These are chat archives for beniz/deepdetect

19th
May 2017
sunatthegilddotcom
@sunatthegilddotcom
May 19 2017 00:18
@beniz , I am trying to following the training instruction on https://www.deepdetect.com/tutorials/train-imagenet/ to train the imagenet data.

@beniz
curl -X PUT "http://localhost:8080/services/imageserv" -d '{
"mllib":"caffe",
"description":"image classification service",
"type":"supervised",
"parameters":{
"input":{
"connector":"image",
"width":224,
"height":224
},
"mllib":{
"nclasses":1000
}
},
"model":{
"templates":"/opt/deepdetect/templates/caffe",
"repository":"/opt/models/ggnet"
}
}'

 curl -X POST "http://localhost:8080/train" -d '{
   "service":"imageserv",
   "async":true,
   "parameters":{
     "mllib":{
       "gpu":true,
       "net":{
         "batch_size":32
       },
       "solver":{
         "test_interval":500,
         "iterations":30000,
         "base_lr":0.001,
         "stepsize":1000,
         "gamma":0.9
       }
     },
     "input":{
       "connector":
       "image",
       "test_split":0.1,
       "shuffle":true,
       "width":224,
       "height":224
     },
     "output":{
       "measure":["acc","mcll","f1"]
     }
   },
   "data":["/opt/deepdetect/ilsvrc12"]
 }'
 on GPU.  After many iteration, I did not see the improvement .  Please see the result below.   Is there anything I need to do to improve the result. {"status":{"code":200,"msg":"OK"},"head":{"method":"/train","job":1,"status":"running","time":80539.0},"body":{"measure":{"iter_time":490.0,"mcll":2.506864451733095,"train_loss":5.627640724182129,"iteration":26000.0,"precision":0.4781695891074822,"recall":0.5783638460336952,"f1":0.5235158555594616,"acc":0.5077821205159571,"accp":0.5077821205159571}}}

{"status":{"code":200,"msg":"OK"},"head":{"method":"/train","job":1,"status":"running","time":80804.0},"body":{"measure":{"iter_time":584.0,"mcll":2.506864451733095,"train_loss":5.367936134338379,"iteration":26500.0,"precision":0.4781695891074822,"recall":0.5783638460336952,"f1":0.5235158555594616,"acc":0.5077821205159571,"accp":0.5077821205159571}}}

Thank you very much

roysG
@roysG
May 19 2017 00:38
How can i load the server 'dede' when my the ubuntu is start?
Emmanuel Benazera
@beniz
May 19 2017 05:21
@roysG see beniz/deepdetect#79
@sunatthegilddotcom are you using the full Imagenet dataset or just a few classes for testing ?
sunatthegilddotcom
@sunatthegilddotcom
May 19 2017 17:12
@beniz I am using the full Imagenet dataset. Thank you very much
@beniz I increased the iteration to 200,000 in the training call after the initial 30,000 training. It appears precision is gradually improved. It is just very slow converged.
sunatthegilddotcom
@sunatthegilddotcom
May 19 2017 17:18
@beniz here is "precision":0.5388656551244995," after additional 15000 iteration. Thanks.
Emmanuel Benazera
@beniz
May 19 2017 18:08
your learning is low, you may want to start around 0.01
sunatthegilddotcom
@sunatthegilddotcom
May 19 2017 23:48
@beniz thank you very much for your help, you mean I need to change "base_lr":0.001, to "base_lr":0.01"?