# AB InBev MLOps Challenge

Jimmy Alexander Pulido Arias
> [name=Jimmy Alexander Pulido Arias]

You can read this doc online in https://hackmd.io/@GM7_PYyKTYimjfqKBc0E2w/Sy2g9xuVF

## Part A: Model selection

I selected the following kaggle challenges solution: Iris - https://www.kaggle.com/caesarmario/iris-classification-using-various-ml-models

## Setup:
1. Clone the following git repository: https://github.com/jiapulidoar/abinbev-mlops.git
2. build the image`docker build -t mlops-airflow:latest . `
3. Run the docker container: `docker run -d -p 8080:8080 -p 8008:8008 mlops-airflow`
4. Enter to Airflow panel: http://localhost:8080/admin/
5. Once the pipeline is executed the FastAPI can be accessed through: http://localhost:8008/

## Part B: Orchestrating the model

For this challenges the following setup was used:

* **Orchestrating tool:** Airflow
* **Language**: python
* **DB**: LiteSQL
* **Classifier**: SVM
* **API**: FastAPI

### Airflow Pipeline:
1. Graph view:

2. Tree view

* **preprocess_data**: Split data into training and test and save in Data base
* **train**: train the model and store it
* **evaludate_model**: evaludate the model using accuracy_score and classification_report and sabe in DB
* **Serve**: Deploy and start API

## DB tables:

* **Iris**

| SepalLengthCm | SepalWidthCm | PetalLengthCm | PetalWidthCm |Species |
| -------- | -------- | -------- | -------- |-------- |
| number | number | number | number |Text |

* **iris_split**

| SepalLengthCm | SepalWidthCm | PetalLengthCm | PetalWidthCm |Species |Train |
| -------- | -------- | -------- | -------- |-------- |-------- |
| number | number | number | number |Text |Boolean |

* **model_evaluation**

| model | accuracy | report |
| -------- | -------- | -------- |
| model name | number | classification_report |

* **model_evaluation**

| SepalLengthCm | SepalWidthCm | PetalLengthCm | PetalWidthCm |Species_prediction|
| -------- | -------- | -------- | -------- |-------- |
| number | number | number | number |Text |

## Part C: Exposing the model as an API

The API was implemented using [FastAPI](https://fastapi.tiangolo.com/)

### Fast API
Once the pipeline is executed and the **serve** is running, we can access the API through the url: `http://localhost:8008/docs`

### Endpoint:

http://localhost:8008/evaluate  POST

* Request sample:

{
  "items": [
    {
      "SepalLengthCm": 150,
      "SepalWidthCm":23,
      "PetalLengthCm": 10,
      "PetalWidthCm": 8
    },
    {
      "SepalLengthCm": 0,
      "SepalWidthCm": 0,
      "PetalLengthCm": 0,
      "PetalWidthCm": 0
    }
  ]
}

* Response sample:

{
  "0": {
    "SepalLengthCm": 150,
    "SepalWidthCm": 23,
    "PetalLengthCm": 10,
    "PetalWidthCm": 8,
    "Species_prediction": "Iris-virginica"
  },
  "1": {
    "SepalLengthCm": 0,
    "SepalWidthCm": 0,
    "PetalLengthCm": 0,
    "PetalWidthCm": 0,
    "Species_prediction": "Iris-setosa"
  }
}

### Sample:
* Request

* Response

## References:

> [[IRIS Dataset](https://www.kaggle.com/caesarmario/iris-classification-using-various-ml-models/data)][Iris Classification using Various ML Models]
> [[10 Minutes to Building a Machine Learning Pipeline with Apache Airflow](https://towardsdatascience.com/10-minutes-to-building-a-machine-learning-pipeline-with-apache-airflow-53cd09268977)]
> [Machine Learning Pipelines with Kubeflow](https://towardsdatascience.com/machine-learning-pipelines-with-kubeflow-4c59ad05522)