PCovC takes forever to fit on sparse fingerprint without data scaling

[mdepper009]

A 2048-bit RDKit fingerprint is performed on X_train.

array([[0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0],
       ...,
       [0, 0, 0, ..., 0, 0, 1],
       [0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0]], shape=(5807, 2048))

Fitting PCovC to this feature array takes around 5-8 minutes to run.

lr = LogisticRegression()
lr.fit(X_train,y_train)

pcovc = PCovC(mixing=0.05,classifier=lr,n_components=2)
pcovc.fit(X_train,y_train)

Now, X_train is fit transformed with a StandardScaler.

array([[-0.59651099, -0.46626218, -0.49935414, ..., -0.34791159,
        -0.4461504 , -0.92512945],
       [-0.59651099, -0.46626218, -0.49935414, ..., -0.34791159,
        -0.4461504 , -0.92512945],
       [-0.59651099, -0.46626218, -0.49935414, ..., -0.34791159,
        -0.4461504 , -0.92512945],
       ...,
       [-0.59651099, -0.46626218, -0.49935414, ..., -0.34791159,
        -0.4461504 ,  1.08092981],
       [-0.59651099, -0.46626218, -0.49935414, ..., -0.34791159,
        -0.4461504 , -0.92512945],
       [-0.59651099, -0.46626218, -0.49935414, ..., -0.34791159,
        -0.4461504 , -0.92512945]], shape=(5807, 2048))

Fitting PCovC to this scaled feature array takes around 15 seconds to run.

lr = LogisticRegression()
lr.fit(X_train_scaled,y_train)

pcovc = PCovC(mixing=0.05,classifier=lr,n_components=2)
pcovc.fit(X_train_scaled,y_train)

It seems strange that the unscaled fingerprint would take so much longer than the scaled fingerprint to fit. What could be causing this?