damjankuznar / pylof Goto Github PK

If the instance you're running lof for is the same distance away from most points, the distances list can have a length < k in which case distances[k - 1][0] returns an index out of range error

Hi

I know the input data must be in a form of tuple. But, I can't apply it on 1 dimensional data:

return lof.outliers(1, data)
File "C:\Users\lof.py", line 162, in outliers
value = l.local_outlier_factor(k, instance)
File "C:\Users\lof.py", line 95, in local_outlier_factor
return local_outlier_factor(min_pts, instance, self.instances, distance_function=self.distance_function)
File "C:\Users\lof.py", line 141, in local_outlier_factor
(k_distance_value, neighbours) = k_distance(min_pts, instance, instances, **kwargs)
File "C:\Users\lof.py", line 104, in k_distance
distance_value = distance_function(instance, instance2)
File "C:\Users\lof.py", line 33, in distance_euclidean
if len(instance1) != len(instance2):
TypeError: object of type 'float' has no len()

Do you have any suggestion?

Also do you have any advice on how to use it on large scale data?

Thanks

Hi Again,

I appreciate if you could put the codes of plotting figures (with outliers as different colors) here too.

Thanks

Hi,

When I'm trying to run example 2 on a different set of data, I got div-by-zero exceptions with some particular data. For example,

import lof
inst = [(1.,2.,3.),(2.,3.,4.),(1.,2.,4.),(1.,2.,1.)]
l = lof.outliers(1, inst)

produces

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/lof.py", line 160, in outliers
    value = l.local_outlier_factor(k, instance)
  File "/lof.py", line 95, in local_outlier_factor
    return local_outlier_factor(min_pts, instance, self.instances, distance_function=self.distance_function)
  File "/lof.py", line 142, in local_outlier_factor
    instance_lrd = local_reachability_density(min_pts, instance, instances, **kwargs)
  File "/lof.py", line 134, in local_reachability_density
    return len(neighbours) / sum(reachability_distances_array)
ZeroDivisionError: float division by zero

I wonder if I'm doing anything wrong? Any help would be appreciated, thanks.

Hi, thanks for sharing this greate project！
There is one line of code I can't understand:
rmse = (sum(map(lambda x: x**2, differences)) / len(differences))**0.5
As far as I know, if x and y are two points in n-dimension space, the euclidean distance between them is

d(x,y) = sqrt((x1-y1)^2 + (x2-y2)^2 + ... + (xn-yn)^2)

So why in the code the sum term is divided by len(differences) ?
Thanks a lot if you could help me.

The performance evaluations of LOF in the paper shows that it takes about 2000 seconds for the algorithm to detect outliers in a 10-dimensional dataset with 200000 data points. However, when I run this implementation of LOF it takes about 2400 seconds to run a 7-dimensional dataset with just 1000 points. Why the huge discrepancy in performance?

I'm just getting started with LOF and just want to experiment with it a little. I'm sorry if I'm just misinterpreting the paper.

Hello~ I just know the LOF algorithm today and I want to use this algorithm on a 3D point set to cluster the outliers. I find your great work on git, but I don't quite understand how to use your lof file. Would you like to show me the way please? Thank you for your patience.