ELKI hierarchical clustering - "mrg_" Cluster object - machine-learning

I'm using ELKI's SimplifiedHierarchyExtraction with AnderbergHierarchicalClustering, LatLngDistanceFunction and minClSize = 100.
I saw that beside the "clu_" Clusters there are also 2 -3 "mrg_" Clusters which have some DBID's, but the number of it is < minClSize.
My question is: what is the best way to handle this "mrg_" Clusters?:
passing its DBID´s to one of its "clu_" children?
taking them as a cluster although they are under the minClSize?
simply ignoring them?

This is a hierarchical result.
You need to include all child clusters into a cluster.
So the mrg_ cluster has some (potentially 0) new objects, plus all those objects in child clusters. In particular, it can have more than one child cluster (that is why it is called merge)

Related

Calculating a full matrix of shortest path-lengths between all nodes

We are trying to find a way to create a full distance matrix in a neo4j database, where that distance is defined as the length of the shortest path between any two nodes. Of course, there is the shortestPath method but using a loop going through all pairs of nodes and calculating their shortestPaths get very slow. We are explicitely not talking about allShortestPaths, because that returns all shortest paths between 2 specific nodes.
Is there a specific method or approach that is fast for a large number of nodes (>30k)?
Thank you!
j.
There is no easier method; the full distance matrix will take a long time to build.
As you've described it, the full distance matrix must contain the shortest path between any two nodes, which means you will have to get that information at some point. Iterating over each pair of nodes and running a shortest-path algorithm is the only way to do this, and the complexity will be O(n) multiplied by the complexity of the algorithm.
But you can cut down on the runtime with a dynamic programming solution.
You could certainly leverage some dynamic programming methods to cut down on the calculation time. For instance, if you are trying to find the shortest path between (A) and (C), and have already calculated the shortest from (B) to (C), then if you happen to encounter (B) while pathfinding from (A), you do not need to recalculate the rest of the cost of that path; it is known.
However, creating a dynamic programming solution of any reasonable complexity will almost certainly be best done in a separate module for Neo4J that is thrown in into a plugin. If what you are doing is a one-time operation or an operation that won't be run frequently, it might be easier to just do the naive solution of calling shortestPath between each pair, but if you plan to be running it fairly frequently on dynamic data, it might be worth authoring a custom plugin. It totally depends on your needs.
No matter what, though, it will take some time to calculate. The dynamic programming solution will cut down on the time greatly (especially in a densely-connected graph), but it will still not be very fast.
What is the end game? Is this a one-time query that resets some property or creates new edges. Or a recurring frequent effort. If it's one-time, you might create edges between the two nodes at each step creating a transitive closure environment. The edge would point between the two nodes and have, as a property, the distance.
Thus, if the path is a>b>c>d, you would create the edges
a>b 1
a>c 2
a>d 3
b>c 1
b>d 2
c>d 1
The edges could be named distinctively to distinguish them from the original path edges. This could create circular paths, which may neither negate this strategy or need a constraint. if you are dealing with directed acyclic graphs it would work well.

In beam custom combine function, does serialization occur even if the object is on "same" machine?

We have a custom combine function (on beam sdk 2.0) in which the millions of objects get accumulated but they do NOT necessarily get reduced....that is, they sometimes get added to a List such that eventually, the List might get quite large (hundreds of megabytes, even gigabytes).
To minimize the problem of having to "pass around" these objects (during merging of accumulators) between nodes, we've created a SINGLE giant node (of 64 cores, tonnes of RAM).
So, in "theory", dataflow does not need to serialize the List object (and any of these big objects in the List) even during "merge accumulator" operations, since all the objects are on the same node. But, does dataflow still serialize even if all the objects of interest are on the same node or is it smart enough to know that an object is on the same node vs separate nodes?
Ideally, when objects are on same node, we can just pass around references to the objects (rather than serializing/deserializing the contents of these objects, which can be very very large.) (I understand, of course, than when dealing with multiple nodes, there's no choice but to serialize/deserialize since the data has to be passed around somehow; but within a node, is beam sdk 2.0 smart enough to not serialize/deserialize during these combine functions, group by's etc.?)
The Dataflow service aggressively optimizes your pipeline to avoid needless serialization. The optimization you are interested in is fusion, described here in the Dataflow documentation. When data moves through a fused "stage" (a sequence of low-level instructions roughly corresponding to steps in your input pipeline), it is not serialized and deserialized.
However, if your CombineFn builds a list, and that list grows large, you should try to rephrase your pipeline to use a raw GroupByKey. Another important optimization is "combiner lifting" or "mapper-side combine" where your CombineFn is applied per-key locally prior to shuffling your data between machines, based on the assumption that the accumulator will be smaller than just a list of elements. So the whole list will be serialized, shuffled, and deserialized prior to completing the Combine transform. If, instead, you use a GroupByKey directly, your elements would be much more efficiently streamed, without serializing an entire list.
I should note that Beam's other runners also perform standard fusion optimization and others. These all generally come from functional programming work in the late 80s / early 90s and was applied to distributed data processing in FlumeJava, circa 2010, so it is a baseline expectation now.

Too many 'steps' when executing a pipeline

We have a large data set which needs to be partition into 1,000 separate files, and the simplest implementation we wanted to use is to apply PartitionFn which, given an element of the data set, returns a random integer between 1 and 1,000.
The problem with this approach is it ends up creating 1,000 PCollections and the pipeline does not launch as there seems to be a hard limit on the number of 'steps' (which correspond to the boxes shown on the job monitoring UI in execution graph).
Is there a way to increase this limit (and what is the limit)?
The solution we are using to get around this issue is to partition the data into a smaller subsets first (say 50 subsets), and for each subset we run another layer of partitioning pipelines to produce 20 subsets of each subset (so the end result is 1000 subsets), but it'll be nice if we can avoid this extra layer (as ends up creating 1 + 50 pipelines, and incurs the extra cost of writing and reading the intermediate data).
Rather than using the Partition transform and introducing many steps in the pipeline consider using either of the following approaches:
Many sinks support the option to specify the number of output shards. For example, TextIO has a withNumShards method. If you pass this 1000 it will produce 1000 separate shards in the specified directory.
Using the shard number as a key and using a GroupByKey + a DoFn to write the results.

NeuroEvolution of Augmenting Topologies (NEAT) and global innovation number

I was not able to find why we should have a global innovation number for every new connection gene in NEAT.
From my little knowledge of NEAT, every innovation number corresponds directly with an node_in, node_out pair, so, why not only use this pair of ids instead of the innovation number? Which new information there is in this innovation number? chronology?
Update
Is it an algorithm optimization?
Note: this more of an extended comment than an answer.
You encountered a problem I also just encountered whilst developing a NEAT version for javascript. The original paper published in ~2002 is very unclear.
The original paper contains the following:
Whenever a new
gene appears (through structural mutation), a global innovation number is incremented
and assigned to that gene. The innovation numbers thus represent a chronology of the
appearance of every gene in the system. [..] ; innovation numbers are never changed. Thus, the historical origin of every
gene in the system is known throughout evolution.
But the paper is very unclear about the following case, say we have two ; 'identical' (same structure) networks:
The networks above were initial networks; the networks have the same innovation ID, namely [0, 1]. So now the networks randomly mutate an extra connection.
Boom! By chance, they mutated to the same new structure. However, the connection ID's are completely different, namely [0, 2, 3] for parent1 and [0, 4, 5] for parent2 as the ID is globally counted.
But the NEAT algorithm fails to determine that these structures are the same. When one of the parents scores higher than the other, it's not a problem. But when the parents have the same fitness, we have a problem.
Because the paper states:
In composing the offspring, genes are randomly chosen from veither parent at matching genes, whereas all excess or disjoint genes are always included from the more fit parent, or if they are equally fit, from both parents.
So if the parents are equally fit, the offspring will have connections [0, 2, 3, 4, 5]. Which means that some nodes have double connections... Removing global innovation counters, and just assign id's by looking at node_in and node_out, you avoid this problem.
So when you have equally fit parents, yes you have optimized the algorithm. But this is almost never the case.
Quite interesting: in the newer version of the paper, they actually removed that bolded line! Older version here.
By the way, you can solve this problem by instead of assigning innovation ID's, assign ID based on node_in and node_out using pairing functions. This creates quite interesting neural networks when fitness is equal:
I can't provide a detailed answer, but the innovation number enables certain functionality within the NEAT model to be optimal (like calculating the species of a gene), as well as allowing crossover between the variable length genomes. Crossover is not necessary in NEAT, but it can be done, due to the innovation number.
I got all my answers from here:
http://nn.cs.utexas.edu/downloads/papers/stanley.ec02.pdf
It's a good read
During crossover, we have to consider two genomes that share a connection between the two same nodes in their personal neural networks. How do we detect this collision without iterating both genome's connection genes over and over again for each step of crossover? Easy: if both connections being examined during crossover share an innovation number, they are connecting the same two nodes because they received that connection from the same common ancestor.
Easy Example:
If I am a genome with a specific connection gene with innovation number 'i', my children that take gene 'i' from me may eventually cross over with each other in 100 generations. We have to detect when these two evolved versions (alleles) of my gene 'i' are in collision to prevent taking both. Taking two of the same gene would cause the phenotype to probably loop and crash, killing the genotype.
When I created my first implementation of NEAT I thought the same... why would you keep a innovation number tracker...? and why would you use it only for one generation? Wouldn't be better to not keep it at all and use a key value par with the nodes connected?
Now that I am implementing my third revision I can see what Kenneth Stanley tried to do with them and why he wanted to keep them only for one generation.
When a connection is created, it will start its optimization in that moment. It marks its origin. If the same connection pops out in another generation, that will start its optimization then. Generation numbers try to separate the ones which come from a common ancestor, so the ones that have been optimized for many generations are not put side to side that one that was just generated. If a same connection is found in two genomes, that means that that gene comes from the same origin and thus, can be aligned.
Imagine then that you have your generation champion. Some of their genes will have 50 percent chance to be lost due that the aligned genes are treated equally.
What is better...? I haven't seen any experiments comparing the two approaches.
Kenneth Stanley also addressed this issue in the NEAT users page: https://www.cs.ucf.edu/~kstanley/neat.html
Should a record of innovations be kept around forever, or only for the current
generation?
In my implementation of NEAT, the record is only kept for a generation, but there
is nothing wrong with keeping them around forever. In fact, it may work better.
Here is the long explanation:
The reason I didn't keep the record around for the entire run in my
implementation of NEAT was because I felt that calling something the same
mutation that happened under completely different circumstances was not
intuitive. That is, it is likely that several generations down the line, the
"meaning" or contribution of the same connection relative to all the other
connections in a network is different than it would have been if it had appeared
generations ago. I used a single generation as a yardstick for this kind of
situation, although that is admittedly ad hoc.
That said, functionally speaking, I don't think there is anything wrong with
keeping innovations around forever. The main effect is to generate fewer species.
Conversely, not keeping them around leads to more species..some of them
representing the same thing but separated nonetheless. It is not currently clear
which method produces better results under what circumstances.
Note that as species diverge, calling a connection that appeared in one species a
different name than one that appeared earlier in another just increases the
incompatibility of the species. This doesn't change things much since they were
incompatible to begin with. On the other hand, if the same species adds a
connection that it added in an earlier generation, that must mean some members of
the species had not adopted that connection yet...so now it is likely that the
first "version" of that connection that starts being helpful will win out, and
the other will die away. The third case is where a connection has already been
generally adopted by a species. In that case, there can be no mutation creating
the same connection in that species since it is already taken. The main point is,
you don't really expect too many truly similar structures with different markings
to emerge, even with only keeping the record around for 1 generation.
Which way works best is a good question. If you have any interesting experimental
results on this question, please let me know.
My third revision will allow both options. I will add more information to this answer when I have results about it.

Creating a linked list using CUDA

Is it possible to create a linked list on a GPU using CUDA?
I am trying to do this and I am encoutering some difficulties.
If I can't allocate dynamic memory in a CUDA kernel, then how can I create a new node and add it to the linked list?
You really don't want to do this if you can help it - the best thing you can do if you can't get away from linked lists is to emulate them via arrays and use array indices rather than pointers for your links.
There are some valid use cases for linked lists on a GPU. Consider using a Skip List as an alternative as they provide faster operations. There are examples of highly concurrent Skip List algorithms available via Google searches.
Check out this link http://www.cse.iitk.ac.in/users/mainakc/lockfree.html/
for CUDA code a PDF and PPT presentation on a number of lock free CUDA data structures.
Link Lists can be constructed in parallel using a reduction algorithm approach. This assumes that ALL members are known at construction time. Each thread starts by connecting 2 nodes. Then half the threads connect the 2 node segments together and so on, reducing the number threads by 2 each iteration. This will build a list in log2 N time.
Memory allocation is a constraint. Pre-allocate all the nodes in an array on the host. Then you can use array subscripts in place of pointers. That has the advantage that the list traversal is valid on the GPU and the host.
For concurrency you need to use CUDA atomic operations. Atomic add/increment to count the nodes used from the node array and Compare and Swap to to set the links between nodes.
Again carefully consider the use case and access patterns. Using one large linked list is very serial. Using 100 - 100's of small Linked list is more parallel. I expect the memory access be uncoalesced unless care is taken to allocate connected nodes in adjacent memory locations.
I agree with Paul, linked lists are a very 'serial' way of thinking. Forget what you've learned about serial operations and just do everything at once : )
take a look at Thrust for the way of doing common operations

Resources