From e2106af5dba4bdbe28f74d609fe409811ce1a89d Mon Sep 17 00:00:00 2001
From: Scott <condronscott@gmail.com>
Date: Wed, 1 Feb 2023 13:52:08 +0000
Subject: [PATCH] Add W&B embedding projector example

---
 examples/Visualizing_embeddings_in_W&B.ipynb | 108 +++++++++++++++++++
 1 file changed, 108 insertions(+)
 create mode 100644 examples/Visualizing_embeddings_in_W&B.ipynb

diff --git a/examples/Visualizing_embeddings_in_W&B.ipynb b/examples/Visualizing_embeddings_in_W&B.ipynb
new file mode 100644
index 0000000..3b11325
--- /dev/null
+++ b/examples/Visualizing_embeddings_in_W&B.ipynb
@@ -0,0 +1,108 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Visualizing the embeddings in W&B\n",
+    "\n",
+    "We will upload the data to [Weights & Biases](http://wandb.ai) and use an [Embedding Projector](https://docs.wandb.ai/ref/app/features/panels/weave/embedding-projector) to visualize the embeddings using common dimension reduction algorithms like PCA, UMAP, and t-SNE. The dataset is created in the [Obtain_dataset Notebook](Obtain_dataset.ipynb)."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 1. Log the data to W&B\n",
+    "\n",
+    "We create a [W&B Table](https://docs.wandb.ai/guides/data-vis/log-tables) with the original data and the embeddings. Each review is a new row and the 1536 embedding floats are given their own column named `emb_{i}`."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "from sklearn.manifold import TSNE\n",
+    "import numpy as np\n",
+    "\n",
+    "# Load the embeddings\n",
+    "datafile_path = \"data/fine_food_reviews_with_embeddings_1k.csv\"\n",
+    "df = pd.read_csv(datafile_path)\n",
+    "\n",
+    "# Convert to a list of lists of floats\n",
+    "matrix = np.array(df.embedding.apply(eval).to_list())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import wandb\n",
+    "\n",
+    "original_cols = df.columns[1:-1].tolist()\n",
+    "embedding_cols = ['emb_'+str(idx) for idx in range(len(matrix[0]))]\n",
+    "table_cols = original_cols + embedding_cols\n",
+    "\n",
+    "with wandb.init(project='openai_embeddings'):\n",
+    "    table = wandb.Table(columns=table_cols)\n",
+    "    for i, row in enumerate(df.to_dict(orient=\"records\")):\n",
+    "        original_data = [row[col_name] for col_name in original_cols]\n",
+    "        embedding_data = matrix[i].tolist()\n",
+    "        table.add_data(*(original_data + embedding_data))\n",
+    "    wandb.log({'openai_embedding_table': table})"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### 2. Render as 2D Projection"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "After navigating to the W&B run link, we click the ⚙️ icon in the top right of the Table and change \"Render As:\" to \"Combined 2D Projection\". "
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Example: http://wandb.me/openai_embeddings"
+   ]
+  }
+ ],
+ "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": "3.9.15"
+  },
+  "vscode": {
+   "interpreter": {
+    "hash": "365536dcbde60510dc9073d6b991cd35db2d9bac356a11f5b64279a5e6708b97"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 4
+}