从Pandas Dataframe有效地提取数据子集

Question

I have 6 pandas dataframes (Patients, Test1, Test2, Test3, Test4, Test5) linked by an ID key.

Each row in the Patients dataframe represents a patient containing a unique ID there are 200000+ patients/rows.

Each row in the Test dataframes represents a test result on a day. The columns for the Test dataframes are ID, DATE, TEST_UNIT, TEST_RESULT. Each of the Test dataframes contains between 6,000,000 to 7,000,000 rows.

I want to loop through all the IDs in the Patients dataframe and in each iteration use the ID to extract relevant test data from each of the 5 Test dataframes and do some processing on them.

If I do

for i in range(len(Patients)):
    ind_id = Patients.ID.iloc[i]
    ind_test1 = Test1[Test1['ID'] == ind_id]
    ind_test2 = Test2[Test2['ID'] == ind_id]
    ind_test3 = Test3[Test3['ID'] == ind_id]
    ind_test4 = Test4[Test4['ID'] == ind_id]
    ind_test3 = Test5[Test5['ID'] == ind_id]

It takes about 3.6 seconds per iteration.

When I tried to speed it up by using the Numpy interface.

Patients_v = Patients.values
Test1_v = Test1.values
Test2_v = Test2.values
Test3_v = Test3.values
Test4_v = Test4.values
Test5_v = Test5.values

for i in range(len(Patients_v)): 
    ind_id = Patients_v[i, ID_idx]
    ind_test1 = Test1_v[Test1_v[:, 0] == ind_id]
    ind_test2 = Test2_v[Test2_v[:, 0] == ind_id] 
    ind_test3 = Test3_v[Test3_v[:, 0] == ind_id] 
    ind_test4 = Test4_v[Test4_v[:, 0] == ind_id] 
    ind_test5 = Test5_v[Test5_v[:, 0] == ind_id]  

It takes about 0.9 seconds per iteration.

How can I speed this up?

Thank you

Answer 1

It is unclear what output you desire. We can only assume that you want patient-specific dataframes.

In any case, your current code will have to hold all dataframes in memory. This is inefficient. Look at, for example, generator functions :

1. Create a list of all IDs

ALL_IDS = Patients.IDs.tolist()                        # Assuming all you need is the ID

2. Create a master dataframe

ALL_DFS = [Test1, Test2, Test3, Test4, Test5]
df_master = pd.concat(ALL_DFS)

3. Create generator function that yields patient-specific dataframes for further processing

def patient_slices(ALL_IDS):                           # Generator
    for ID in ALL_IDS:
        df_slice = df_master[df_master.ID == ID]
        yield df_slice

df_slice = patient_slices(ALL_IDS)                      
for _ in xrange(len(ALL_IDS)):                         # Call the generator n times
    sinlge_patient = next(df_slice)                    # Next patient for every call    
    your_processing(sinlge_patient)                    # Do your magic