WenheLI
WenheLI
Thanks for your excellent work! I got an error of `AttributeError: module 'tensorflow' has no attribute 'get_default_graph` when I try to run your code. It seems to be a dependency...
Right now, our pipeline is bound with a fixed set of hyper-parameter and this is not convenient to run multi-job with different sets of hyper-parameter. We should allow passing hyper-parameter...
We should track the building, execution, and installing time for every release and visualize them. This could give us a clear clue on what is improved and what could be...
After investigation on tvm, I found that we could use tvm to tune a model & export it to a wasm executable in both nodejs and browsers. Later, I will...
In our implementation right now, we apply `dataProcess` to every single sample despite what the batch size is. If we can apply the `dataProcess` at the batch level, we can...
As the project becomes more complexed, we might need to figure out some improvements on developing experience. Below is something comes into my mind. - [ ] Unit test for...
To develop a text summary model, three key plugins are required `modelDefine`, `modelTrain`, and `modelEval`. - `modelDefine`: this plugin requires you to write the definition of a model (mainly network...
While developing plugins, I found that the functionalities of `dataProcess` & `dataAccess` have some overlaps since they both functioned for `data manipulation` with different scopes (`dataProcess` for one sample &...
We should support/add unit test for plugins to ensure the contributed plugins work well and leverage plugin developers' workload. - For `data process`, `data collect`, and `data access`, we can...
Currently, `pipboard` can display numerical data(accuracy, recall rate, and losses). However, most model evaluation process requires images (I.E. generated image in GAN model) or tables (I.E. Confusion Matrix in Classification...