Solved. Input to predict(row) must be processed, numeric data, not the raw, alpha-numeric data in df_final_nn above.
Thus learn.predict(learn.dls.train.all_cols.iloc[0]) works, coz’ the fields are all numeric:
YearMade 1.000000
Coupler_System 0.000000
ProductSize 3.000000
fiProductClassDesc 48.000000
ModelID 893.000000
fiSecondaryDesc 43.000000
Enclosure 1.000000
Hydraulics_Flow 0.000000
fiModelDesc 2293.000000
ProductGroup 5.000000
Drive_System 0.000000
Hydraulics 1.000000
Tire_Size 0.000000
saleElapsed 1.225075
SalePrice 10.043249
p.s. learn.predict(learn.dls.train.iloc[0]) doesn’t work, coz’ it’s a TabularPandas, not a panda.series.Series