Divide by zero encountered in true_divide invalid value encountered in matmul

Hi all,

I am dealing with a panel estimation and am fitting the following model:

est = SparseLinearDML(featurizer=PolynomialFeatures(degree=4, include_bias=False),model_y= LassoCV(cv=[(fold00, fold11), (fold11, fold00)], tol=0.0001,n_alphas = 1000, fit_intercept=False,max_iter=100000), 
                                       model_t= MultiTaskElasticNetCV(cv=[(fold00, fold11), (fold11, fold00)],n_alphas = 1000, fit_intercept=False,max_iter=100000),
                                       cv = [(fold0, fold1), (fold1, fold0)],
                                       #linear_first_stages=True,
                                       fit_cate_intercept=False)

However, the error in the title appears. I know it is something about the presence of Nan values or inf but my data look good actually (see attached file data.csv). In particular, I am using the following variables: ln_stdunits ln_price avg_price_atc sales_firms, avg_price_atc, generics, quarter_formatted as pd.datetime() variable, num_prd to equally split the variables for each Productlocal within a molecule.

EDIT: the X is as you suggested for panel data estimation, i.e. a mix individual and time dummies:

for g in np.arange(n_groups):
            for i in np.arange(n_products_per_group):
                for j in np.arange(n_months):
                    if n_products_per_group>1:
                        index_i = n_products_per_group*g + i
                        minus_i = (np.arange(n_products) >= (n_products_per_group*g))
                        minus_i &= (np.arange(n_products) < (n_products_per_group*(g+1)))
                        minus_i &= (np.arange(n_products) != (index_i))
                        p_minus[index_i, j] = np.mean(p[minus_i, j])
                    
        for g in np.arange(n_groups):
            for i in np.arange(n_products_per_group):
                for j in np.arange(n_months):
                    index_i = n_products_per_group*g + i
                    #INDICE INDIVIDUO PER TUTTI GLI INDIVIDUI COMPRESO IL PRIMO
                    #INFATTI IL PRIMO ERA PRESO DI BASELINE SOLO IN CASO DI COLLINEARITA (I.E. SE USO LINEARDML)
                    X[index_i, j, :(n_products)] = 1.0 * (np.arange(0, n_products) == (index_i))
                    #INDICE GRUPPO TRANNE PRIMO GRUPPO
                    X[index_i, j, (n_products):(n_products + n_groups - 1)] =  1.0 * (np.arange(1, n_groups) == g)
                    #INDICE QUARTER TRANNE PRIMO QUARTER
                    X[index_i, j, (n_products + n_groups - 1):] = 1.0 * (np.arange(1, n_months) == j)

Start by reproducing the SparseLinearDML setup with the LassoCV and MultiTaskElasticNetCV models shown, using the attached data.csv and the constructed X panel features. Trace the divide-by-zero and invalid matmul warnings through the fitting path and verify whether the reported inputs contain problematic values. Done means identifying a reproducible failure cause and documenting or testing the expected behavior.