Skip to content

Latest commit

 

History

History
45 lines (42 loc) · 1.72 KB

python_client.md

File metadata and controls

45 lines (42 loc) · 1.72 KB
Raw

Python client

For Python client example we'll use tfaas_client.py. Similar to Curl client use case we need to upload our model to TFaaS server. This can be done by creating upload.json file with our upload parameters:

{
  "model": "/path/tf_model.pb", "labels": "/path/labels.txt",
  "name": "myModel", "params":"/path/params.json"
}

It includes full path to our TF model, labels and parameters files as well as name of our TF model. Now we can run our python client:

# define url for TFaaS
url=http://localhost:8083

# upload our model
tfaas_client.py --url=$url --upload=upload.json

# view registered models in TFaaS server
tfaas_client.py --url=$url --models

Finally, we can ask for predictions by preparing input.json file which contains our keys (the list of names of our parameters), their values (the list of numerical values) and model name we want to use for inference, e.g.:

{"keys":["attr1", "attr2", ...], "values":[1,2,...], "name":"myModel"}

We can place the following call to get our predictions:

tfaas_client.py --url=$url --predict=input.json

When we deal with images, i.e. perform image classification, we use the following API:

# send request to classify given image (in this case we uploaded HEP_images model)
tfaas_client.py --url=$url --image=/path/file.png --model=HEP_images
# the server response will looks like this (here the labels contains
# three classes: higgs, qcd and muons)
{"filename":"/path/file.png","labels":[{"label":"higgs","probability":1},{"label":"qcd","probability":2.815438e-8},{"label":"muons","probability":4.65911e-18}]}