|
// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
// See the LICENSE file in the project root for more information.
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using Microsoft.ML.TorchSharp.Extensions;
using TorchSharp;
namespace Microsoft.ML.TorchSharp.Utils
{
internal class MetricUtils
{
/// <summary>
/// Get top k answer spans for QA task.
/// </summary>
/// <param name="logits">Model output logits</param>
/// <param name="k">Number of highest scores</param>
/// <param name="input0Len"></param>
/// <param name="input1Len"></param>
public static IList<(int start, int end, float score)> ComputeTopKSpansWithScore(torch.Tensor logits, int k, int input0Len, int input1Len)
{
var splitLogits = logits.split(1, dim: -1);
var startLogits = splitLogits[0].squeeze(-1).contiguous(); //[maxseqlen]
var endLogits = splitLogits[1].squeeze(-1).contiguous(); //[maxseqlen]
var (predictStartScores, predictStarts) = startLogits.topk(k);
var (predictEndScores, predictEnds) = endLogits.topk(k);
var startScores = predictStartScores.ToArray<float>();
var endScores = predictEndScores.ToArray<float>();
var starts = predictStarts.ToArray<long>();
var ends = predictEnds.ToArray<long>();
var topK = new List<(int start, int end, float score)>();
for (var i = 0; i < starts.Length; ++i)
{
for (var j = 0; j < ends.Length; ++j)
{
if (starts[i] <= input0Len + 1 || starts[i] >= input0Len + input1Len + 2 || ends[j] <= input0Len + 1 || ends[j] >= input0Len + input1Len + 2 || starts[i] > ends[j])
{
continue;
}
topK.Add(((int)starts[i], (int)ends[j], startScores[i] + endScores[j]));
}
}
topK = topK.OrderByDescending(tuple => tuple.score).Take(k).ToList();
return topK;
}
}
}
|