avalon-api

Avalon API backend application for DS&ML Project
Includes utilization of supervised, unsupervised algorithms and Computer Vision using Neural Networks to solve Machine Learning tasks

Our project is dedicated for analyzing, building infrastructure and providing comprehensive UI for interaction to client, using those datasets:

• Global Development indicators for supervised type of analysis
• UK Road Traffic Collision Dataset for unsupervised type of analysis
• CelebFaces Attributes (CelebA) Dataset for training and validating the CNN MobilenetV2 for Computer Vision features extraction.

Installation

Docker Compose

Simply run in the root directory of the project:

docker compose up --build

That's all, application will be available at localhost:8000

Note: If you have ARM arch, change the image of tf-serving service in docker-compose.yml to emacski/tensorflow-serving

Manual

Clone the repository

git clone https://github.com/withoutsecondD/avalon-api.git

Setup venv and activate it

cd avalon-api
python -m venv .venv

source .venv/bin/activate
or
.\.venv\Scripts\activate (Windows)

Install requirements

pip install -r requirements.txt

Create .env file at the root of the project and fill following fields

# address of TF Serving container
TF_SERVING_BASE_URL="http://localhost:8501"

# set the same model name as in TF Serving container
MODEL_NAME="smile"

# address of frontend app, only needed for CORS 
FRONTEND_ADDR="http://localhost:3000"

# paths to results of supervised/unsupervised learning
# algorithms results 
PATH_TO_SUPERVISED_RESULTS="model/results/supervised.json"
PATH_TO_UNSUPERVISED_RESULTS="model/results/unsupervised.json"

Go to api directory and run the app

fastapi dev api.py

Application will be available at localhost:8000

Usage

API

---

GET /resources/{file_name}

Returns a static file specified by file_name

Example request:

curl localhost:8000/resources/dummy.csv

Example response

• Response body contains the file itself

---

POST /predict

After uploading a photo with a person, endpoint returns confidence score which represents how possible is that the person on uploaded photo is smiling

Example request:

• POST query with image as FormData
  Ensure that you append the file to FormData with key image

Example response:

{
  "data": {
    "confidence": "93.31%"
  },
  "success": true,
  "message": ""
}

---

GET /results?model={model}

Returns a results of specified model as json, results includes various metrics, such as accuracy, f2 score, optimal k and so on, response structure may vary depending on choice of the model

Parameters:

• model - results of which model to return, possible values are supervised, unsupervised

Example request:

curl localhost:8000/results?model=supervised

Example response:

Successfull response

{
    "data": <MODEL_RESULTS>,
    "success": true,
    "message": ""
}

Unsuccessfull response

{
    "data": null,
    "success" false,
    "message" <ERROR_MESSAGE>
}

Note:

MODEL_RESULTS structure depends on choice of the model:

• supervised:

{
  "regression": {
    "linear": {
      "r2": 0.038026047739232904,
      "rmse": 261.6237722008474,
      "mae": 4.1611325886124755,
      "mse": 261.6237722008474
    },
    "dtree_reg": {
      "d2mae": 0.6506618433481275,
      "cv_results": {
        "d2mae_results": [0.167, 0.187, ..., 0.644],
        "best_maxdepth": 20
      }
    },
    "randomforest_reg": {
      "mae": 3.0555189593971828
    },
    "knn": {
      "mae": 4.842358534082274,
      "cv_results": {
        "mae_results": [5.657, 5.190, ..., 4.873],
        "best_k": 19
      }
    }
  },
  "classification": {
    "logistic": {
      "accuracy": 0.9508098380323935,
      "roc_auc": 0.3973070708625067,
      "log_loss": 0.6220748882638015
    },
    "dtree_clf": {
      "accuracy": 0.978069498069498,
      "cv_results": {
        "accuracy_results": [0.0, 0.794, 0.946, ..., 0.978],
        "best_maxdepth": 5
      }
    },
    "randomforest_clf": {
      "accuracy": 0.9994001199760048
    },
    "naiveB": {
      "accuracy": 0.9070185962807439
    },
    "knn": {
      "accuracy": 0.9524838275194425,
      "cv_results": {
        "accuracy_results": [0.908, 0.945, ..., 0.952],
        "best_k": 9
      }
    },
    "svc": {
      "accuracy": 0.9508098380323935,
      "f1_score": 0.0
    }
  }
}

• unsupervised:

{
    "decomposition": {
        "pca": {
            "explained_variance_ratio": [
                0.17691734853507976,
                0.06470805720522751,
                0.0615497440233952,
                0.05049413014730652,
                0.04790211649338024
            ],
            "noise_variance": 0.0012627757554276168,
            "n_components": 5,
            "accuracy": 0.8977539389875964,
            "f1_score": 0.31537419772713887
        },
        "nmf": {
            "explained_variance_ratio": [
                0.17691734853507976,
                0.06470805720522751,
                0.0615497440233952,
                0.05049413014730652,
                0.04790211649338024
            ],
            "noise_variance": 0.0012627757554276168,
            "n_components": 5,
            "accuracy": 0.8977539389875964,
            "f1_score": 0.31537419772713887
        },
        "isomap": {
            "n_components": 5,
            "accuracy": 0.8977539389875964,
            "f1_score": 0.31537419772713887
        },
        "tsne": {
            "n_components": 5,
            "accuracy": 0.8977539389875964,
            "f1_score": 0.31537419772713887
        }
    },
    "clustering": {
        "inertia_results": [
            58016.19292814988,
            51369.142981542005,
            43204.366726449094,
            38669.96911231872
        ],
        "accuracy": 0.8977539389875964,
        "f1_score": 0.31537419772713887
    }
}