TheAlgorithms/C-Sharp · error · ArgumentException

Feature count mismatch.

Error message

Feature count mismatch.

What it means

PredictProbability validates that a sample has the same number of features as the weight vector learned during Fit. A mismatched vector cannot be dotted with the weights, so the library throws ArgumentException instead of computing a garbage result.

Solutions

1. Ensure each prediction sample has exactly the same features, in the same order, as training data.
2. Check x.Length against the model's feature count (e.g. weights.Length) before predicting.
3. If passing multiple samples, call the batch Predict API rather than looping the wrong-shaped data into PredictProbability.

Example fix

// before
model.PredictProbability(new double[] { 1.0, 2.0, 3.0 }); // trained on 2 features

// after
model.PredictProbability(new double[] { 1.0, 2.0 }); // matches weights.Length

if (sample.Length != expectedFeatureCount)
{
    throw new ArgumentException($"Expected {expectedFeatureCount} features, got {sample.Length}.");
}
var p = model.PredictProbability(sample);

static bool MatchesFeatureCount(double[] x, int n) => x != null && x.Length == n;

try
{
    var p = model.PredictProbability(sample);
}
catch (ArgumentException ex) when (ex.Message == "Feature count mismatch.")
{
    logger.LogError("Sample has {Len} features, model expects {Exp}", sample.Length, model.FeatureCount);
    throw;
}

When it happens

Trigger: Calling PredictProbability(double[] x) (directly or via Predict) with a vector whose Length differs from the feature count used in Fit, e.g. forgetting a bias/intercept column or passing multiple samples instead of one.

Common situations: Serving-time feature engineering differs from training (extra/dropped column); passing a 2D row set to the single-sample API; changing the dataset schema after the model was fitted.

Understand the failure class

Background: Tensor shape mismatch errors ("must have shape", "expected shape ... got ..."): when tensor dimensions disagree with what an op or layer was told to expect — this error's family across 6 libraries.

Related errors

• Number of samples and labels must match.
• k must be at least 1.
• Input features cannot be empty.
• message
• key must be non-empty string

AI-assisted analysis of TheAlgorithms/C-Sharp@96e2905cab (2026-09-13). Data as JSON: /api/errors/525601d2ed5f575c. Report an issue: GitHub.

Appendix: source

Thrown at Algorithms/MachineLearning/LogisticRegression.cs:73

            }

            for (int j = 0; j < nFeatures; j++)
            {
                weights[j] -= learningRate * dw[j] / nSamples;
            }

            bias -= learningRate * db / nSamples;
        }
    }

    /// <summary>
    /// Predict probability for a single sample.
    /// </summary>
    public double PredictProbability(double[] x)
    {
        if (x.Length != weights.Length)
        {
            throw new ArgumentException("Feature count mismatch.");
        }

        return Sigmoid(Dot(x, weights) + bias);
    }

    /// <summary>
    /// Predict class label (0 or 1) for a single sample.
    /// </summary>
    public int Predict(double[] x) => PredictProbability(x) >= 0.5 ? 1 : 0;

    private static double Sigmoid(double z) => 1.0 / (1.0 + Math.Exp(-z));

    private static double Dot(double[] a, double[] b) => a.Zip(b).Sum(pair => pair.First * pair.Second);
}

View on GitHub (pinned to 96e2905cab)