pandas displaying one column's value in the order of top score

You can gropby('device') later work with every group separatelly to create Series with devide as name, and concatenate with other columns.

import pandas as pd

data = {
    'device': ['A', 'A','B','A','C'], 
    'category': ['soccer','basketball','basketball','soccer','tenis'],
    'score': [10,8,9,9,0],
}

df = pd.DataFrame(data)
print(df)

new = pd.DataFrame()

for key, val in df.groupby('device'):
    data = val.groupby('category').mean().sort_values(by='score', ascending=False)
    print(data)
    s = pd.Series(data.index)
    s.name = key
    new = pd.concat([new, s], axis=1)
    
print(new)   

Result:

  device    category  score
0      A      soccer     10
1      A  basketball      8
2      B  basketball      9
3      A      soccer      9
4      C       tenis      0

            score
category         
soccer        9.5
basketball    8.0

            score
category         
basketball    9.0

          score
category       
tenis       0.0

            A           B      C
0      soccer  basketball  tenis
1  basketball         NaN    NaN

Using numpy:

# Calculate the mean score per category and device type
tmp = df.pivot_table(
    index="category",
    columns="deviceType",
    values="score",
    aggfunc="mean",
)

# Extract the score and category matrix into numpy
score = tmp.to_numpy()
category = np.tile(tmp.index, (len(tmp.columns), 1)).T

# Sort the score descending. numpy always sort ascending so to do descending
# sort, we need to negate the `score` array
index = np.argsort(-score, axis=0)

# Sort score and category
ordered_score = np.take_along_axis(score, index, axis=0)
ordered_category = np.take_along_axis(category, index, axis=0)

# Result
pd.DataFrame(
    np.where(np.isnan(ordered_score), "", ordered_category), columns=tmp.columns
)

Another solution with pandas:

(
    # Calculate the mean score
    df.groupby(["deviceType", "category"], as_index=False)["score"]
    .mean()
    # Sort by the mean score despondingly and assign an order
    .sort_values("score", ascending=False)
    .assign(order=lambda x: x.groupby("deviceType").cumcount())
    # Pivot for final result
    .pivot_table(index="order", columns="deviceType", values="category", aggfunc="first")
)