Integrate AWS Chalice

.github/labeler.yml

@@ -1,9 +1,16 @@
 # Add 'docs' to any changes within 'docs' folder or any subfolders
+# https://stackoverflow.com/questions/34691809/regex-match-folder-and-all-subfolders
 documentation:
 - docs/**/*
 
 example:
-- example/**/*
+- examples($|/.*)
 
 test:
 - tests/**/*
+
+serve:
+- chitra/serve($|/.*)
+
+cli:
+- chitra/cli($|/.*)

.github/workflows/main.yml

@@ -1,4 +1,4 @@
-name: Python Main CI
+name: pytest
 on:
   push:
     branches: [ master ]
@@ -12,6 +12,11 @@ jobs:
     strategy:
       matrix:
         os: [ ubuntu-latest, macos-latest ]
+        include:
+          - os: ubuntu-latest
+            path: ~/.cache/pip
+          - os: macos-latest
+            path: ~/Library/Caches/pip
     env:
       OS: ${{ matrix.os }}
       PYTHON: '3.7'
@@ -27,6 +32,15 @@ jobs:
         with:
           python-version: 3.7
 
+      - name: Cache pip
+        uses: actions/cache@v2
+        with:
+          path: ${{ matrix.path }}
+          key: ${{ runner.os }}-pip-${{ hashFiles('**/requirements.txt') }}
+          restore-keys: |
+            ${{ runner.os }}-pip-
+            ${{ runner.os }}-
+
       - name: Installation
         run: |
           python --version
@@ -35,6 +49,7 @@ jobs:
           pip install ".[serve]"
           pip install ".[converter]"
           pip install ".[test]"
+          pip install ".[example]"
           pip list
         shell: bash
 

.gitignore

@@ -138,3 +138,7 @@ checklink/cookies.txt
 
 # .gitconfig is now autogenerated
 .gitconfig
+
+*.pth
+*.ckpt
+*.h5

Makefile

@@ -1,6 +1,4 @@
-SRC = $(wildcard ./*.ipynb)
-
-all: chitra docs
+.PHONY: build_docs clean style build pypi
 
 build_docs:
 	cp README.md docs/index.md
@@ -18,7 +16,7 @@ coverage:  ## Run tests with coverage
 		coverage xml
 
 clean:
-	rm -rf dist
+	rm -rf dist/
 	find . -type f -name "*.DS_Store" -ls -delete
 	find . | grep -E "(__pycache__|\.pyc|\.pyo)" | xargs rm -rf
 	find . | grep -E ".pytest_cache" | xargs rm -rf
@@ -29,10 +27,10 @@ style:
 	black chitra tests examples
 	isort chitra tests examples
 
-dist: clean
+build: style clean
 	flit build
 
-pypi: dist
+pypi: build
 	flit publish
 
 push:

chitra/__init__.py

@@ -1,4 +1,4 @@
 """Deep Learning library for Model Building, Interpretability, Visualization, API Building & Deployment."""
 
-__version__ = "0.1.1"
+__version__ = "0.2.0a1"
 __license__ = "Apache License 2.0"

chitra/image.py

@@ -1,3 +1,4 @@
+import io
 import os
 from io import BytesIO
 from pathlib import Path
@@ -89,6 +90,9 @@ def _load_image(data: DATA_FORMATS, cache: bool):
         if isinstance(data, Image.Image):
             return data
 
+        if isinstance(data, bytes):
+            return Image.open(io.BytesIO(data))
+
         if isinstance(data, (tf.Tensor, torch.Tensor)):
             data = data.numpy().astype("uint8")
 
@@ -147,6 +151,19 @@ def draw_boxes(
             self.numpy()[..., :3], color=color, size=marker_size
         )
 
+    def resize(self, *args, **kwargs) -> Image.Image:
+        """
+        Calls PIL.Image.resize method and passes the arguments
+        Args:
+            *args:
+            **kwargs:
+
+        Returns:
+            resized PIL.Image
+        """
+        self.image = self.image.resize(*args, **kwargs)
+        return self.image
+
     def resize_image_with_bbox(self, size: List[int]):
         old_size = self.shape
         self.image = self.image.resize(size)

chitra/serve/__init__.py

@@ -1,3 +1,3 @@
 from chitra.serve.api import API, create_api
 from chitra.serve.app import GradioApp
-from chitra.serve.model_server import ModelServer
+from chitra.serve.base import ModelServer

chitra/serve/api.py

@@ -5,8 +5,8 @@
 
 from chitra.__about__ import documentation_url
 from chitra.serve import schema
+from chitra.serve.base import ModelServer
 from chitra.serve.constants import IMAGE_CLF, OBJECT_DETECTION, QNA, TXT_CLF
-from chitra.serve.model_server import ModelServer
 
 
 class API(ModelServer):

chitra/serve/app.py

@@ -5,11 +5,14 @@
 
 from chitra.__about__ import documentation_url
 from chitra.serve import constants as const
-from chitra.serve.model_server import ModelServer
+from chitra.serve.base import ModelServer
 
 
 class GradioApp(ModelServer):
-    API_TYPES = {"VISION": (const.IMAGE_CLF, const.OBJECT_DETECTION)}
+    API_TYPES = {
+        "VISION": (const.IMAGE_CLF, const.OBJECT_DETECTION),
+        "NLP": (const.TXT_CLF,),
+    }
 
     def __init__(
         self,
@@ -47,8 +50,12 @@ def setup(
         self,
         **kwargs,
     ):
-
-        self.api_type_func[const.IMAGE_CLF] = self.image_classification
+        if self.api_type in (const.IMAGE_CLF, const.OBJECT_DETECTION):
+            self.api_type_func[self.api_type] = self.single_x_classification
+        elif self.api_type == const.TXT_CLF:
+            self.api_type_func[self.api_type] = self.single_x_classification
+        else:
+            raise NotImplementedError(f"api_type={self.api_type} not implemented yet!")
 
         if not self.input_types:
             self.input_types = self.get_input_type(**kwargs)
@@ -67,7 +74,7 @@ def get_input_type(self, **kwargs):
             )
         raise NotImplementedError(f"{self.api_type} API Type is not implemented yet!")
 
-    def image_classification(self, x: np.ndarray):
+    def single_x_classification(self, x: np.ndarray):
         data_processor = self.data_processor
 
         if data_processor.preprocess_fn:

chitra/serve/model_server.py → chitra/serve/base.py

@@ -1,3 +1,4 @@
+import abc
 import itertools
 from typing import Callable, List, Optional
 
@@ -21,10 +22,19 @@ def __init__(
         model: Callable,
         preprocess_fn=None,
         postprocess_fn=None,
+        preprocess_conf: Optional[dict] = None,
+        postprocess_conf: Optional[dict] = None,
         **kwargs,
     ):
+        if not preprocess_conf:
+            preprocess_conf = {}
+        if not postprocess_conf:
+            postprocess_conf = {}
+
         self.api_type = api_type.upper()
         self.model = model
+        self.preprocess_conf = preprocess_conf
+        self.postprocess_conf = postprocess_conf
         self.data_processor: Optional[DataProcessor] = self.set_data_processor(
             preprocess_fn, postprocess_fn
         )
@@ -50,11 +60,8 @@ def set_default_processor(self) -> DataProcessor:
         elif api_type in ModelServer.API_TYPES.get("NLP"):
             self.data_processor = DefaultTextProcessor.nlp
         else:
-            raise UserWarning(
+            raise NotImplementedError(
                 f"{api_type} is not implemented! Available types are -\
                  {ModelServer.get_available_api_types()}"
             )
         return self.data_processor
-
-    def run(self, *_, **__):
-        raise NotImplementedError

chitra/serve/cloud/__init__.py

@@ -0,0 +1 @@
+from .aws_serverless import ChaliceServer

chitra/serve/cloud/aws_serverless.py

@@ -0,0 +1,68 @@
+from typing import Callable, List, Optional
+
+from chalice import Chalice, Rate
+
+from chitra.logging import logger
+from chitra.serve.cloud.base import CloudServer
+
+S3 = "s3"
+GCS = "gcs"
+
+RATE_UNIT = {"m": Rate.MINUTES, "h": Rate.HOURS, "d": Rate.DAYS}
+
+
+class ChaliceServer(CloudServer):
+    INVOKE_METHODS = ("route",)
+
+    def __init__(
+        self,
+        api_type: str,
+        model_path: str,
+        model_loader: Callable,
+        preprocess_fn: Callable = None,
+        postprocess_fn: Callable = None,
+        **kwargs,
+    ):
+        super().__init__(
+            api_type,
+            model_path=model_path,
+            model_loader=model_loader,
+            preprocess_fn=preprocess_fn,
+            postprocess_fn=postprocess_fn,
+            **kwargs,
+        )
+
+        self.app = Chalice(app_name=kwargs.get("name", "chitra-server"))
+
+    @staticmethod
+    def index():
+        return {"hello": "world"}
+
+    def predict(self) -> dict:
+
+        data_processor = self.data_processor
+        x = self.app.current_request.raw_body
+        logger.debug(f"raw body type={type(x)}")
+        if data_processor.preprocess_fn:
+            x = data_processor.preprocess(x, **self.preprocess_conf)
+        x = self.model(x)
+        if data_processor.postprocess_fn:
+            x = data_processor.postprocess(x, **self.postprocess_conf)
+        return x
+
+    def run(self, invoke_method: str, content_types: Optional[List] = None, **kwargs):
+        invoke_method = invoke_method.lower()
+        if not content_types:
+            content_types = []
+
+        if invoke_method not in self.INVOKE_METHODS:
+            raise NotImplementedError(
+                f"invoke method={invoke_method} not implemented yet. Please select {self.INVOKE_METHODS}"
+            )
+
+        if invoke_method == "route":
+            route_path = kwargs.get("path", "/predict")
+            self.app.route("/", methods=["GET"])(self.index)
+            self.app.route(route_path, methods=["POST"], content_types=content_types)(
+                self.predict
+            )

chitra/serve/cloud/base.py

@@ -0,0 +1,47 @@
+import abc
+import io
+from abc import ABC
+from typing import Callable, Optional
+
+import smart_open
+
+from chitra.serve.base import ModelServer
+
+
+class CloudServer(ModelServer, ABC):
+    def __init__(
+        self,
+        api_type: str,
+        model_path: str,
+        model_loader: Callable,
+        preprocess_fn: Optional[Callable] = None,
+        postprocess_fn: Optional[Callable] = None,
+        **kwargs
+    ):
+        raw_model = self.download_model(model_path, **kwargs)
+        model = model_loader(raw_model)
+
+        super().__init__(
+            api_type,
+            model,
+            preprocess_fn=preprocess_fn,
+            postprocess_fn=postprocess_fn,
+            **kwargs
+        )
+
+    @staticmethod
+    def download_model(path: str, **kwargs) -> io.BytesIO:
+        """
+        Download model from cloud
+        ref: http://5.9.10.113/67706477/load-pytorch-model-from-s3-bucket
+        Args:
+            path:
+            **kwargs:
+
+        Returns:
+
+        """
+
+        with smart_open.open(path, mode="rb", **kwargs) as fr:
+            data = io.BytesIO(fr.read())
+        return data

docs/index.md

@@ -42,7 +42,7 @@ Easily create UI for Machine Learning models or Rest API backend that can be dep
 
 ### Using pip (recommended)
 
-`pip install -U chitra==0.1.0`
+`pip install -U chitra`
 
 ### From source