SQL sample groups - performance

Question

I have a sqlite database that I can read as:
In [42]: df = pd.read_sql("SELECT * FROM all_vs_all", engine)
In [43]:
In [43]: df.head()
Out[43]:
user_data user_model \
0 037d05edbbf8ebaf0eca#172.16.199.165 037d05edbbf8ebaf0eca#172.16.199.165
1 037d05edbbf8ebaf0eca#172.16.199.165 060210bf327a3e3b4621#172.16.199.33
2 037d05edbbf8ebaf0eca#172.16.199.165 1141259bd36ba65bef02#172.21.44.180
3 037d05edbbf8ebaf0eca#172.16.199.165 209627747e2af1f6389e#172.16.199.181
4 037d05edbbf8ebaf0eca#172.16.199.165 303a1aff4ab6e3be82ab#172.21.112.182
score Class time_secs model_name bin_id
0 0.283141 0 1514764800 Flow 0
1 0.999300 1 1514764800 Flow 0
2 1.000000 1 1514764800 Flow 0
3 0.206360 1 1514764800 Flow 0
4 1.000000 1 1514764800 Flow 0
As the table is too big I rather than reading the full table I select a random subset of rows:
This can be done very quckly as:
random_query = "SELECT * FROM all_vs_all WHERE abs(CAST(random() AS REAL))/9223372036854775808 < %f AND %s" % (ratio, time_condition)
df = pd.read_sql(random_query, engine)
The problem is that for each triplet [user_data, user_model, time_secs] I want to get all the rows containing that triplet. Each triplet appears 1 or 2 times.
A possible way to do it is to firstly sample a random set of triplets and then get all the rows that have one of the selected triplets but this seems to be too slow.
Is there an efficient way to do it?
EDIT: If I could load all the data in pandas I would have done something like:
selected_groups = []
for group in df.groupby(['user_data', 'user_model', 'time_secs']):
if np.random.uniform(0,1) > ratio:
selected_groups.append(group)
res = pd.concat(selected_groups)

Answer 1

Few sample join and sql query:
currently admitted :
Select p.patient_no, p.pat_name,p.date_admitted,r.room_extension,p.date_discharged FROM Patient p JOIN room r ON p.room_location = r.room_location where p.date_discharged IS NULL ORDER BY p.patient_no,p.pat_name,p.date_admitted,r.room_extension,p.date_discharged;
vacant rooms:
SELECT r.room_location, r.room_accomadation, r.room_extension FROM room r where r.room_location NOT IN (Select p.room_location FROM patient.p where p.date_discharged IS NULL) ORDER BY r.room_location, r.room_accomadation, r.room_extension;
no charges yet :
SELECT p.patient_no, p.pat_name, COALESCE (b.charge,0.00) charge FROM patient p LEFT JOIN billed b on p.patient_no = b.patient_no WHERE p.patient_no NOT IN (SELECT patient_no FROM billed) group by p.patient_no ORDER BY p.patient_no, p.pat_name,charge;
max salarised :
SELECT phy_id,phy_name, salary FROM physician where salary in (SELECT MAX(salary) FROM physician) UNION
SELECT phy_id,phy_name, salary FROM physician where salary in (SELECT MIN(salary) FROM physician) ORDER BY phy_id,phy_name, salary;
various item consumed by:
select p.pat_name, i.discription, count (i.item code) as item code from patient p join billed b on p.patient no = b. patient no join item i on b.item code = i.item code group by p.patient no, i.item code order by..
patient not receivede treatment:
SELECT p.patient_no,p.pat_name FROM patient p where p.patient_no NOT IN (SELECT t.patient_no FROM treats t)
ORDER BY p.patient_no,p.pat_name;
2 high paid :
Select phy_id, phy_name, date_of_joining, max(salary) as salary from physician group by salary having salary IN (Select salary from physician)
Order by phy_id, phy_name, date_of_joining, salary limit 2;
over 200:
select patient_no, sum (charge), as total charge from billed group by patient no having total charges > 200 order by patient no, total charges