DeepLearning API with pytorch and flask

基于flask打造的深度学习API，帮助你快速发布部署神经网络， 实现训练或测试模型， 框架目前支持多种机器学习与深度学习模型的构建。

1. 模型训练

可用模型

模型	接口
随机森林	http://127.0.0.1:5000/model/random_forest
XGBoost	http://127.0.0.1:5000/model/xgboost
K-Means	http://127.0.0.1:5000/model/k_means
KNN	http://127.0.0.1:5000/model/knn

通用可选参数

必要参数

POST参数	具体值
dataset	数据集文件名（str）
parameter	模型的入参（dict）

可选参数

模型的入参（dict）	具体值
predict	预测的属性（表格字段名）
learning_rate	学习率（1 - 0.01）
test_data_ratio	测试集的比例（1 - 0.01）
batch_size	batch大小（512 - 16）

训练范例

Endpoint:

Method: POST
Type: RAW
URL: http://127.0.0.1:5000/model/xgboost

Body:

{
    "dataset": "diabetes_binary_health_indicators_BRFSS2015.csv",
    "parameter": {
        "predict": "Diabetes_binary",
        "learning_rate": 0.1
    }
}

2. 预测

Endpoint:

通用预测接口

Method: GET
Type: RAW
URL: http://127.0.0.1:5000/model/predict

使用已保存模型

{
    "model": "D:/SourceCode/PythonCode/DeepLearningAPI/saves/test.model",
    "data": {
        "HighBP": 1,
        "HighChol": 1,
        "CholCheck": 1,
        "BMI": 30,
        "Smoker": 1,
        "Stroke": 1,
        "HeartDiseaseorAttack": 1,
        "PhysActivity": 1,
        "Fruits": 0,
        "Veggies": 0,
        "HvyAlcoholConsump": 0,
        "AnyHealthcare": 1,
        "NoDocbcCost": 0,
        "GenHlth": 4,
        "MentHlth": 0,
        "PhysHlth": 20,
        "DiffWalk": 1,
        "Sex": 1,
        "Age": 68,
        "Education": 3,
        "Income": 1
    }
}

测试样例

{
    "data": {
        "HighBP": 1,
        "HighChol": 1,
        "CholCheck": 1,
        "BMI": 30,
        "Smoker": 1,
        "Stroke": 1,
        "HeartDiseaseorAttack": 1,
        "PhysActivity": 1,
        "Fruits": 0,
        "Veggies": 0,
        "HvyAlcoholConsump": 0,
        "AnyHealthcare": 1,
        "NoDocbcCost": 0,
        "GenHlth": 4,
        "MentHlth": 0,
        "PhysHlth": 20,
        "DiffWalk": 1,
        "Sex": 1,
        "Age": 7,
        "Education": 3,
        "Income": 1
    }
}

返回结果

{
    "data": {
        "0": {
            "precision": 0.0,
            "recall": 0.0,
            "f1-score": 0.0,
            "support": 0
        },
        "1": {
            "precision": 1.0,
            "recall": 0.6443878633566731,
            "f1-score": 0.783741935483871,
            "support": 14139
        },
        "accuracy": 0.6443878633566731,
        "macro avg": {
            "precision": 0.5,
            "recall": 0.32219393167833654,
            "f1-score": 0.3918709677419355,
            "support": 14139
        },
        "weighted avg": {
            "precision": 1.0,
            "recall": 0.6443878633566731,
            "f1-score": 0.783741935483871,
            "support": 14139
        }
    }
}

3. 上传数据集

Endpoint:

Method: POST
Type: RAW
URL: http://127.0.0.1:5000/dataset

Body:

{
    "name": "test.csv",
    "data": [
        {
            "address": "华山路31号",
            "addressExtend": "屯溪老街"
        },
        {
            "address": "金城镇 珠山82号",
            "addressExtend": "101"
        }
    ]
}

BRFS2015数据集预测效果

WideDeep网络

              precision    recall  f1-score   support
	       0       0.88      0.97      0.93     43552
	       1       0.56      0.21      0.31      7184
	
	accuracy                           0.86     50736
   macro avg       0.72      0.59      0.62     50736
weighted avg       0.84      0.86      0.84     50736

DeepFM网络

              precision    recall  f1-score   support

           0       0.88      0.97      0.92     43552
           1       0.53      0.24      0.33      7184

    accuracy                           0.86     50736
   macro avg       0.71      0.60      0.63     50736
weighted avg       0.84      0.86      0.84     50736

DeepFM在”Census Income”数据集上表现

              precision    recall  f1-score   support
   
       <=50K       0.92      0.82      0.87      4942
        >50K       0.58      0.79      0.67      1571
    accuracy                           0.81      6513
   macro avg       0.75      0.80      0.77      6513
weighted avg       0.84      0.81      0.82      6513