func (fft *FastFtrlTrainer) Train(
alpha float64,
beta float64,
l1 float64,
l2 float64,
dropout float64,
model_file string,
train_file string,
test_file string) error {
if !fft.Init {
fft.log4fft.Error("[FastFtrlTrainer-Train] Fast ftrl trainer initialize error.")
return errors.New("[FastFtrlTrainer-Train] Fast ftrl trainer initialize error.")
}
if !util.FileExists(train_file) || !util.FileExists(test_file) {
fft.log4fft.Error("[FastFtrlTrainer-Train] Train file or test file is not exist.")
return errors.New("[FastFtrlTrainer-Train] Train file or test file is not exist.")
}
feat_num, line_cnt, _ := read_problem_info(train_file, fft.CacheFeatureNum, fft.NumThreads)
if feat_num == 0 {
fft.log4fft.Error("[FastFtrlTrainer-Train] The number of features is zero.")
return errors.New("[FastFtrlTrainer-Train] The number of features is zero.")
}
err := fft.ParamServer.Initialize(alpha, beta, l1, l2, feat_num, dropout)
if err != nil {
fft.log4fft.Error(fmt.Sprintf("[FastFtrlTrainer-Train] Parameter server initializing error.%s", err.Error()))
return errors.New(fmt.Sprintf("[FastFtrlTrainer-Train] Parameter server initializing error.%s", err.Error()))
}
return fft.TrainImpl(model_file, train_file, line_cnt, test_file)
}
func StreamRun(model_file string, instances []string) (string, error) {
log := util.GetLogger()
if !util.FileExists(model_file) || len(instances) == 0 {
log.Error("[Predictor-StreamRun] Model file or instances error.")
return fmt.Sprintf(errorjson, "[Predictor-StreamRun] Model file or instances error."), errors.New("[Predictor-StreamRun] Model file or instances error.")
}
var rtstr string
var model solver.LRModel
model.Initialize(model_file)
for i := 0; i < len(instances); i++ {
res, _, x := util.ParseSample(instances[i])
if res != nil {
break
}
pred := model.Predict(x)
pred = math.Max(math.Min(pred, 1.-10e-15), 10e-15)
if i == len(instances)-1 {
rtstr += strconv.FormatFloat(pred, 'f', 6, 64)
} else {
rtstr += strconv.FormatFloat(pred, 'f', 6, 64) + ","
}
}
return fmt.Sprintf(streamjson, rtstr), nil
}