of the Hyena recurrence. Defaults to 2\n", " dropout: (float): Dropout probability. Defaults to 0.0\n", " filter_dropout: (float): Dropout probability for the filter. Defaults to 0.0\n", " \"\"\"\n", " super().__init__()\n", " self.d_model = d_model\n", " self.l_max = l_max\n", " self.order = order\n", " inner_width = d_model * (order + 1)\n", " self.dropout = nn.Dropout(dropout)\n", " self.in_proj = nn.Linear(d_model, inner_width)\n", " self.out_proj = nn.Linear(d_model, d_model)\n", " \n", " self.short_filter = nn.Conv1d(\n", " inner_width, \n", " inner_width, \n", " 3,\n", " padding=2,\n", " groups=inner_width\n", " )\n", " self.filter_fn = HyenaFilter(\n", " d_model * (order - 1), \n", " order=filter_order, \n", " seq_len=l_max,\n", " channels=1, \n", " dropout=filter_dropout, \n", " **filter_args\n", " ) \n", "\n", " def forward(self, u, *args, **kwargs):\n", " l = u.size(-2)\n", " l_filter = min(l, self.l_max)\n", " u = self.in_proj(u)\n", " u = rearrange(u, 'b l d -> b d l')\n", " \n", " uc = self.short_filter(u)[...,:l_filter] \n", " *x, v = uc.split(self.d_model, dim=1)\n", " \n", " k = self.filter_fn.filter(l_filter)[0]\n", " k = rearrange(k, 'l (o d) -> o d l', o=self.order - 1)\n", " bias = rearrange(self.filter_fn.bias, '(o d) -> o d', o=self.order - 1)\n", " \n", " for o, x_i in enumerate(reversed(x[1:])):\n", " v = self.dropout(v * x_i)\n", " v = self.filter_fn(v, l_filter, k=k[o], bias=bias[o])\n", "\n", " y = rearrange(v * x[0], 'b d l -> b l d')\n", "\n", " y = self.out_proj(y)\n", " return y\n", "\n", " \n", " \n", "if __name__ == \"__main__\":\n", " layer = HyenaOperator(\n", " \n", " d_model=128, \n", " l_max=1024, \n", " order=2, \n", " filter_order=64\n", " )\n", " x = torch.randn(1, 1024, 128, requires_grad=True)\n", " y = layer(x)\n", " \n", " print(x.shape, y.shape)\n", " \n", " grad = torch.autograd.grad(y[:, 10, :].sum(), x)[0]\n", " print('Causality check: gradients should not flow \"from future to past\"')\n", " print(grad[0, 11, :].sum(), grad[0, 9, :].sum())\n" ] }, { "cell_type": "code", "execution_count": 2, "id": "032ef08a-8cc6-491a-9eb8-4a6b3f2d165e", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "torch.Size([1, 1023, 1]) torch.Size([1, 1])\n" ] } ], "source": [ "class HyenaOperatorAutoregressive1D(nn.Module):\n", " def __init__(\n", " self,\n", " d_model,\n", " l_max,\n", " order=2, \n", " filter_order=64,\n", " dropout=0.0, \n", " filter_dropout=0.0, \n", " **filter_args,\n", " ):\n", " super().__init__()\n", "\n", " self.l_max = l_max\n", " self.d_model = d_model\n", " self.l_max = l_max\n", " self.order = order\n", " inner_width = d_model * (order + 1)\n", "\n", " self.dropout = nn.Dropout(dropout)\n", " self.in_proj = nn.Linear(d_model, inner_width)\n", " self.out_proj = nn.Linear(d_model, d_model)\n", " self.fc_before = nn.Linear(1, d_model) # Fully connected layer before the main layer\n", " self.fc_after = nn.Linear(d_model, 1) # Fully connected layer after the main layer\n", "\n", " self.operator = HyenaOperator(\n", " d_model=d_model,\n", " l_max=l_max,\n", " order=order, \n", " filter_order=filter_order,\n", " dropout=dropout, \n", " filter_dropout=filter_dropout, \n", " **filter_args,\n", " )\n", "\n", " def forward(self, u, *args, **kwargs):\n", " # Increase the channel dimension from 1 to d_model\n", " u = self.fc_before(u) \n", " # Pass through the operator\n", " #[B,1024,128] --> [B,1024,128]\n", " u = self.operator(u)\n", " \n", " last_state = u[:,-1,:]\n", " # Decrease the channel dimension back to 1\n", " y = self.fc_after(last_state)\n", " return y,last_state\n", "\n", "\n", "if __name__ == \"__main__\":\n", " layer = HyenaOperatorAutoregressive1D(\n", " d_model=128, \n", " l_max=1024, \n", " order=2, \n", " filter_order=64\n", " )\n", "\n", " x = torch.randn(1, 1023, 1, requires_grad=True) # 1D time series input\n", " y, last_state = layer(x)\n", "\n", " #import pdb;pdb.set_trace()\n", " print(x.shape, y.shape) # should now be [1, 1024, 1]\n", "\n", " #grad = torch.autograd.grad(y[:, 10, 0].sum(), x)[0]\n", " #print('Causality check: gradients should not flow \"from future to past\"')\n", " #print(grad[0, 11, 0].sum(), grad[0, 9, 0].sum())" ] }, { "cell_type": "code", "execution_count": 3, "id": "80cde67b-992f-4cb0-8824-4a6b7e4984ca", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Train Epoch: 1 [0/640 (0%)]\tLoss: 0.446847\n", "Train Epoch: 2 [0/640 (0%)]\tLoss: 0.077979\n", "Train Epoch: 3 [0/640 (0%)]\tLoss: 0.021656\n", "Train Epoch: 4 [0/640 (0%)]\tLoss: 0.007355\n", "Train Epoch: 5 [0/640 (0%)]\tLoss: 0.004926\n", "Train Epoch: 6 [0/640 (0%)]\tLoss: 0.006014\n", "Train Epoch: 7 [0/640 (0%)]\tLoss: 0.003400\n", "Train Epoch: 8 [0/640 (0%)]\tLoss: 0.003720\n", "Train Epoch: 9 [0/640 (0%)]\tLoss: 0.004267\n", "Train Epoch: 10 [0/640 (0%)]\tLoss: 0.004081\n" ] } ], "source": [ "import torch\n", "import torch.optim as optim\n", "import torch.nn.functional as F\n", "from torch.utils.data import DataLoader, Dataset\n", "import numpy as np\n", "\n", "def generate_sine_with_noise(n_points, frequency, phase, amplitude, noise_sd):\n", " # Generate an array of points from 0 to 2*pi\n", " x = np.linspace(0, 2*np.pi, n_points)\n", " \n", " # Generate the sine wave\n", " sine_wave = amplitude * np.sin(frequency * x + phase)\n", " \n", " # Generate Gaussian noise\n", " noise = np.random.normal(scale=noise_sd, size=n_points)\n", " \n", " # Add the noise to the sine wave\n", " sine_wave_noise = sine_wave + noise\n", " \n", " # Stack the sine wave and the noisy sine wave into a 2D array\n", " output = np.column_stack((sine_wave, sine_wave_noise))\n", " \n", " return output\n", " \n", " \n", "class SineDataset(Dataset):\n", " def __init__(self, n_samples, n_points, frequency_range, phase_range, amplitude_range, noise_sd_range):\n", " self.n_samples = n_samples\n", " self.n_points = n_points\n", " self.frequency_range = frequency_range\n", " self.phase_range = phase_range\n", " self.amplitude_range = amplitude_range\n", " self.noise_sd_range = noise_sd_range\n", "\n", " def __len__(self):\n", " return self.n_samples\n", "\n", " def __getitem__(self, idx):\n", " # Generate random attributes\n", " frequency = np.random.uniform(*self.frequency_range)\n", " phase = np.random.uniform(*self.phase_range)\n", " amplitude = np.random.uniform(*self.amplitude_range)\n", " noise_sd = np.random.uniform(*self.noise_sd_range)\n", "\n", " # Generate sine wave with the random attributes\n", " sine_wave = generate_sine_with_noise(self.n_points, frequency, phase, amplitude, noise_sd)\n", "\n", " # Return the sine wave and the parameters\n", " return torch.Tensor(sine_wave[:-1, 1, None]), torch.Tensor(sine_wave[-1:, 0]), torch.Tensor([frequency, phase, amplitude, noise_sd])\n", "\n", "\n", "\n", "# Usage:\n", "dataset = SineDataset(640, 1025, (1, 3), (0, 2*np.pi), (0.5, 1.5), (0.05, 0.15))\n", "\n", "def train(model, device, train_loader, optimizer, epoch):\n", " model.train()\n", " for batch_idx, (data, target, params) in enumerate(train_loader):\n", " #data = data[...,None]\n", " data, target = data.to(device), target.to(device)\n", " optimizer.zero_grad()\n", " output,last_state = model(data)\n", " #import pdb;pdb.set_trace()\n", "\n", " loss = F.mse_loss(output, target)\n", " loss.backward()\n", " optimizer.step()\n", " if batch_idx % 10 == 0:\n", " print('Train Epoch: {} [{}/{} ({:.0f}%)]\\tLoss: {:.6f}'.format(\n", " epoch, batch_idx * len(data), len(train_loader.dataset),\n", " 100. * batch_idx / len(train_loader), loss.item()))\n", "\n", "if __name__ == \"__main__\":\n", " device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n", "\n", " model = HyenaOperatorAutoregressive1D(\n", " d_model=128, \n", " l_max=1024, \n", " order=2, \n", " filter_order=64\n", " ).to(device)\n", "\n", " optimizer = optim.Adam(model.parameters())\n", "\n", " # Assume 10000 samples in the dataset\n", " #dataset = SineDataset(10000, 1025, 2, 0, 1, 0.1)\n", " train_loader = DataLoader(dataset, batch_size=64, shuffle=True)\n", "\n", " for epoch in range(1, 11): # Train for 10 epochs\n", " train(model, device, train_loader, optimizer, epoch)\n" ] }, { "cell_type": "code", "execution_count": null, "id": "cc9f9031-5ee1-49f8-a70f-ad85ca015596", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 4, "id": "90330622-8b44-4b45-8158-6840538f768c", "metadata": {}, "outputs": [], "source": [ "from sklearn.linear_model import LinearRegression\n", "from sklearn.metrics import r2_score\n", "\n", "def fit_and_evaluate_linear_regression(outputs_and_params):\n", " # Split the data into inputs (last_states) and targets (params)\n", " inputs = np.concatenate([x[0] for x in outputs_and_params])\n", " targets = np.concatenate([x[1] for x in outputs_and_params])\n", " \n", " r2_scores = []\n", " param_names = [\"frequency\", \"phase\", \"amplitude\", \"noise_sd\"]\n", " \n", " # Fit the linear regression model for each parameter and calculate the R^2 score\n", " for i in range(targets.shape[1]):\n", " model = LinearRegression().fit(inputs, targets[:, i])\n", " pred = model.predict(inputs)\n", " score = r2_score(targets[:, i], pred)\n", " r2_scores.append(score)\n", " print(f\"R^2 score for {param_names[i]}: {score:.2f}\")\n", " \n", " return r2_scores" ] }, { "cell_type": "code", "execution_count": 5, "id": "5eb62a22-cad8-43c4-b757-f36b6a01e9be", "metadata": {}, "outputs": [], "source": [ "def generate_outputs(model, device, data_loader):\n", " model.eval()\n", " outputs_and_params = []\n", " with torch.no_grad():\n", " for data, target, params in data_loader:\n", " data, target = data.to(device), target.to(device)\n", " output, last_state = model(data)\n", " outputs_and_params.append((last_state.cpu().numpy(), params.cpu().numpy()))\n", " return outputs_and_params\n" ] }, { "cell_type": "code", "execution_count": 6, "id": "a95ee542-1c39-4f04-9184-e26c6983a018", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "R^2 score for frequency: 0.77\n", "R^2 score for phase: 0.66\n", "R^2 score for amplitude: 0.99\n", "R^2 score for noise_sd: 0.97\n" ] } ], "source": [ "outputs_and_params = generate_outputs(model, device, train_loader)\n", "\n", "# Fit the linear regression model and print the R^2 score for each parameter\n", "r2_scores = fit_and_evaluate_linear_regression(outputs_and_params)" ] }, { "cell_type": "markdown", "id": "de65d0d2-b0c6-4ac5-a87f-70fa7f90480b", "metadata": {}, "source": [ "# Autoregressive" ] }, { "cell_type": "code", "execution_count": 120, "id": "8ee139a6-aee5-4309-8685-bf0c28893279", "metadata": {}, "outputs": [], "source": [ "def predict_autoregressive(model, initial_data, n_steps):\n", " model.eval()\n", " predictions = []\n", " current_input = initial_data\n", " with torch.no_grad():\n", " for _ in range(n_steps):\n", " # Get the prediction for the next step and save it\n", " next_output, last_state = model(current_input)\n", " predictions.append(next_output)\n", " #import pdb;pdb.set_trace()\n", "\n", " # Prepare the input for the next step\n", " next_input = torch.cat((current_input[:, 1:, :], next_output[:, None,:]), dim=1)\n", "\n", " current_input = next_input\n", "\n", " return torch.cat(predictions, dim=1)" ] }, { "cell_type": "code", "execution_count": null, "id": "272040d6-4dfb-438b-a432-744e950effd1", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 167, "id": "9f49cbf8-08e8-428a-af80-bcb2515a6327", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "torch.Size([1, 1024, 1])" ] }, "execution_count": 167, "metadata": {}, "output_type": "execute_result" } ], "source": [ "initial_data = dataset[0]\n", "initial_data[0][None,...].shape" ] }, { "cell_type": "code", "execution_count": 168, "id": "c75464e3-e44d-4325-8804-29d8081e3a45", "metadata": {}, "outputs": [], "source": [ "autoregressive_out = predict_autoregressive(model,initial_data[0][None,...],500)" ] }, { "cell_type": "code", "execution_count": 169, "id": "90a37c56-59f3-49bc-bfbe-d9e126c42ed1", "metadata": {}, "outputs": [], "source": [ "total = np.concatenate([initial_data[0].squeeze().numpy() ,autoregressive_out.squeeze().numpy()])" ] }, { "cell_type": "code", "execution_count": 170, "id": "3d2ad0ee-7a0e-4b6f-8b40-dd565c8af84d", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "[]" ] }, "execution_count": 170, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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", "text/plain": [ "
" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "import matplotlib.pyplot as plt\n", "\n", "plt.plot(total)" ] }, { "cell_type": "code", "execution_count": null, "id": "eb368900-34b3-4c36-a916-62ed3a97173b", "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.10" } }, "nbformat": 4, "nbformat_minor": 5 }