SciSharp / NumSharp

High Performance Computation for N-D Tensors in .NET, similar API to NumPy.
https://github.com/SciSharp
Apache License 2.0
1.38k stars 192 forks source link

np.save fails with IndexOutOfRangeException for jagged arrays #410

Open Jmerk523 opened 4 years ago

Jmerk523 commented 4 years ago

Hi, I am seeing this exception when trying to save a jagged array to a stream:

Index was outside the bounds of the array. at NumSharp.np.d__109`1.MoveNext() at NumSharp.np.writeValueJagged(BinaryWriter reader, Array matrix, Int32 bytes, Int32[] shape) at NumSharp.np.Save(Array array, Stream stream)

I haven't pinpointed the root cause specifically, but it seems that there is some issue with jagged arrays when the outer array has rank 1:

        static IEnumerable<T> Enumerate<T>(Array a, int[] dimensions, int pos)
        {
            if (pos == dimensions.Length - 1)
            {
                for (int i = 0; i < dimensions[pos]; i++)
                    yield return (T)a.GetValue(i);
            }
            else
            {
                for (int i = 0; i < dimensions[pos]; i++)
                    foreach (var subArray in Enumerate<T>(a.GetValue(i) as Array, dimensions, pos + 1))
                        yield return subArray;
            }
        }

When pos == 0 and dimensions[] has length 0, the else will execute and pos(0) will be out of bounds. dimensions[] has length 0 from the caller, which passes: int[] first = shape.Take(shape.Length - 1).ToArray(); so first[] will be an empty array when shape has length 1 (as in, the single dimension of the 1 dimensional outer jagged array).