Merge pull request #20 from A3Data/apigunicorn

Apigunicorn

README.md

@@ -40,6 +40,7 @@ This is also our way of reinforcing our position that women should be taking mor
 
 - Anaconda or Miniconda Python (>= 3.6)
 - conda (>= 4.8)
+- **docker**
 
 Hermione depends on conda to build and manage virtual conda environments. If you don't have it installed, please visit 
 <a href="https://www.anaconda.com/products/individual" target="_blank">Anaconda website</a> or 
@@ -61,9 +62,11 @@ After installed Hermione:
 hermione new project_hermione
 ```
 
-1. Enter “y” if you want to start with an example code
+1. Hit Enter if you want to start with an example code
 
-![](https://cdn-images-1.medium.com/max/800/1*TJoFVA-Nio2O3XvxBN4MUQ.png)
+```
+Do you want to start with an implemented example (recommended) [y/n]? [y]: 
+```
 
 3. Hermione already creates a conda virtual environment for the project. Activate it
 
@@ -81,9 +84,16 @@ pip install -r requirements.txt
 
 6. After that, a mlflow experiment is created. To verify the experiment in mlflow, type: mlflow ui. The application will go up.
 
-![](https://cdn-images-1.medium.com/max/800/1*DReyAtL9eJ0fiwxaVo3Yfw.png)
+```
+mlflow ui
+```
 
-7. To access the experiment, just enter the path previously provided in your preferred browser. Then it is possible to check the trained models and their metrics.
+    [2020-10-19 23:23:12 -0300] [15676] [INFO] Starting gunicorn 19.10.0
+    [2020-10-19 23:23:12 -0300] [15676] [INFO] Listening at: http://127.0.0.1:5000 (15676)
+    [2020-10-19 23:23:12 -0300] [15676] [INFO] Using worker: sync
+    [2020-10-19 23:23:12 -0300] [15678] [INFO] Booting worker with pid: 15678
+
+1. To access the experiment, just enter the path previously provided in your preferred browser. Then it is possible to check the trained models and their metrics.
 
 ![](https://cdn-images-1.medium.com/max/800/1*c_rDEqERZR6r8JVI3TMTcQ.png)
 
@@ -100,17 +110,42 @@ hermione predict
 Do you want to create your **project from scratch**? There click [here](tutorial_base.md) to check a tutorial.
 
 
-## Docker
+# Docker
 
-Hermione comes with a default `Dockerfile` that should work fine without editing if you have an autonomous `predict.py` file. It was designed to make batch predictions, not to serve the model with an API (coming soon...). To build correctly, you should be at the top project folder, not inside `src` folder.
+Hermione comes with a default `Dockerfile` which implements a Flask + Gunicorn API that serves your ML model. You should take a look at the `api/app.py` module and rewrite `predict_new()` function as you see fit.  
 
-You can build the docker image and run it with the following commands:
+Also, in the newest version, hermione brings two CLI commands that helps us abstract a little bit the complexity regarding docker commands. To build an image (remember you should have docker installed), you should be in the project's root directory. Than, do:
 
+```bash
+hermione build <IMAGE_NAME>
 ```
-docker build -f src/Dockerfile -t myprediction:latest .
-docker run --rm myprediction:latest
+
+After you have built you're docker image, run it with:
+
+```bash
+hermione run <IMAGE_NAME>
 ```
 
+    [2020-10-20 02:13:20 +0000] [1] [INFO] Starting gunicorn 20.0.4
+    [2020-10-20 02:13:20 +0000] [1] [INFO] Listening at: http://0.0.0.0:5000 (1)
+    [2020-10-20 02:13:20 +0000] [1] [INFO] Using worker: sync
+    [2020-10-20 02:13:20 +0000] [7] [INFO] Booting worker with pid: 7
+    [2020-10-20 02:13:20 +0000] [8] [INFO] Booting worker with pid: 8
+    [2020-10-20 02:13:20 +0000] [16] [INFO] Booting worker with pid: 16
+
+**THAT IS IT!** You have a live model up and running. To test your API, hermione provides a `api/myrequests.py` module. *This is not part of the project*; it's a "ready to go" code to make requests to the API. Help yourself!
+
+```bash
+cd src/api
+python myrequests.py
+```
+
+    Sending request for model...
+    Data: {"Pclass": [3, 2, 1], "Sex": ["male", "female", "male"], "Age": [4, 22, 28]}
+    Response: "[0.24630952 0.996      0.50678968]"
+
+Play a little with the 'fake' data and see how far can the predictions go.
+
 
 ## Documentation
 This is the class structure diagram that Hermione relies on:

hermione/_version.py

@@ -1 +1 @@
-__version__ = '0.2'
+__version__ = '0.3'

hermione/cli.py

@@ -30,8 +30,8 @@ def info():
 @cli.command()
 @click.argument('project_name')
 @click.option('-p', '--python-version', 'python_version', default='3.7', show_default=True)
-@click.option('-imp', '--implemented', 'implemented', prompt='Do you want to start with an implemented example? [y/n] Default:', 
-            default='n', show_default=True)
+@click.option('-imp', '--implemented', 'implemented', prompt='Do you want to start with an implemented example (recommended) [y/n]?', 
+            default='y', show_default=True)
 def new(project_name, python_version, implemented):
     """
     Create a new hermione project
@@ -63,14 +63,13 @@ def new(project_name, python_version, implemented):
     # Write config file
     write_config_file(LOCAL_PATH, project_name)
     #write_logging_file(LOCAL_PATH, project_name)
-    #write_endpoints_file(LOCAL_PATH, project_name)
     write_requirements_txt(LOCAL_PATH, project_name, file_source)
     write_gitignore(LOCAL_PATH, project_name, file_source)
     write_readme_file(LOCAL_PATH, project_name, file_source)
-    #write_application_file(LOCAL_PATH, project_name)
     #write_application_config(LOCAL_PATH, project_name)
     write_src_util_file(LOCAL_PATH, project_name, file_source)
-    #write_wsgi_file(LOCAL_PATH, project_name)
+    write_wsgi_file(LOCAL_PATH, project_name, file_source)
+    write_app_file(LOCAL_PATH, project_name, file_source)
     write_visualization_file(LOCAL_PATH, project_name, file_source)
     write_normalization_file(LOCAL_PATH, project_name, file_source)
     write_preprocessing_file(LOCAL_PATH, project_name, file_source)
@@ -90,6 +89,7 @@ def new(project_name, python_version, implemented):
 
     if implemented in ['yes', 'ye', 'y', 'Yes', 'YES', 'Y']:
         write_titanic_data(LOCAL_PATH, project_name, file_source)
+        write_myrequests_file(LOCAL_PATH, project_name, file_source)
 
 
     write_test_file(LOCAL_PATH, project_name, file_source)
@@ -127,4 +127,31 @@ def predict():
     else:
         print("Making predictions: ")
         os.system('python ./predict.py')
-
+
+
+@click.argument('image_name')
+@click.option('-t', '--tag', 'tag', default='latest', show_default=True)
+@cli.command()
+def build(image_name, tag):
+    """
+    Build a docker image with given image_name. Only run if you have docker installed.
+    One should be at the root directory.
+    """
+    if not os.path.exists('src/Dockerfile'):
+        click.echo("You gotta have an src/Dockerfile file. You must be at the root folder.")
+    else:
+        os.system(f'docker build -f src/Dockerfile -t {image_name}:{tag} .')
+
+
+@click.argument('image_name')
+@click.option('-t', '--tag', 'tag', default='latest', show_default=True)
+@cli.command()
+def run(image_name, tag):
+    """
+    Run a container with given image_name. 
+    Only run if you have docker installed.
+    """
+    if not os.path.exists('src/Dockerfile'):
+        click.echo("You gotta have an src/Dockerfile file. You must be at the root folder.")
+    else:
+        os.system(f'docker run --rm -p 5000:5000 {image_name}:{tag}')

hermione/file_text/app.txt

@@ -0,0 +1,41 @@
+from flask import Flask, request, redirect, url_for, flash, jsonify
+import numpy as np
+import pandas as pd
+from joblib import load
+import json
+import logging
+
+logging.getLogger().setLevel(logging.INFO)
+
+app = Flask(__name__)
+
+def predict_new(X, probs=True):
+    model = load('model/titanic_model_rf.pkl')
+    p = model.get_preprocessing()
+
+    X = p.clean_data(X)
+    X = p.categ_encoding(X)
+
+    columns = model.get_columns()
+    for col in columns:
+        if col not in X.columns:
+            X[col] = 0
+    if probs:
+        return model.predict_proba(X)[:,1]
+    else:
+        return model.predict(X)
+
+@app.route('/invocations', methods=['POST'])
+def predict():
+    data = pd.read_json(request.json)
+    predictions = np.array2string(predict_new(data, probs=True))
+    return jsonify(predictions)
+
+@app.route('/health', methods=['GET'])
+def health_check():
+    resp = jsonify(success=True)
+    return resp
+
+
+if __name__ == "__main__":
+    app.run(host='0.0.0.0')

hermione/file_text/dockerfile.txt

@@ -25,15 +25,14 @@ COPY requirements.txt /opt/ml/code/src/requirements.txt
 RUN pip3 install --no-cache -r /opt/ml/code/src/requirements.txt
 
 # Copy project files
-COPY output/titanic_model_rf.pkl /opt/ml/output/titanic_model_rf.pkl
+COPY output/titanic_model_rf.pkl /opt/ml/code/src/api/model/titanic_model_rf.pkl
 COPY src/api/ /opt/ml/code/src/api/
 COPY src/config/ /opt/ml/code/src/config/
 COPY src/ml/ /opt/ml/code/src/ml/
 COPY src/util.py /opt/ml/code/src/util.py
-COPY src/predict.py /opt/ml/code/src/predict.py
 
 # Change working directory
-WORKDIR /opt/ml/code/src
+WORKDIR /opt/ml/code/src/api
 
 
 # Environment variables
@@ -43,5 +42,7 @@ ENV PYTHONDONTWRITEBYTECODE=1 \
 
 ENV PYTHONPATH="/opt/ml/code/src:${PATH}"
 
+EXPOSE 5000
+
 # Execution command
-CMD python3 predict.py
+CMD ["gunicorn", "-w", "3", "-b", ":5000", "-t", "360", "--reload", "wsgi:app"]