Wu Yi

Model zoo TODO list

1

We are implementing the model zoo currently. Here is the [overall design](https://github.com/sql-machine-learning/sqlflow/blob/develop/doc/design/model_zoo.md) and [how users contribute custom models under this design](https://github.com/sql-machine-learning/sqlflow/blob/develop/doc/design/contribute_models_new.md) - [x] Model zoo server @typhoonzero - [x] GRPC...

As a compiler, SQLFlow compiles a SQL program to target ML code. We can do more to leverage the power of compiler: - Compile the whole SQL program to a...

Implement support for mathematical programming in SQLFlow

1

Please check out the design doc before starting to contribute: https://github.com/sql-machine-learning/sqlflow/blob/develop/doc/design/programming_solver.md TODO: - [x] Add parser support to parse the programming extended syntax - [x] Add IR struct to represent...

[Discussion] two ways of implementing Python API and connect to Go parser

6

## proposal one Python API: ```python import sqlflow sqlflow.train("SELECT ...", feature_columns=[...], attrs={...}, engine="pai") sqlflow.predict(...) sqlflow.explain(...) sqlflow.evaluate(...) ``` Go code using code generator to generate python code: ```go train_template=`import sqlflow feature_columns...

/kind bug **What steps did you take and what happened:** Following https://github.com/kserve/kserve/blob/master/docs/samples/outlier-detection/alibi-detect/cifar10/cifar10_outlier.ipynb, the sample model `gs://kfserving-examples/tfserving/cifar10/resnet32` mentioned does not exist.

Unify the docker images build method

1

We now have the following docker images need to be built to run on the cluster: - EDL controller - Cloud server - PFS server - Cloud job runtime image...

paddlecloud logs jobname error

4

paddle cloud logs: ``` [ERROR 2017-08-24 02:51:17,126 @ 2976] - Internal Server Error: /api/v1/logs/ Traceback (most recent call last): File "/usr/local/lib/python2.7/site-packages/django/core/handlers/exception.py", line 41, in inner response = get_response(request) File "/usr/local/lib/python2.7/site-packages/django/core/handlers/base.py",...

It's hard to find pserver fail logs for now. We can add a API calling kubectl logs -p to get previous log or just turn pserver ReplicaSet to a Job.

Current HDFS cluster usage does not need to login into a server to prepare training data.

Running fault tolernet jobs encounter panic: context deadline exceeded

1

running with case1 at: https://github.com/PaddlePaddle/cloud/issues/399 ``` label selector: paddle-job-pserver=mnist5, desired: 10 label selector: paddle-job-master=mnist5, desired: 1 Starting training job: /pfs/mulan/home/[email protected]/jobs/mnist5, num_gradient_servers: 5, version: I1102 03:27:33.970584 30 Util.cpp:166] commandline: --num_gradient_servers=5 --ports_num_for_sparse=1...