Dataframe fillna conditional based on Index & Column Name

Question

I want to be able to use the df.fillna() function on a Dataframe, but apply a conditional to it based on the Index & Column name of that particular cell.

I am trying to create a heatmap of hockey linemate data based on the following dataset (apologies for the large dictionary below) -

linemates_toi = {
    'Player 1': {'Player 2': 0.25, 'Player 3': 7.95, 'Player 4': 0.6333, 'Player 5': 9.95, 'Player 6': 0.6333, 'Player 7': 0.8, 'Player 8': 4.2667, 'Player 9': 7.8833, 'Player 10': 0.3, 'Player 11': 11.2333, 'Player 12': 3.35, 'Player 13': 0.2167},
    'Player 10': {'Player 14': 2.3, 'Player 18': 1.2667, 'Player 2': 6.8333, 'Player 4': 5.5833, 'Player 5': 0.9, 'Player 16': 6.9167, 'Player 6': 4.9667, 'Player 7': 4.15, 'Player 15': 1.0, 'Player 8': 0.3167, 'Player 17': 5.3167, 'Player 1': 0.3, 'Player 11': 1.6167, 'Player 12': 0.6833, 'Player 13': 12.7167}, 
    'Player 12': {'Player 14': 4.5333, 'Player 18': 4.3333, 'Player 2': 3.1167, 'Player 3': 1.2333, 'Player 4': 5.7333, 'Player 5': 3.5167, 'Player 16': 3.0, 'Player 6': 3.0167, 'Player 7': 2.4, 'Player 15': 2.0167, 'Player 8': 11.6667, 'Player 17': 2.2667, 'Player 9': 0.1167, 'Player 1': 3.35, 'Player 10': 0.6833, 'Player 11': 3.35}, 
    'Player 17': {'Player 14': 4.55, 'Player 18': 1.65, 'Player 2': 0.8833, 'Player 3': 2.85, 'Player 5': 0.0333, 'Player 16': 2.9167, 'Player 6': 7.8167, 'Player 7': 6.0833, 'Player 8': 3.8, 'Player 9': 2.25, 'Player 10': 5.3167, 'Player 12': 2.2667, 'Player 13': 5.7833},
    'Player 7': {'Player 18': 0.3667, 'Player 2': 0.6667, 'Player 3': 1.55, 'Player 4': 0.3333, 'Player 5': 0.15, 'Player 16': 1.2167, 'Player 6': 6.8333, 'Player 15': 0.3333, 'Player 8': 3.0667, 'Player 17': 6.0833, 'Player 9': 1.8833, 'Player 1': 0.8, 'Player 10': 4.15, 'Player 11': 1.0, 'Player 12': 2.4, 'Player 13': 4.4333}, 
    'Player 16': {'Player 14': 2.2833, 'Player 2': 8.5333, 'Player 3': 2.7, 'Player 4': 8.0167, 'Player 5': 0.45, 'Player 6': 0.4, 'Player 7': 1.2167, 'Player 8': 2.3, 'Player 17': 2.9167, 'Player 9': 2.15, 'Player 10': 6.9167, 'Player 11': 0.1333, 'Player 12': 3.0, 'Player 13': 6.5833},
    'Player 18': {'Player 14': 10.05, 'Player 2': 0.75, 'Player 3': 5.0, 'Player 4': 3.45, 'Player 5': 0.3333, 'Player 6': 0.8333, 'Player 7': 0.3667, 'Player 15': 5.2, 'Player 8': 5.8167, 'Player 17': 1.65, 'Player 9': 4.3833, 'Player 10': 1.2667, 'Player 11': 1.5, 'Player 12': 4.3333, 'Player 13': 1.5333},
    'Player 13': {'Player 14': 3.0333, 'Player 18': 1.5333, 'Player 2': 5.9167, 'Player 3': 0.7333, 'Player 4': 4.95, 'Player 5': 0.8167, 'Player 16': 6.5833, 'Player 6': 5.1333, 'Player 7': 4.4333, 'Player 15': 1.2667, 'Player 8': 0.2833, 'Player 17': 5.7833, 'Player 1': 0.2167, 'Player 10': 12.7167, 'Player 11': 1.5333},
    'Player 5': {'Player 18': 0.3333, 'Player 2': 0.8333, 'Player 3': 8.0333, 'Player 16': 0.45, 'Player 6': 0.3333, 'Player 7': 0.15, 'Player 8': 3.0167, 'Player 17': 0.0333, 'Player 9': 6.7333, 'Player 1': 9.95, 'Player 10': 0.9, 'Player 11': 11.2333, 'Player 12': 3.5167, 'Player 13': 0.8167},
    'Player 15': {'Player 14': 4.5667, 'Player 18': 5.2, 'Player 2': 0.4667, 'Player 3': 2.35, 'Player 6': 0.1667, 'Player 7': 0.3333, 'Player 8': 2.0167, 'Player 9': 2.0833, 'Player 10': 1.0, 'Player 12': 2.0167, 'Player 13': 1.2667},
    'Player 2': {'Player 18': 0.75, 'Player 3': 2.65, 'Player 4': 8.6, 'Player 5': 0.8333, 'Player 16': 8.5333, 'Player 6': 0.8333, 'Player 7': 0.6667, 'Player 15': 0.4667, 'Player 8': 2.3333, 'Player 17': 0.8833, 'Player 9': 1.9167, 'Player 1': 0.25, 'Player 10': 6.8333, 'Player 11': 1.6167, 'Player 12': 3.1167, 'Player 13': 5.9167},
    'Player 8': {'Player 14': 5.8333, 'Player 18': 5.8167, 'Player 2': 2.3333, 'Player 3': 1.1167, 'Player 4': 5.6833, 'Player 5': 3.0167, 'Player 16': 2.3, 'Player 6': 4.2667, 'Player 7': 3.0667, 'Player 15': 2.0167, 'Player 17': 3.8, 'Player 9': 1.1333, 'Player 1': 4.2667, 'Player 10': 0.3167, 'Player 11': 3.8167, 'Player 12': 11.6667, 'Player 13': 0.2833},
    'Player 4': {'Player 14': 3.2833, 'Player 18': 3.45, 'Player 2': 8.6, 'Player 3': 2.0667, 'Player 16': 8.0167, 'Player 6': 0.8333, 'Player 7': 0.3333, 'Player 8': 5.6833, 'Player 9': 1.85, 'Player 1': 0.6333, 'Player 10': 5.5833, 'Player 11': 0.85, 'Player 12': 5.7333, 'Player 13': 4.95},
    'Player 9': {'Player 14': 4.5167, 'Player 18': 4.3833, 'Player 2': 1.9167, 'Player 3': 14.35, 'Player 4': 1.85, 'Player 5': 6.7333, 'Player 16': 2.15, 'Player 6': 0.8833, 'Player 7': 1.8833, 'Player 15': 2.0833, 'Player 8': 1.1333, 'Player 17': 2.25, 'Player 1': 7.8833, 'Player 11': 9.0667, 'Player 12': 0.1167},
    'Player 14': {'Player 18': 10.05, 'Player 3': 5.7167, 'Player 4': 3.2833, 'Player 16': 2.2833, 'Player 6': 1.8833, 'Player 15': 4.5667, 'Player 8': 5.8333, 'Player 17': 4.55, 'Player 9': 4.5167, 'Player 10': 2.3, 'Player 11': 0.9833, 'Player 12': 4.5333, 'Player 13': 3.0333},
    'Player 11': {'Player 14': 0.9833, 'Player 18': 1.5, 'Player 2': 1.6167, 'Player 3': 9.7667, 'Player 4': 0.85, 'Player 5': 11.2333, 'Player 16': 0.1333, 'Player 6': 0.5, 'Player 7': 1.0, 'Player 8': 3.8167, 'Player 9': 9.0667, 'Player 1': 11.2333, 'Player 10': 1.6167, 'Player 12': 3.35, 'Player 13': 1.5333},
    'Player 6': {'Player 14': 1.8833, 'Player 18': 0.8333, 'Player 2': 0.8333, 'Player 3': 1.1333, 'Player 4': 0.8333, 'Player 5': 0.3333, 'Player 16': 0.4, 'Player 7': 6.8333, 'Player 15': 0.1667, 'Player 8': 4.2667, 'Player 17': 7.8167, 'Player 9': 0.8833, 'Player 1': 0.6333, 'Player 10': 4.9667, 'Player 11': 0.5, 'Player 12': 3.0167, 'Player 13': 5.1333},
    'Player 3': {'Player 14': 5.7167, 'Player 18': 5.0, 'Player 2': 2.65, 'Player 4': 2.0667, 'Player 5': 8.0333, 'Player 16': 2.7, 'Player 6': 1.1333, 'Player 7': 1.55, 'Player 15': 2.35, 'Player 8': 1.1167, 'Player 17': 2.85, 'Player 9': 14.35, 'Player 1': 7.95, 'Player 11': 9.7667, 'Player 12': 1.2333, 'Player 13': 0.7333}
}

df = pd.DataFrame(linemates_toi)

What I am trying to achieve now is to use df.fillna(0) and apply a conditional so the only NaN that are replaced is when the Index and Column name don't match because I want those cells to remain NaN so that when I plot them into a Heatmap they don't have any color in the cmap applied from Matplotlib.

If I were writing pseudo code, it would look like this -

df.fillna(0, df.cell.Index.Name != df.cell.Column.Name)

Thanks in advance!

Answer 1

Using some broadcasting and NaN -masking

mask = df.index.to_numpy() == df.columns.to_numpy()[:, None]

df.fillna(0).mask(mask)

---

>>> df.head()

           Player 1  Player 10  Player 12  Player 17  Player 7  Player 16  \
Player 1        NaN     0.3000     3.3500     0.0000    0.8000     0.0000   
Player 10    0.3000        NaN     0.6833     5.3167    4.1500     6.9167   
Player 11   11.2333     1.6167     3.3500     0.0000    1.0000     0.1333   
Player 12    3.3500     0.6833        NaN     2.2667    2.4000     3.0000   
Player 13    0.2167    12.7167     0.0000     5.7833    4.4333     6.5833   

           Player 18  Player 13  Player 5  Player 15  Player 2  Player 8  \
Player 1      0.0000     0.2167    9.9500     0.0000    0.2500    4.2667   
Player 10     1.2667    12.7167    0.9000     1.0000    6.8333    0.3167   
Player 11     1.5000     1.5333   11.2333     0.0000    1.6167    3.8167   
Player 12     4.3333     0.0000    3.5167     2.0167    3.1167   11.6667   
Player 13     1.5333        NaN    0.8167     1.2667    5.9167    0.2833   

           Player 4  Player 9  Player 14  Player 11  Player 6  Player 3  
Player 1     0.6333    7.8833     0.0000    11.2333    0.6333    7.9500  
Player 10    5.5833    0.0000     2.3000     1.6167    4.9667    0.0000  
Player 11    0.8500    9.0667     0.9833        NaN    0.5000    9.7667  
Player 12    5.7333    0.1167     4.5333     3.3500    3.0167    1.2333  
Player 13    4.9500    0.0000     3.0333     1.5333    5.1333    0.7333 

Answer 2

You could alternatively do the following:

df.fillna(0, inplace=True)
for col in df:
    df.loc[col, col] = np.nan

Explanation:

1. Fill NaNs
2. Set NaN where row name is equal to column name

Answer 3

Use df.apply to map a lambda over each column:

df = df.apply(lambda col: col.where((col.name == col.index) | col.notnull(), 0))

col.where(condition, value_if_false) returns the original value in col if condition is true. Otherwise it returns value_if_false