根据 df1 上的条件创建 pd 系列，并报告来自 df2 或 df3 的值

Question

First post here. I'm new to Python, but have made alot of progress leveraging the answers posted here to others questions. Unfortunately i'm having trouble with what seems to be an easy task. I have 3 pandas series, indexed on dates

df1 = {'signal': [0,0,1,1,0,0,1]}  #binary trading signal

df2 = {'SPX': [5,0,5,1,0,5,2]}     #S&P 500 returns

df3 = {'UST': [-1,1,1,0,1,-1,0]}   #10yr Treasury returns

I am trying to create a new series df4 that will represent the return profile of the trading signal. If the signal = 1, get the df3 value on that day, else give me the df2 value (which is for all the zeros)

I've found plenty of posts regarding this topic, which seems very simple, but have struggled to make them work. I tried a simple if statement...

df4 = df1
    if df1 == 1:
        df4.replace(1, df3)
    else:
        df4.replace(0, df2)

But I get ValueError: The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all(). If I add df1.any(), no change is made

I've also tried and failed to use other solutions...

df4 = df1.apply(lambda x: df2 if x == 0 else df3, axis=1)

df4 = df1.loc[df1 == 1, df3] == df2

df4 = df1.select([df1 > 0], [df3], default=df2)

One thing i'm concerned about is that if I replace all the 1s in df4 with a return from df3 and at some point it just so happens the value is a 0... then if I do a second replace for all the 0s in df4, I may place a 0 that should be left along.

Any help to educate me on the most efficient way to do this is very much appreciated.

Answer 1

use Series.where() , specify the column names.

see https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.Series.where.html

>>> df3.where(df1.signal == 1, other=df2.SPX, axis=0)
  UST
0   5
1   0
2   1
3   0
4   0
5   5
6   0

Answer 2

Using numpy.where with the DataFrame values

df1 = pd.DataFrame({'signal': [0,0,1,1,0,0,1]})  #binary trading signal
df2 = pd.DataFrame({'SPX': [5,0,5,1,0,5,2]})     #S&P 500 returns
df3 = pd.DataFrame({'UST': [-1,1,1,0,1,-1,0]})   #10yr Treasury returns

data = np.where(df1.values,df3.values,df2.values)
df4 = pd.DataFrame(data)
#df4 = pd.DataFrame(np.where(df1.values,df3.values,df2.values)) 

---

If the DataFrames actually have more columns you would need to specify - .values isn't actually necessary

pd.DataFrame(np.where(df1['signal'],df3['UST'],df2['SPX']))
# or
pd.DataFrame(np.where(df1.signal,df3.UST,df2.SPX))

---

Using numpy.where is pretty fast compared to DataFrame.where