I’m attempting to use the onnxruntime wrapper for Go to inference an ONNX model.

Model visualization:

The model inference works fine when using the onnxruntime in Python, which does not require specifying tensor sizes. However, when attempting to run in Go, I get the following:

Error running the model: Error running network: Invalid rank for
output: predictions Got: 2 Expected: 1 Please fix either the
inputs/outputs or the model.

package main

import (
    "fmt"

    ort "github.com/yalue/onnxruntime_go"
)

func main() {
    ort.SetSharedLibraryPath("/opt/homebrew/Cellar/onnxruntime/1.17.1/lib/libonnxruntime.dylib")
    err := ort.InitializeEnvironment()
    defer ort.DestroyEnvironment()
    if err == nil {
        fmt.Println("Initialized")
    }

    inputData := [][]float32{
        {546.0, 410, 4.63, 0, 8592, 5192, 34728, 9000, 3, 0, 2, 55461, 3, 20, 3},
        {386.0, 398, 4.52, 0, 6699, 4217, 34728, 9000, 3, 0, 2, 55253, 3, 20, 3},
        {69.0, 2, 5.0, 0, 2191, 1105, 34728, 9000, 3, 0, 2, 55152, 3, 20, 3},
        {359.0, 387, 4.48, 0, 5262, 3053, 34728, 9000, 3, 0, 2, 55101, 3, 20, 3},
        {213.0, 396, 4.63, 0, 4545, 2614, 34728, 9000, 3, 0, 2, 55013, 3, 20, 3},
        {165.0, 398, 4.63, 0, 3715, 2522, 34728, 9000, 3, 0, 2, 54449, 3, 20, 3},
    }

    rows := len(inputData)
    cols := len(inputData[0])
    // Flatten the two-dimensional slice into a one-dimensional slice
    flatData := make([]float32, 0, rows*cols)
    for _, row := range inputData {
        flatData = append(flatData, row...)
    }

    inputShape := ort.NewShape(int64(rows), int64(cols))
    inputTensor, err := ort.NewTensor(inputShape, flatData)
    if err != nil {
        fmt.Println(err)
    }
    fmt.Println(inputTensor)

    outputShape := ort.NewShape(int64(rows), 1)
    outputTensor, err := ort.NewEmptyTensor[float32](outputShape)
    if err != nil {
        fmt.Println("Failed to create output tensor:", err)
        return
    }

    defer outputTensor.Destroy()

    // Create the session with the model and specify input/output details
    session, err := ort.NewSession[float32]("CatBoost_RMSE.onnx",
        []string{"features"}, []string{"predictions"},
        []*ort.Tensor[float32]{inputTensor}, []*ort.Tensor[float32]{outputTensor})
    if err != nil {
        fmt.Println("Failed to create session:", err)
        return
    }
    defer session.Destroy()

    // Run the model
    err = session.Run()
    if err != nil {
        fmt.Println("Error running the model:", err)
        return
    }

    // Retrieve and print the output data
    outputData := outputTensor.GetData()
    fmt.Println("Model output data:", outputData)
}

I have tried updating ort.NewShape(int64(rows), 1) to ort.NewShape(int64(rows))

but that results in a new error:

Error running the model: Error running network: Non-zero status code
returned while running TreeEnsembleRegressor node. Name:” Status
Message:
/tmp/onnxruntime-20240227-5139-v94cng/onnxruntime/core/framework/execution_frame.cc:173
Status
onnxruntime::IExecutionFrame::GetOrCreateNodeOutputMLValue(const int,
int, const TensorShape *, OrtValue *&, const Node &) shape &&
tensor.Shape() == *shape was false. OrtValue shape verification
failed. Current shape:{6} Requested shape:{6,1}