Fix QDA converter crashing on string labels and incorrect shape calculation

[alexarntzen]

Fixes #1168

[xadupre]

I wonder if we can have mixed types?

[alexarntzen]

Yes, sklearn supports mixed types. However, the other converters in sklearn-onnx, like LinearDiscriminantAnalysis, do not. For instance:

from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from skl2onnx.common.data_types import DoubleTensorType
from skl2onnx import convert_sklearn

X = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]])
y = np.array([1, 1, "b", "b"])
skl_model = LinearDiscriminantAnalysis()
skl_model.fit(X, y)
skl_model.classes_ = [1, "b"]  # force mixed types
onnx_model = convert_sklearn(
    skl_model,
    "scikit-learn",
    initial_types=[("inputs", DoubleTensorType([None, X.shape[1]]))],
    target_opset=16,
    options={"zipmap": False},
)

outputs:

ValueError: Label types must be all integers or all strings.

[xadupre]

Can you add a unit test for the new model you added support?

[xadupre]

Please add a unit test.

[alexarntzen]

I added an unit test with string labels for QDA: #6706027