Updating content for Graph Embeddings

Author: rajiv.sambasivan@gmail.com

notebooks/.ipynb_checkpoints/Graph_Embeddings-checkpoint.ipynb

@@ -78,7 +78,10 @@
  "2. The learning is _transductive_. The embeddings that can be determined are limited to the nodes seen during training. We will not be able to determine an embedding for nodes that were not seen as part of the training process.\n",
  "3. Inability to incorporate node features in determining the embedding. \n",
  "\n",
- "Instead of just using a simple _encoder_ and _similarity_ measure to determine a Euclidean representation, it is possible to use more sophisticated approach to determining a representation for the graph. In contrast, we can use a _deeper_ approach using _Graph Neural Networks (GNN)_. In a _GNN_ nodes aggregate information from their neighbors using a _neural network_ . This approach eliminates some of the draw backs observed with the shallow approach. A schematic of the computational approach for _GNN's_ is shown below. ![](./img/gnn_schematic.png)\n"
+ "Instead of just using a simple _encoder_ and _similarity_ measure to determine a Euclidean representation, it is possible to use more sophisticated approach to determining a representation for the graph. In contrast, we can use a _deeper_ approach using _Graph Neural Networks (GNN)_. In a _GNN_ nodes aggregate information from their neighbors using a _neural network_ . This approach eliminates some of the draw backs observed with the shallow approach. A schematic of the computational approach for _GNN's_ is shown below. ![](./img/gnn_schematic.png)\n",
+ "\n",
+ "\n",
+ "This notebook illustrates the shallow approach to embedding using _Node2vec_. An illustration of the deeper approach will follow shortly."
  ]
  },
  {

notebooks/.ipynb_checkpoints/Graph_Retail_EDA_II-checkpoint.ipynb

@@ -1132,32 +1132,6 @@
  "ap.log_run(run_info)"
  ]
  },
- {
- "cell_type": "code",
- "execution_count": null,
- "metadata": {
- "colab": {
- "base_uri": "https://localhost:8080/"
- },
- "id": "TReWeBuAHZnU",
- "outputId": "fd031b0c-a863-4e11-d2e3-0e603685a841"
- },
- "outputs": [],
- "source": [
- "conn_params"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": null,
- "metadata": {
- "id": "hupHf5NOHczI"
- },
- "outputs": [],
- "source": [
- "! mkdir -p data/retail_freq_cust_data_dump"
- ]
- },
  {
  "cell_type": "code",
  "execution_count": null,
@@ -1171,7 +1145,7 @@
  "outputs": [],
  "source": [
  "\n",
- "!./tools/arangodump -c none --server.endpoint http+ssl://{conn_params[\"DB_service_host\"]}:{conn_params[\"DB_service_port\"]} --server.username {conn_params[\"username\"]} --server.database {conn_params[\"dbName\"]} --server.password {conn_params[\"password\"]} --output-directory \"data/retail_freq_cust_data_dump\""
+ "#!./tools/arangodump -c none --server.endpoint http+ssl://{conn_params[\"DB_service_host\"]}:{conn_params[\"DB_service_port\"]} --server.username {conn_params[\"username\"]} --server.database {conn_params[\"dbName\"]} --server.password {conn_params[\"password\"]} --output-directory \"data/retail_freq_cust_data_dump\""
  ]
  },
  {

notebooks/Graph_Embeddings.ipynb

@@ -78,7 +78,10 @@
  "2. The learning is _transductive_. The embeddings that can be determined are limited to the nodes seen during training. We will not be able to determine an embedding for nodes that were not seen as part of the training process.\n",
  "3. Inability to incorporate node features in determining the embedding. \n",
  "\n",
- "Instead of just using a simple _encoder_ and _similarity_ measure to determine a Euclidean representation, it is possible to use more sophisticated approach to determining a representation for the graph. In contrast, we can use a _deeper_ approach using _Graph Neural Networks (GNN)_. In a _GNN_ nodes aggregate information from their neighbors using a _neural network_ . This approach eliminates some of the draw backs observed with the shallow approach. A schematic of the computational approach for _GNN's_ is shown below. ![](./img/gnn_schematic.png)\n"
+ "Instead of just using a simple _encoder_ and _similarity_ measure to determine a Euclidean representation, it is possible to use more sophisticated approach to determining a representation for the graph. In contrast, we can use a _deeper_ approach using _Graph Neural Networks (GNN)_. In a _GNN_ nodes aggregate information from their neighbors using a _neural network_ . This approach eliminates some of the draw backs observed with the shallow approach. A schematic of the computational approach for _GNN's_ is shown below. ![](./img/gnn_schematic.png)\n",
+ "\n",
+ "\n",
+ "This notebook illustrates the shallow approach to embedding using _Node2vec_. An illustration of the deeper approach will follow shortly."
  ]
  },
  {

notebooks/Graph_Retail_EDA_II.ipynb

@@ -1132,32 +1132,6 @@
  "ap.log_run(run_info)"
  ]
  },
- {
- "cell_type": "code",
- "execution_count": null,
- "metadata": {
- "colab": {
- "base_uri": "https://localhost:8080/"
- },
- "id": "TReWeBuAHZnU",
- "outputId": "fd031b0c-a863-4e11-d2e3-0e603685a841"
- },
- "outputs": [],
- "source": [
- "conn_params"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": null,
- "metadata": {
- "id": "hupHf5NOHczI"
- },
- "outputs": [],
- "source": [
- "! mkdir -p data/retail_freq_cust_data_dump"
- ]
- },
  {
  "cell_type": "code",
  "execution_count": null,
@@ -1171,7 +1145,7 @@
  "outputs": [],
  "source": [
  "\n",
- "!./tools/arangodump -c none --server.endpoint http+ssl://{conn_params[\"DB_service_host\"]}:{conn_params[\"DB_service_port\"]} --server.username {conn_params[\"username\"]} --server.database {conn_params[\"dbName\"]} --server.password {conn_params[\"password\"]} --output-directory \"data/retail_freq_cust_data_dump\""
+ "#!./tools/arangodump -c none --server.endpoint http+ssl://{conn_params[\"DB_service_host\"]}:{conn_params[\"DB_service_port\"]} --server.username {conn_params[\"username\"]} --server.database {conn_params[\"dbName\"]} --server.password {conn_params[\"password\"]} --output-directory \"data/retail_freq_cust_data_dump\""
  ]
  },
  {