def trn_validate(df, trn_end, val_end, layer_sizes):
    df_train = split_at(df, end_date=trn_end)
    df_val = split_at(df, start_date=trn_end, end_date=val_end)

    numeric_features_train, scaling_mean, scaling_std = scale(df_train)
    numeric_features_val, _, _ = scale(df_val, scaling_mean, scaling_std)
    nan_trn = numeric_features_train.isnull()
    nan_val = numeric_features_val.isnull()
    imputer = build_imputer(numeric_features_train)

    trn_imputed = apply_imputer(imputer, numeric_features_train)
    val_imputed = apply_imputer(imputer, numeric_features_val)

    trn_imputed = pd.concat((trn_imputed, nan_trn.astype(float)), 1)
    val_imputed = pd.concat((val_imputed, nan_val.astype(float)), 1)

    trn_sold, trn_Y = build_Y(df_train, trn_end)

    val_sold, val_Y = build_Y(df_val, dataset_end)

    n_features = trn_imputed.shape[1]

    model = Model(n_features, layer_sizes=layer_sizes)
    yhat = model.train(
        trn_imputed.values,
        val_imputed.values,
        trn_Y.values,
        val_Y.values,
        trn_sold.values,
        val_sold.values,
        epochs=10)
    model.visualize(
        'Loss and r2_scores ' + str(trn_end)[:10] + ' to \n' +
        str(val_end)[:10] + ' layers: ' + str(layer_sizes),
        str(trn_end)[:7] + ':' + str(val_end)[:7])
    return model.val_losses[-1], yhat, val_Y