Construct models for scikit-learn without programming, only with json files.
git clone https://github.com/saromanov/scikit-json
cd scikit-json
python scikit_json.py --json jsonfiles/example.json
json file example with datasets from sklearn
{
"class1" : {
"dataset": "load_iris",
"method": "neighbors.KNeighborsClassifier",
"predict": [5.8,6.7,2.5,1.6]
}
}
example of json file with link to your dataset
{
"class1" : {
"dataset_file": {
"path": "./toy.txt",
"data": [0,3],
"labels": [4,5],
"split": ","
},
"method": "neighbors.KNeighborsClassifier",
"predict": [7,8,5]
}
}
example of json file with model parameters
{
"class1" : {
"dataset": "load_iris",
"method": {
"name": "svm.SVC",
"params": {
"kernel": "rbf",
"gamma" : 0.001,
"max_iter": 100
}
},
"predict": [5.8,6.7,2.5,1.6]
}
}
example of json file with saving model
{
"class1" : {
"dataset": "load_iris",
"method": "neighbors.KNeighborsClassifier",
"predict": [5.8,6.7,2.5,1.6],
"save": "./model1.pkl"
},
"class2": {
"dataset": "load_iris",
"method": "svm.SVC",
"predict": [5.8,6.7,2.5,1.6],
"save": "./model2.pkl"
}
}
example of json file with loading model
{
"class1": {
"load": "./model1.pkl",
"predict": [5.8,6.7,2.5,1.6]
}
}
class1 - title of experiment. Example: "Experiment1", "experiment2", "foobar", etc
dataset - title of dataset from scikit Example: "load_iris", "load_digits", "load_boston"
dataset_file - user dataset object
- path - path to your dataset
- data - [startindex, endindex] - Get
- labels - [startindex, endindex] - Get labels
- split - spliting string symbol
Example:
"dataset_file": {
"path": "data.csv",
"data": [0,3],
"labels": [4],
"split": ","
}
method - Method name from scikit learn
Example:
"method": "svm.SVC"
"method": "neighbors.KNeighborsClassifier"
Or, use extended definition of method
"method": {
"name": "svm.SVC",
"params": {
"kernel": "rbf",
"gamma" : 0.001,
"max_iter": 100
}
}
predict - Data for prediction (in list type) Example: [1,2,3], [1], [0.8,0.5,0.6,0.7]
MIT