| layout | title |
|---|---|
default |
Seldon Algorithm Configuration |
Algorithm configuration is added via ZooKeeper as per the configuration page.
Provided below is a default set of algorithm configurations to add. Put this at /config/default_strategy
{% highlight javascript %} {"algorithms":[{"name":"mfRecommender","tag":"MATRIX_FACTOR","includers":["recentItemsIncluder"],"filters":[],"config":[{"name":"io.seldon.algorithm.inclusion.itemsperincluder","value":200}],{"name":"recentItemsRecommender","tag":"RECENT_ITEMS","includers":[]}],"combiner":"firstSuccessfulCombiner"} {% endhighlight %}
This provides a base set of algorithms. If you want to alter this then please refer to the configuration page.
The table below shows the nodes in zookeeper where the list of clients using a model is specified and where the location of a model for a particular client is specified.
| model | client list node | model location node |
|---|---|---|
| matrix factorization | /config/mf |
/all_clients/<client_name>/mf |
| word2vec | /config/word2vec |
/all_clients/<client_name>/word2vec |
| user clusters | /config/usercluster |
/all_clients/<client_name>/usercluster |
The output of the item similarity job is presently placed directly in the db so is not controlled in this way.
To activate a matrix factorization client for client test1 ensure the client name is added e.g.,
/config/mf => some_other_client1,test1,some_other_client2
The seldon server will look for the model location in:
/all_clients/test1/mf