ExampleUse.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generate Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from scipy import signal\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fs = 48e3\n",
    "f0 = 20\n",
    "f1 = 20e3\n",
    "t = np.linspace(0, 60, 60*fs)\n",
    "\n",
    "train_input = signal.chirp(t=t, f0=f0, t1=60, f1=f1, method='logarithmic') + np.random.normal(scale=5e-2, size=len(t))\n",
    "\n",
    "fc = 2e3\n",
    "sos = signal.butter(N=2, Wn=fc/fs, output='sos')\n",
    "train_target = signal.sosfilt(sos, train_input)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Declare Model\n",
    "Example will only be shown for this model, the other models are analogous."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "model_type = \"SVF\" #or \"TDF-II\" or \"LSS\" or \"DOPOZ\"\n",
    "\n",
    "device = torch.device('cpu')\n",
    "\n",
    "if model_type == \"SVF\":\n",
    "    from dsvf_models import DSVF\n",
    "    model = DSVF().to(device)\n",
    "elif model_type == 'TDF-II':\n",
    "    from differentiable_tdf2_model import DTDFII\n",
    "    model = DTDFII().to(device)\n",
    "elif model_type == 'LSS':\n",
    "    from linear_state_space_model import LinearStateSpaceModel\n",
    "    model = LinearStateSpaceModel(num_states=2).to(device)\n",
    "elif model_type == \"DOPOZ\":\n",
    "    from differentiable_one_pole_one_zero import DOnePole\n",
    "    model = DOnePole().to(device)\n",
    "else:\n",
    "    print(\"Please pick valid model type\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Initialize Dataloader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from iir_dataset import DIIRDataSet\n",
    "from torch.utils.data import DataLoader\n",
    "\n",
    "batch_size = 1024\n",
    "sequence_length = 512\n",
    "\n",
    "loader = DataLoader(dataset=DIIRDataSet(train_input, train_target, sequence_length), batch_size=batch_size, shuffle = True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Define optimizer and criterion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "from torch.optim import Adam\n",
    "\n",
    "n_epochs = 1000\n",
    "lr = 1e-3\n",
    "\n",
    "optimizer = Adam(model.parameters(), lr=lr, betas=(0.9, 0.999), eps=1e-08, weight_decay=0, amsgrad=False)\n",
    "\n",
    "criterion = nn.MSELoss()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Define train loop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(criterion, model, loader, optimizer):\n",
    "    model.train()\n",
    "    device = next(model.parameters()).device\n",
    "    total_loss = 0\n",
    "    for batch in loader:\n",
    "        input_seq_batch = batch['input'].to(device)\n",
    "        target_seq_batch = batch['target'].to(device)\n",
    "        optimizer.zero_grad()\n",
    "        predicted_output = model(input_seq_batch)\n",
    "        loss = criterion(target_seq_batch, predicted_output)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        total_loss += loss.item()\n",
    "\n",
    "    total_loss /= len(loader)\n",
    "    return total_loss"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for epoch in range(n_epochs):\n",
    "    loss = train(criterion, model, loader, optimizer)\n",
    "    print(\"Epoch {} -- Loss {:3E}\".format(epoch, loss))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}