Пример #1
0
func AlgorithmRunOnDataSet(classifier algo.Classifier, train_dataset, test_dataset *core.DataSet, pred_path string, params map[string]string) (float64, []*eval.LabelPrediction) {

	if train_dataset != nil {
		classifier.Train(train_dataset)
	}

	predictions := []*eval.LabelPrediction{}
	var pred_file *os.File
	if pred_path != "" {
		pred_file, _ = os.Create(pred_path)
	}
	for _, sample := range test_dataset.Samples {
		prediction := classifier.Predict(sample)
		if pred_file != nil {
			pred_file.WriteString(strconv.FormatFloat(prediction, 'g', 5, 64) + "\n")
		}
		predictions = append(predictions, &(eval.LabelPrediction{Label: sample.Label, Prediction: prediction}))
	}
	if pred_path != "" {
		defer pred_file.Close()
	}

	auc := eval.AUC(predictions)
	return auc, predictions
}
Пример #2
0
func predict(res http.ResponseWriter, req *http.Request) {
	driverID, _ := strconv.Atoi(req.URL.Query().Get(":id"))
	var model algo.Classifier
	if _, ok := driversModels[driverID]; !ok {
		model = NewModel(driverID)
	} else {
		model = driversModels[driverID]
	}
	fs := make(map[string]float64)
	fs["hour"], _ = strconv.ParseFloat(req.URL.Query().Get("hour"), 64)
	fs["dayOfWeek"], _ = strconv.ParseFloat(req.URL.Query().Get("dayOfWeek"), 64)
	fs["distance_from_order_on_creation"], _ = strconv.ParseFloat(req.URL.Query().Get("distance_from_order_on_creation"), 64)
	fs["driver_location_key"], _ = strconv.ParseFloat(req.URL.Query().Get("driver_location_key"), 64)
	fs["driver_latitude"], _ = strconv.ParseFloat(req.URL.Query().Get("driver_latitude"), 64)
	fs["driver_longitude"], _ = strconv.ParseFloat(req.URL.Query().Get("driver_longitude"), 64)
	fs["origin_location_key"], _ = strconv.ParseFloat(req.URL.Query().Get("origin_location_key"), 64)
	fs["origin_latitude"], _ = strconv.ParseFloat(req.URL.Query().Get("origin_latitude"), 64)
	fs["origin_longitude"], _ = strconv.ParseFloat(req.URL.Query().Get("origin_longitude"), 64)
	sample := NewSample(fs)
	pr := model.Predict(sample)
	renderJSON(res, http.StatusOK, map[string]interface{}{"predict": pr})
}