brightcoder01
brightcoder01
##### SQLFlow Built-in TSFresh High-Level Python Api ```python def add_extracted_ts_features( iterator, column_id, column_time, column_values, windows, extract_setting): """ Extract the features from the input containing the time series data and then...
MaxCompute SQL: ```SQL SELECT app_id, time, uv, LAG(uv, 1) OVER (PARTITION BY app_id ORDER BY time) AS uv_1, LAG(uv, 2) OVER (PARTITION BY app_id ORDER BY time) AS uv_2, pv,...
We will add `TO RUN` statement in SQLFlow. #2355 We can wrap the logic of convert the data of sliding windows to rows into python program and build it into...
> How about `TO RUN a_docker_registry/sqlflow-run:v1/binning.py` I think it's a good idea. It's aligned with the model training statement.
Current ```SQL SELECT * FROM sqlflow_tutorial_sample_binning TO RUN a_docker_registry/sqlflow-run:binning-v0.1.0 CMD "binning.py", "column=amount", "bin_method=log_bucket", "bin_num=10" INTO sqlflow_tutorial_statistics,sqlflow_tutorial_binned_prob,sqlflow_tutorial_binned_cumsum_prob; ``` Proposed from @typhoonzero (More aligned with TO TRAIN statement) ```SQL SELECT * FROM...
**The key challenge is: how to save the preprocess logic into the serialized model for serving (TorchScript or ONNX).** Proposal Options: - Develop some custom some PyTorch OP to do...
Additional Options: - Write customized ops to process the data of string type (such as hash_bucket, lookup_vocabulary). TorchScript can support the string type. Export the preproc into TorchScript. - Use...
> Can we write libtorch dataset transform functions to achieve this? I'm afraid that the transform logic in Dataset cannot be saved to model and used in serving.
> The feature derivation stage may generate: > > * the TensorFlow/XGBoost feature column codes, or, > * the transforming DAG as described in [this design doc](https://github.com/sql-machine-learning/elasticdl/blob/develop/docs/designs/data_transform.md#expand-the-transformop). Add one point:...
> Reasonable. In previous discussions with @Yancey1989 and @shendiaomo, we've been talked about this. > > Yet, We can somehow derive the feature columns in compile-time. > > ```sql >...