graph book

books/graph_ml_for_engineers/notebooks/1_graphs_intro.ipynb

@@ -0,0 +1,128 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Graphs\n",
+    "\n",
+    "Graphs are a general mechanism for describing and analyzing entities with relations and interactions\n",
+    "\n",
+    "Rather than thinking of the world as a set of isolated data points:\n",
+    "\n",
+    "| Entities    | Feature 1   | Feature 2   |\n",
+    "| ----------- | ----------- | ----------- |\n",
+    "| A           | 1000        | 1001        |\n",
+    "| B           | 2000        | 2002        |\n",
+    "\n",
+    "We think of the these entities and their networks and relations between other entities\n",
+    "\n",
+    "| Entities    | Feature 1   | Feature 2   |     \n",
+    "| ----------- | ----------- | ----------- |    \n",
+    "| A           | 1000        | 1001        |    \n",
+    "| B           | 2000        | 2002        |   \n",
+    "\n",
+    "➕\n",
+    "\n",
+    "| Src Entity    | Dst Entity    | Relation Type | Feature 2     | Feature 2     |   \n",
+    "| ------------- | ------------- | ------------- | ------------- | ------------- | \n",
+    "| A             | B             | Is Child Of   | 1             | 1             |    \n",
+    "| B             | A             | Is Parent Of  | 2             | 2             |\n",
+    "\n",
+    "\n",
+    "\n",
+    "## What is the issue with non-graph related ML Toolbox?\n",
+    "\n",
+    "Designed for tabular data, grids of data, or sequences of data.\n",
+    "- Text / Audio sequences has a notion of left & right\n",
+    "- Images has a notion of up / down & left / right\n",
+    "\n",
+    "Graphs have arbritary size and arbritary topology and has no spatial locality.\n",
+    "\n",
+    "\n",
+    "In traditional ML, we take our nodes, links and entire graph and represent them as vectors; then we train a classical ML model on them i.e. Random forest, SVM, NN, etc.\n",
+    "Thus that when a new node / link / graph appears we can obtain its features to make a prediction. Traditional ML uses hand crafted features which is what we will talk about below.\n",
+    "\n",
+    "Traditional way to do node prediction:\n",
+    "\n",
+    "Given a graph $G$, set of vertices $V$ and Edges $E$, where $G = (V, E)$, we want to learn a function $f : V \\rightarrow \\Reals$\n",
+    "\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### What are some features we can extract for nodes?\n",
+    "\n",
+    "- Node Degree\n",
+    "  - The degree $d_v$ of a node $v$ is the number of edges the nodes has.\n",
+    "  - CONS: We treat all neighboring nodes equally.\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "- Node Centrality\n",
+    "  -  Try to capture the importance of a node, can be modeled by:\n",
+    "     -  Engienvector centrality\n",
+    "        -  \n",
+    "     -  Betweenness centrality\n",
+    "     -  Closeness centrality"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "- Clustering Coefficient"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "- Clustering Coefficient"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## What are we trying to do?\n",
+    "\n",
+    "Given node $u$, we are trying to learn a nueral network $f$, such that we can generate a $d$ dimensional vector representaiton of the node $u$ : $f(u) \\rightarrow \\Reals^d$ ; where similar nodes have their vector representation spatially close to each other."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": []
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3.9.12 ('graph-ml-for-engineers-9GQyHo6a-py3.9')",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python",
+   "version": "3.9.12"
+  },
+  "orig_nbformat": 4,
+  "vscode": {
+   "interpreter": {
+    "hash": "c5835645cbd39e77e80fd28b6a8a6b63c0a1f33699bc9c2aaafa2cbac9764660"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

books/graph_ml_for_engineers/notebooks/2_node_embeddings.ipynb

@@ -0,0 +1,49 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## What are we trying to do?\n",
+    "\n",
+    "Input Graph --> Feature Engineering --> Apply some learning algorithm --> Drive some prediction\n",
+    "Wtih Graph represeatnation learning we want to eliminate the \"Feature Engineering\" step, and automatically learn the features\n",
+    "\n",
+    "\n",
+    "## So whats the idea?\n",
+    "\n",
+    "Learn a $f: u \\rightarrow \\mathbb{R}^d$ ; a function $f$ that given a node $u$ maps that node to a $d$ dimensional vector in the $\\mathbb{R}$ (real) space\n",
+    "\n",
+    "\n",
+    "Given that we learn the function $f$, what can we do?\n",
+    "You can use them for downstream gtasks such as Node classification, Link prediction, Graph classification, Clustering, Anomoly detection, et al.\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    " "
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3.9.12 ('graph-ml-for-engineers-9GQyHo6a-py3.9')",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python",
+   "version": "3.9.12"
+  },
+  "orig_nbformat": 4,
+  "vscode": {
+   "interpreter": {
+    "hash": "c5835645cbd39e77e80fd28b6a8a6b63c0a1f33699bc9c2aaafa2cbac9764660"
+   }
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

books/graph_ml_for_engineers/pyproject.toml

@@ -0,0 +1,19 @@
+[tool.poetry]
+name = "graph-ml-for-engineers"
+version = "0.1.0"
+description = "Learning Graph Machine Learning made easy for Seasoned Engineers"
+authors = ["shubhamvij <reachme@shubhamvij.com>"]
+readme = "README.md"
+packages = [{include = "graph_ml_for_engineers"}]
+
+[tool.poetry.dependencies]
+python = "^3.9"
+matplotlib = "networkx"
+
+[tool.poetry.group.dev.dependencies]
+networkx = "^2.8.6"
+matplotlib = "^3.5.3"
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"