forked from intel-machine-learning/DistML
-
Notifications
You must be signed in to change notification settings - Fork 0
/
CHANGES.txt
28 lines (22 loc) · 1.03 KB
/
CHANGES.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
[2016-01-18]
1. Reimplemented data transfer with pure socket
[2016-01-08]
1. Implemented LightLDA and MLR, now LR/LightLDA/MLR all works fine
2. Add api support for Double Array/Matrix and Int Array/Matrix
3. Add scala api for DoubleArray
[2015-12-14]
1. Simplify APIs, allow users to use naive data for customized algorithms
2. With API change, only LR and LDA are remained to proove API is working well
3. Run examples on yarn and can use spark submit
[2015-10-15]
1. Add environment variable "PS_NETWORK_PREFIX", which is used by parameter servers
2. Fix issue in handling messages after iteration done
[2015-10-12]
1. Add serialization support for CNN classes
[2015-08-21]
1. add draft lda support
2. support iterative UDF in serverside
[2015-07-01]
1. removed feature of worker group to reduce code complexity, a new branch "workergrou" is created
2. use akka-io for data transport, which allows big data block transferring
3. start parameter servers in a separated thread, allows iterative training without resparting parameter servers