Update to the latest version of hydrosdk (#6)

* Update to the latest version of hydrosdk (2.3.2) to fix automatic numpy conversion

* Allow returning TensorProto as well as Numpy arrays/scalars in a dict from a runtime executable

* add usage example to the README file

README.md

@@ -1,10 +1,44 @@
 # hydro-serving-python
 Python runtime for [Hydrosphere Serving](https://github.com/Hydrospheredata/hydro-serving).
-Provides GRPC API for a Python scripts.
-Supported versions are: python-3.4 python-3.5 python-3.6
+Provides a GRPC API for Python scripts.
+
+Supported versions are: python-3.6 python-3.7 python-3.8 
 
 ## Build commands
 - `make test`
 - `make python` - build docker runtime with python:latest-alpine base image
 - `make python-${VERSION}` - build docker runtime with python:${VERSION}-alpine base image
 - `make clean` - clean repository from temp files
+
+## Usage
+
+This runtime uses `src/func_main.py` script as an entry point.
+You may create any arbitrary Python application within,
+just keep in mind that the entry point of your script has to be located in
+   `src/func_main.py`.
+
+
+Example of a `func_main.py`:
+
+```python
+import pandas as pd
+from joblib import load
+
+# Load an ML model during runtime initialisation
+clf = load('/model/files/classification_model.joblib')
+
+# This function is called on each request
+# Input and output must comply with your model's signature 
+def predict(**kwargs):
+    # kwargs is a dict with Numpy arrays or scalars you've specified in a signature
+    x = pd.DataFrame.from_dict({"request": kwargs}).T
+    predicted = clf.predict(x)
+    return {"income": int(predicted)}
+```
+
+or if you wish to work with proto messages:
+```python
+    return {"income": TensorProto(int_val=[int(predicted)],
+                                  dtype=DT_INT32,
+                                  tensor_shape=TensorShapeProto())}
+```

requirements.txt

@@ -1,8 +1,7 @@
 grpcio~=1.19
-hydro-serving-grpc==2.1.0rc1
+hydro-serving-grpc==2.2.1
 docker~=3.0
 pytest
-hydrosdk==2.3.1
-
+hydrosdk==2.3.2
 pandas~=1.0.3
 numpy~=1.18.3

src/PythonRuntimeService.py

@@ -5,9 +5,9 @@
 
 import grpc
 import hydro_serving_grpc as hs
-import numpy as np
+from hydro_serving_grpc import TensorProto
 from hydro_serving_grpc.tf.api import PredictionServiceServicer
-from hydrosdk.data.conversions import tensor_proto_to_nparray, nparray_to_tensor_proto
+from hydrosdk.data.conversions import tensor_proto_to_np, np_to_tensor_proto
 
 
 class PythonRuntimeService(PredictionServiceServicer):
@@ -46,11 +46,15 @@ def Predict(self, request, context):
         else:
             self.logger.info("Received inference request: {}".format(request)[:256])
             try:
-                numpy_request_inputs: Dict[str, np.array] = {k: tensor_proto_to_nparray(t) for k, t in request.inputs.items()}
-                numpy_outputs: Dict[str, np.array] = self.executable(**numpy_request_inputs)
+
+                numpy_request_inputs: Dict[str] = {k: tensor_proto_to_np(t) for k, t in request.inputs.items()}
+                numpy_outputs: Dict[str] = self.executable(**numpy_request_inputs)
 
                 try:
-                    tensor_proto_outputs = {k: nparray_to_tensor_proto(v) for k, v in numpy_outputs.items()}
+                    # If TensorProto is returned, than pass it. If Numpy is returned, cast it to TensorProto
+                    tensor_proto_outputs: Dict[str, TensorProto] = {k: (v if isinstance(v, TensorProto) else np_to_tensor_proto(v))
+                                                                    for k, v in numpy_outputs.items()}
+
                     result = hs.PredictResponse(outputs=tensor_proto_outputs)
                     self.logger.info("Answer: {}".format(result)[:256])
                     return result

version

@@ -1 +1 @@
-2.3.1
+2.3.2