Backported serialization to python 3.6

ros1_roboy/Dockerfile

@@ -6,13 +6,12 @@ WORKDIR /ros1
 COPY . .
 
 RUN apt update
-RUN apt install python3.7 python3.7-dev ffmpeg iputils-ping -y
+RUN apt install ffmpeg iputils-ping -y
 
 RUN git clone https://github.com/Roboy/sonosco.git
-RUN cd sonosco; git checkout demo; python3.7 -m pip install -e .
-RUN python3.7 -m pip install webrtcvad
+RUN cd sonosco; git checkout develop; pip3 install .
 RUN chmod +x STT_server.py
 RUN chmod +x STT_client.py
-#RUN .  ~/melodic_ws/devel/setup.bash
+#RUN source ~/melodic_ws/devel/setup.bash
 #RUN roscore &
 #ENTRYPOINT [ "bash", "-c", "source /opt/ros/melodic/setup.bash; python3 STT_srv.py" ]

ros1_roboy/STT_server.py

@@ -4,7 +4,7 @@
 
 from sonosco.models import DeepSpeech2
 
-from sonosco.models.deepspeech2_inference import DeepSpeech2Inference
+from sonosco.inference.deepspeech2_inference import DeepSpeech2Inference
 from sonosco.ros1.server import SonoscoROS1
 from roboy_cognition_msgs.srv import RecognizeSpeech
 from roboy_control_msgs.msg import ControlLeds
@@ -14,7 +14,7 @@
 
 model_path = "pretrained/deepspeech_final.pth"
 
-asr = DeepSpeech2Inference(DeepSpeech2.load_model(model_path))
+asr = DeepSpeech2Inference(model_path)
 leave = False
 got_a_sentence = False
 

setup.py

@@ -20,9 +20,10 @@
     include_package_data=True,
     dependency_links=[],
     classifiers=[
+        "Programming Language :: Python :: 3.6",
         "Programming Language :: Python :: 3.7",
         "License :: OSI Approved :: BSD License",
     ],
-    python_requires='>=3.7',
-    install_requires = required
+    python_requires='>=3.6',
+    install_requires=required
 )

sonosco/inference/dummp_asr.py

@@ -3,9 +3,11 @@
 
 class DummyASR(SonoscoASR):
 
-
     def __init__(self) -> None:
-        super().__init__(None)
+        super().__init__("")
 
     def infer(self, sound_bytes):
         return "dummy transcript"
+
+    def infer_from_path(self, path: str) -> str:
+        return "dummy transcript"

sonosco/model/serialization.py

@@ -3,12 +3,13 @@
 from sonosco.common.constants import CLASS_MODULE_FIELD, CLASS_NAME_FIELD, SERIALIZED_FIELD
 from dataclasses import _process_class, _create_fn, _set_new_attribute, fields
 import typing
+from typing import List, Set, Tuple, Dict
 from torch import nn
 import torch
 
 __primitives = {int, float, str, bool}
 # TODO support for dict with Union value type
-__iterables = [list, set, tuple, dict]
+__iterables = [list, set, tuple, dict, List, Set, Tuple, Dict]
 
 
 # TODO: Prevent user from serializing lambdas.
@@ -115,7 +116,7 @@ def __create_serialize_body(fields_to_serialize: typing.Iterable, model: bool, e
         body_lines.append(f"    {c.name} = {{")
         __encode_serializable_serialization(body_lines, c)
         body_lines.append(f"}}")
-        body_lines.append(f"else: raise TypeError(\"Callable must be a function for now\")")
+        body_lines.append(f"else: raise TypeError(f\"Callable must be a function for now: {c.name}\")")
 
     for field in callable_iterables:
         body_lines.append(f"{field.name} = []")
@@ -275,4 +276,6 @@ def __is_callable(obj: any) -> bool:
         objs = list(obj.__args__)
     else:
         objs = [obj]
-    return all([hasattr(obj, '__origin__') and obj.__origin__ == collections.abc.Callable for obj in objs])
+    return all([hasattr(obj, '__origin__') and (
+            obj.__origin__ == collections.abc.Callable or obj.__origin__ == typing.Callable)
+                for obj in objs])

tests/test_model_trainer_serialization.py

@@ -20,6 +20,7 @@
 SOS = '#'
 PADDING_VALUE = '%'
 
+
 def test_mode_trainer_serialization():
     config_path = "model_trainer_config_test.yaml"
     config = parse_yaml(config_path)["train"]
@@ -60,4 +61,5 @@ def test_mode_trainer_serialization():
         'test_data_loader': test_loader,
     }, with_config=True)
     assert trainer_deserialized is not None
-    assert deserialized_config == config
+    assert deserialized_config == config
+