MachineLearningHoods

Uses AWS Machine Learning to predict which neighborhood a string of text most likely originates from.

How

• Using a dataset of ~1G of geo-tagged tweets, we create a CSV with two columns: text and neighborhood.
• After training and evaluating a model with this data, we expose the real-time endpoint via this elixir application.

Takeaways

• Molding the training data to create a better model is the real challenge here.
• Does my data even have statistical correlations or is it just noise?
• Iterate, iterate, iterate on the model and evaluation data is what it seems people do.

Input Schema

{
  "version": "1.0",
    "rowId": null,
    "rowWeight": null,
    "targetAttributeName": "Neighborhood",
    "dataFormat": "CSV",
    "dataFileContainsHeader": true,
    "attributes": [
    {
      "attributeName": "Text",
      "attributeType": "TEXT"
    },
    {
      "attributeName": "Neighborhood",
      "attributeType": "CATEGORICAL"
    }
    ],
    "excludedAttributeNames": []
}