feat: #112 model singleton #136
Conversation
| app.state.primary = args.primary | ||
| app.state.reference = args.reference | ||
| model = Model(primary_dataset_name=args.primary, reference_dataset_name=args.reference) | ||
| app = create_app(model, graphiql=config.graphiql) | ||
|
|
||
| uvicorn.run(app, port=args.port) |
There was a problem hiding this comment.
I figured out how to run with automatic reloads using uvicorn.run("main:app", reload=True, port=args.port), but this does not work now that we are creating app on the fly rather than importing a static app object. Haven't figured out how to get automatic server reloads working with the current setup.
src/phoenix/core/model.py
Outdated
| def _infer_dimension_data_type(self, dimension_name: str) -> DimensionDataType: | ||
| # TODO: verify corresponding dimension of reference dataset has same type | ||
| dataframe_dtype = self.primary_dataset.dataframe[dimension_name].dtype | ||
| if is_numeric_dtype(dataframe_dtype): # type: ignore | ||
| return DimensionDataType.NUMERIC | ||
| elif is_object_dtype(dataframe_dtype): # type: ignore | ||
| return DimensionDataType.CATEGORICAL | ||
| raise ValueError("Unrecognized dimension type") |
There was a problem hiding this comment.
First crack at inferring dimension data type from pandas data type. The types on these methods are rough, ignoring for now.
app/schema.graphql
Outdated
| actual | ||
| feature | ||
| tag | ||
| embedding |
There was a problem hiding this comment.
we might want to nix this. Not important for now
src/phoenix/core/model.py
Outdated
| for name in primary_actual_column_names: | ||
| if name is not None: | ||
| dimensions.append( | ||
| Dimension( | ||
| name=name, | ||
| data_type=self._infer_dimension_data_type(name), | ||
| type=DimensionType.ACTUAL, | ||
| ) | ||
| ) |
There was a problem hiding this comment.
this bit of code is repeated quite a bit, there's gonna be a way for you to merge the schemas, and to infer the types in one go. Not urgent but I think this code can read a bit cleaner. We can pair on this as a follow-up
There was a problem hiding this comment.
Took a crack at condensing this, let me know if this looks better.
src/phoenix/core/model.py
Outdated
| primary_schema = primary_dataset.schema | ||
|
|
||
| # add actual dimensions | ||
| primary_actual_column_names = [ |
There was a problem hiding this comment.
Might consider deferring on this, though nice to have as a strawman
There was a problem hiding this comment.
Do you mean to remove actuals? Don't we want to compute cardinality on actual columns?
| from phoenix.core.model import Model | ||
| from phoenix.metrics.cardinality import cardinality | ||
|
|
||
| T = TypeVar("T") |
There was a problem hiding this comment.
Similar to comments below, when using Generics, let's just nix using single variables. There's no need make the type signature more complicated. T, like foo/bar, is commonly used in generics but hurts code readability. Here you can make this actually humanly understandable - it is the entity you are loading - e.g. DataItem or other. Naming along these lines is important as this code must be understandable by many.
There was a problem hiding this comment.
Simplified this section and no longer needed generics.
| return Loader(cardinality=cardinality_dataloader) | ||
|
|
||
|
|
||
| def _get_metric_dataloader( |
There was a problem hiding this comment.
Not sure if this is really worthy of an abstraction quite yet - the way it uses partial application of the function requires a distinct function signature. Not a big deal but I would caution against building abstractions when there is no need.
|
|
||
| from phoenix.core.model import Model | ||
|
|
||
| from .loader import Loader |
There was a problem hiding this comment.
@axiomofjoy can we use plurals here?
resolves #112
phoenix.core.modelsingleton to strawberry info.contextphoenix.metricsmodel.dimensionspandasdata types