Merge pull request #8 from InsightDataCommunity/dev-dualdocker-20190204

Dev dualdocker 20190204

Author: hundredblocks

.gitignore

@@ -1,6 +1,8 @@
 results/
 .vscode/
 
+src/app/models/SentimentV1/uncased_L-12_H-768_A-12/
+*.zip
 # Byte-compiled / optimized / DLL files
 .DS_store
 __pycache__/
@@ -14,6 +16,7 @@ __pycache__/
 
 # model files
 *.h5
+*.bin
 
 # Distribution / packaging
 .Python

README.md

@@ -1,7 +1,10 @@
 # Sherlock
 Sherlock is a web platform that allows user to create a image classifier for custom images, based on pre-trained CNN models. It also allows to use the customized CNN to pre-label images, and re-train the customized CNN when more training data become avaliable.
 
-Sherlock is currently serveing as RESTful APIs.
+Sherlock is currently serving as RESTful APIs.
+
+- [Sherlock for NLP](#sherlock-for-nlp)
+
 
 [Here](http://bit.ly/michaniki_demo) are the slides for project Sherlock (previously called Michaniki).
 
@@ -75,6 +78,10 @@ Move to the directory where you cloned *Sherlock* , and run:
 ```bash
 docker-compose up --build
 ```
+Training using BERT runs much faster on GPU with >12GB RAM (Tested with Nvidia K80). To train with GPU run:
+```bash
+docker-compose -f docker-compose-gpu.yml up --build
+```
 
 If everything goes well, you should start seeing the building message of the docker containers:
 ```
@@ -253,3 +260,66 @@ curl -X POST \
   -F train_bucket_prefix=S3_BUCEKT_PREFIX/sha
 ```
 
+## Sherlock for NLP
+
+### 1. Predict Sentiment of Text with Simple run_classifier
+```bash
+curl -X POST \
+  http://127.0.0.1:3031/sentimentV1/predict \
+  -H 'Cache-Control: no-cache' \
+  -H 'Postman-Token: eeedb319-2218-44b9-86eb-63a3a1f62e14' \
+  -H 'content-type: multipart/form-data; boundary=----WebKitFormBoundary7MA4YWxkTrZu0gW' \
+  -F textv='the movie was bad' \
+  -F model_name=base
+```
+
+### 2. Train a new classification model using pre-trained BERT model
+
+**The new text dataset should be stored at S3 first, the directory architecture in S3 should look like this**:
+```
+.
+├── YOUR_BUCKET_NAME
+│   ├── train.tsv
+│   ├── val.tsv
+│   ├── test.tsv		
+```
+The folder name you give to *YOUR_MODEL_NAME* will be used to identify this model once it get trained.
+
+The name of train, val and test files  **can't be changed**.
+The train and dev file should have below format (without header)-
+id label None Sentence
+1  0     NC   Text
+The test.tsv file should only have id and sentence column (with header)
+**The S3 folders should have public access permission**.
+
+To call this API, do:
+```bash
+curl -X POST \
+  http://127.0.0.1:3031/sentimentV1/trainbert \
+  -H 'Cache-Control: no-cache' \
+  -H 'Postman-Token: 4e90e1d6-de18-4501-a82c-f8a878616b12' \
+  -H 'content-type: multipart/form-data; boundary=----WebKitFormBoundary7MA4YWxkTrZu0gW' \
+  -F train_bucket_name=YOUR_BUCKET_NAME \
+  -F train_bucket_prefix=YOUR_MODEL_NAME
+```
+### 3. Lable all text in a csv file using pre-trained BERT model
+
+**The new test tsv file should be stored at the same S3 bucket as above for that model, directory architecture in S3 should look like this**:
+```
+.
+├── YOUR_BUCKET_NAME
+│   ├── train.tsv
+│   ├── val.tsv
+│   ├── test.tsv		
+```
+To call this API do:
+```bash
+curl -X POST \
+  http://127.0.0.1:3031/sentimentV1/testbert \
+  -H 'Cache-Control: no-cache' \
+  -H 'Postman-Token: 4e90e1d6-de18-4501-a82c-f8a878616b12' \
+  -H 'content-type: multipart/form-data; boundary=----WebKitFormBoundary7MA4YWxkTrZu0gW' \
+  -F test_bucket_name=YOUR_BUCKET_NAME \
+  -F test_bucket_prefix=YOUR_MODEL_NAME
+```
+At the end of prediction a file named 'test_results.csv' will be uploaded to the same S3 bucket.

docker-compose-gpu.yml

@@ -0,0 +1,99 @@
+version: '2.3'
+
+services:
+  michaniki_client:
+    build:
+      context: ./src
+      dockerfile: Dockerfile-gpu
+    runtime: nvidia
+    ports:
+      - "3031:3031"
+    environment:
+      - PORT=3031
+      - FLAS_APP=app/__init__.py
+      - FLASK_DEBUG=1
+      - REDIS_URL="redis://redis"
+      - REDIS_PORT=6379
+      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
+      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+
+    volumes:
+      - ./src:/opt/src
+
+    command: ./entryPoint.sh
+    depends_on:
+      - redis
+    networks:
+      - michaniki
+
+  inference_server:
+    build:
+      context: ./src
+      dockerfile: Dockerfile-gpu
+    runtime: nvidia
+    environment:
+      - REDIS_URL="redis://redis"
+      - REDIS_PORT=6379
+      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
+      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+      
+    command: ['python', 'app/models/InceptionV3/inception_inference_server.py']
+    volumes:
+      - ./src:/opt/src
+
+    networks:
+      - michaniki
+    depends_on:
+      - michaniki_client
+      - redis
+
+  sentiment_inference_server:
+    build:
+      context: ./src
+      dockerfile: Dockerfile-sentiment
+    runtime: nvidia
+    environment:
+      - REDIS_URL="redis://redis"
+      - REDIS_PORT=6379
+      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
+      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+    command: ['python', 'app/models/SentimentV1/sentiment_infer_server.py']
+    volumes:
+      - ./src:/opt/src
+    networks:
+      - michaniki
+    depends_on:
+      - michaniki_client
+      - redis
+
+  celery_worker:
+    build:
+      context: ./src
+      dockerfile: Dockerfile-gpu
+    runtime: nvidia
+    command: ['celery', '-A', 'app.celeryapp:michaniki_celery_app', 'worker', '-l', 'info']
+    volumes:
+      - ./src:/opt/src
+    networks:
+      - michaniki
+    depends_on:
+      - michaniki_client
+    environment:
+      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
+      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+      - DB_HOST=db
+      - DB_USERNAME=root
+      - DB_PASSWORD=michaniki
+      - DB_NAME=michanikidb
+      - BROKER_URL=redis://redis:6379/0
+     
+
+  redis:
+    image: redis:4.0.5-alpine
+    command: ["redis-server", "--appendonly", "yes"]
+    hostname: redis
+    networks:
+      - michaniki
+
+networks:
+    michaniki:

docker-compose.yml

@@ -1,4 +1,4 @@
-version: '3'
+version: '2.3'
 
 services:
   michaniki_client:
@@ -13,8 +13,10 @@ services:
       - REDIS_PORT=6379
       - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
       - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+
     volumes:
       - ./src:/opt/src
+
     command: ./entryPoint.sh
     depends_on:
       - redis
@@ -28,15 +30,35 @@ services:
       - REDIS_PORT=6379
       - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
       - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+      
     command: ['python', 'app/models/InceptionV3/inception_inference_server.py']
     volumes:
       - ./src:/opt/src
+
     networks:
       - michaniki
     depends_on:
       - michaniki_client
       - redis
-  
+
+  sentiment_inference_server:
+    build:
+      context: ./src
+      dockerfile: Dockerfile-sentiment
+    environment:
+      - REDIS_URL="redis://redis"
+      - REDIS_PORT=6379
+      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
+      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+    command: ['python', 'app/models/SentimentV1/sentiment_infer_server.py']
+    volumes:
+      - ./src:/opt/src
+    networks:
+      - michaniki
+    depends_on:
+      - michaniki_client
+      - redis
+
   celery_worker:
     build: ./src
     command: ['celery', '-A', 'app.celeryapp:michaniki_celery_app', 'worker', '-l', 'info']
@@ -54,14 +76,14 @@ services:
       - DB_PASSWORD=michaniki
       - DB_NAME=michanikidb
       - BROKER_URL=redis://redis:6379/0
-  
+     
+
   redis:
     image: redis:4.0.5-alpine
     command: ["redis-server", "--appendonly", "yes"]
     hostname: redis
     networks:
       - michaniki
-        
+
 networks:
     michaniki:
-    

src/Dockerfile-gpu

@@ -0,0 +1,21 @@
+FROM tensorflow/tensorflow:1.12.0-gpu
+
+# utils
+RUN apt-get update && apt-get install -y --no-install-recommends apt-utils
+
+#RUN conda install gxx_linux-64
+
+#RUN apt-get install -y --force-yes default-libmysqlclient-dev mysql-client build-essential
+
+# Grab requirements.txt.
+COPY requirements-gpu.txt /tmp/requirements-gpu.txt
+
+# Install dependencies
+RUN pip install -qr /tmp/requirements-gpu.txt
+
+# create a user for web server
+RUN adduser --disabled-password --gecos "" foo
+
+COPY ./ /opt/src
+
+WORKDIR /opt/src

src/Dockerfile-sentiment

@@ -0,0 +1,24 @@
+FROM continuumio/miniconda:4.4.10
+
+# utils
+RUN apt-get update && apt-get install -y --no-install-recommends apt-utils
+
+RUN conda install gxx_linux-64
+
+
+RUN apt-get install -y --force-yes default-libmysqlclient-dev mysql-client build-essential
+
+# Grab requirements.txt.
+COPY requirementssenti.txt /tmp/requirementssenti.txt
+
+# Install dependencies
+RUN pip install -qr /tmp/requirementssenti.txt
+
+RUN pip install fasttext
+
+# create a user for web server
+RUN adduser --disabled-password --gecos "" foo
+
+COPY ./ /opt/src
+
+WORKDIR /opt/src

src/__init__.py

@@ -69,7 +69,10 @@ def download_a_dir_from_s3(bucket_name, bucket_prefix, local_path):
                 os.makedirs(save_path)
             except OSError:
                 pass
-            mybucket.download_file(obj.key, os.path.join(save_path, filename))
+            try:
+                mybucket.download_file(obj.key, os.path.join(save_path, filename))
+            except OSError:
+                pass  
 
     print "* Helper: Images Loaded at: {}".format(output_path)
     return output_path