仅当列没有值时，Pandas DataFrame 从另一个 dataframe 更新

Question

I am trying to update a dataframe using the values in another dataframe but I would like the update to happen only if a particular column does not have a value.

from datetime import datetime
import pandas as pd

dr = pd.bdate_range(periods=3, end=datetime.now().date())

df1 = pd.DataFrame([1, 2], columns=['myid'])
for d in dr:                                                                                                                                   
    df1[d.to_pydatetime()] = pd.np.nan
df1.loc[df1['myid'] == 1, dr[2]] = 4.0
df1 = df1.set_index('myid')


df1
      2019-11-13 00:00:00  2019-11-14 00:00:00  2019-11-15 00:00:00
myid                                                               
1                     NaN                  NaN                  4.0
2                     NaN                  NaN                  NaN

df2 = pd.DataFrame([1, 2], columns=['myid'])
for d in dr:                                                                                                                                   
    df2[d.to_pydatetime()] = pd.np.nan
df2.loc[df2['myid'] == 2, dr[2]] = 4.0
df2.loc[df2['myid'] == 1, dr[0]] = 6.0
df2 = df2.set_index('myid')

df2
      2019-11-13 00:00:00  2019-11-14 00:00:00  2019-11-15 00:00:00
myid                                                               
1                     6.0                  NaN                  NaN
2                     NaN                  NaN                  4.0

I would like to update df1 with values in df2 if df1 does not have a value for dr[2] (current date)

So in the above example only the second row in df1 should get updated.

I tried update as follows but not sure how to filter based on whether the column has a value or not

df1.update(df2, overwrite=False)

I did look at filter_func that update takes but again unable to make this work with it. Any help is much appreciated. Thanks

EDIT:

Expected output:

Row 1 should not be touched because it already has a value in column 2019-11-15 00:00:00

df1
      2019-11-13 00:00:00  2019-11-14 00:00:00  2019-11-15 00:00:00
myid                                                               
1                     NaN                  NaN                  4.0
2                     NaN                  NaN                  4.0

Answer 1

Update: This seems to be an obvious use for the filter_func argument. Update only rows where all columns of df1 are null:

df1.update(df2, filter_func=lambda df: df1.isnull().all(1))
#      2019-11-13 00:00:00  2019-11-14 00:00:00  2019-11-15 00:00:00
#myid                                                               
#1                     NaN                  NaN                  4.0
#2                     NaN                  NaN                  4.0

---

Old answer, more hands-on:

You can separate which rows to update, update only those rows then combine. update operates inplace so we need to split things out.

m = df1.notnull().any(1)

# These get updated
u = df1[~m].copy()
u.update(df2)

df1 = pd.concat([df1[m], u])
#      2019-11-13 00:00:00  2019-11-14 00:00:00  2019-11-15 00:00:00
#myid                                                               
#1                     NaN                  NaN                  4.0
#2                     NaN                  NaN                  4.0

---

Alternatively, you could use combine_first , then mask rows that shouldn't have been updated and reset them back to the original values in df1

df1.combine_first(df2).mask(df1.notnull().any(1)).fillna(df1)