Part 2 - Sanity Check.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch import nn, optim\n",
    "from jcopdl.callback import Callback, set_config\n",
    "\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "device"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Dataset & Dataloader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from src.dataset import VCTKSpeakerDataset, VCTKSpeakerDataloader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "config = torch.load(\"model/configs.pth\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs = 64\n",
    "dataset = VCTKSpeakerDataset(\"vctk_dataset/wav48/\", \"vctk_dataset/txt/\", n_speaker=25, n_each_speaker=100, sr=config.sr, min_dur=config.min_dur)\n",
    "dataloader = VCTKSpeakerDataloader(dataset, batch_size=bs, num_workers=2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Architecture & Config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from src.model import Encoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "weight = torch.load(\"model/weights_best.pth\", map_location=\"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Encoder(config.ndim).to(device)\n",
    "model.load_state_dict(weight)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Embed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tqdm.auto import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "embed = []\n",
    "label = []\n",
    "with torch.no_grad():\n",
    "    model.eval()\n",
    "    for images, labels in tqdm(dataloader):\n",
    "        images = images.to(device)\n",
    "        output = model(images)\n",
    "        embed.extend(output)\n",
    "        label.extend(labels)\n",
    "embed = torch.stack(embed).cpu().numpy()\n",
    "label = torch.stack(label).cpu().numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visualize Embedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from umap import UMAP\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import plotly.express as px"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "umap = UMAP(n_neighbors=3, min_dist=0.01, a=1, b=1)\n",
    "X = umap.fit_transform(embed)\n",
    "\n",
    "df = pd.DataFrame(X, columns=[\"umap1\", \"umap2\"])\n",
    "df[\"text\"] = label.astype(str)\n",
    "\n",
    "fig = px.scatter(df, x=\"umap1\", y=\"umap2\", color=\"text\")\n",
    "fig.update_traces(textposition='top center')\n",
    "fig.update_layout(\n",
    "    height=800,\n",
    "    title_text='Reduced Vec Visualization'\n",
    ")\n",
    "fig.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:speaker_embedding] *",
   "language": "python",
   "name": "conda-env-speaker_embedding-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}