Enable WitWidget for AI Platform including xgboost models #2194
Conversation
|
@tolga-b please review |
| if 'adjust_prediction' in config else None) | ||
| self.compare_adjust_fn = ( | ||
| config.get('adjust_prediction_2') | ||
| if 'adjust_prediction_2' in config else None) |
There was a problem hiding this comment.
would be better if we are consistent with names (adjust_prediction_2 vs compare_adjust_fn) as in compare_custom_predict_fn.
| if 'custom_predict_fn' in copied_config: | ||
| del copied_config['custom_predict_fn'] | ||
| if 'compare_custom_predict_fn' in copied_config: | ||
| del copied_config['compare_custom_predict_fn'] | ||
| if 'adjust_prediction' in copied_config: |
There was a problem hiding this comment.
same comment about naming here.
| # functions. | ||
| if self.config.get('use_aip'): | ||
| self.custom_predict_fn = self._predict_aip_model | ||
| if self.config.get('use_aip_2'): |
There was a problem hiding this comment.
same comment about naming here.
| @@ -14,4 +14,4 @@ | |||
|
|
|||
| """Contains the version string.""" | |||
|
|
|||
| VERSION = '1.0.1' | |||
| VERSION = '1.0.2' | |||
There was a problem hiding this comment.
Bump this to 1.0.3 if you already pushed the previous PR to pip. Is the best practice having different version string for each PR?
There was a problem hiding this comment.
haven't pushed out 1.0.2. we should only bump version on PRs where we plan to push to pypi when they are submitted.
| feat = feature_names[i] | ||
| else: | ||
| feat = str(i) | ||
| if isinstance(json_ex[i], (int, long)): |
There was a problem hiding this comment.
nit, you can pull this code chunk into a function add_single_feature(key, value, ex) to avoid repetition.
| self.set_examples(self._add_feature_names_to_examples()) | ||
| return self | ||
|
|
||
| def _add_feature_names_to_examples(self): |
There was a problem hiding this comment.
Does this fail if I call set_feature_names twice as int(feat) is now casting feature name string to an integer?
There was a problem hiding this comment.
after discussion, made this logic simpler by having feature_names optionally provided with examples, so no need to ever convert existing tf.Examples to have the new feature name mapping. its all done at set_examples time now.
Continuing work on CMLE mode for WitWidget, which has been rebranded to AI Platform by Cloud. Added support for xgboost models in AI Platform as well.
Rename CMLE to AI Platform serving
Allow examples to be provided as JSON lists (which is true for xgboost models), not just JSON dicts.
Allow separate provided of feature names for display when examples are provided as flat lists.
Add better prediction result parsing logic for AI Platform use-case, handling more models out of the box.
Add ability for user to specify custom prediction adjustment function for when the AI Platform model returns results in a format that is unknown to WIT.
Screenshots of UI changes
N/A
Detailed steps to verify changes work correctly (as executed by you)
Verified changes with two AI Platform models in a colab, including one that uses a custom prediction adjustment function.