I have arrays similar to the following:
a=[["tennis","tennis","golf","federer","cricket"],
["federer","nadal","woods","sausage","federer"],
["sausage","lion","prawn","prawn","sausage"]]
I then have a matrix of the following weights
w=[[1,3,3,4,5],
[2,3,2,3,4],
[1,2,1,1,1]]
What I am looking to then do is to sum the weights based on the labels of matrix a for each row and take the top 3 labels from that row. So at the end I would like something like this:
res=[["cricket","tennis","federer"],
["federer","sausage","nadal"],
["lion","sausage","prawn"]]
In my actual data set ties would be highly unlikely and are not really a concern, also for cases where say the entire row is:
["federer","federer","federer","federer","federer"]
Ideally I would like this to be returned as ["federer","",""].
Any guidance would be appreciated.
See piRSquared answer for numpy arrays.
This is a pure python approach:
for i in range(4):
if a[i].count(a[i][0]) == len(a[i]):
res = [a[1][0], "", ""]
else:
res = [x[0] for x in sorted(zip(a[i], w[i]), key=lambda c: c[1], reverse=True)[:3]]
print(res)
Try:
print pd.DataFrame(
{i: a.loc[i, row.sort_values(ascending=False).index[:3]].values for i, row in w.iterrows()}
).T
0 1 2
0 cricket federer golf
1 federer sausage nadal
2 lion sausage prawn
I managed to get it to work using below code:
def myf(a,w):
lookupTable, indexed_dataSet = np.unique(a, return_inverse=True)
y= np.bincount(indexed_dataSet,w)
lookupTable[y.argsort()]
res=(lookupTable[y.argsort()][::-1][:3])
ret=np.empty((3))
ret.fill(res[-1])
ret[0:res.shape[0]]=res
return ret
result = np.empty_like(knearest_labels[:,0:3])
for i,(x,y) in enumerate(zip(a,w)):
result[i] = myf(x,y)
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.