An open source project from Data to AI Lab at MIT.

AutoML for Renewable Energy Industries.

GreenGuard

• Free software: MIT license
• Documentation: https://D3-AI.github.io/GreenGuard
• Homepage: https://github.com/D3-AI/GreenGuard

Overview

The GreenGuard project is a collection of end-to-end solutions for machine learning problems commonly found in monitoring wind energy production systems. Most tasks utilize sensor data emanating from monitoring systems. We utilize the foundational innovations developed for automation of machine Learning at Data to AI Lab at MIT.

The salient aspects of this customized project are:

• A set of ready to use, well tested pipelines for different machine learning tasks. These are vetted through testing across multiple publicly available datasets for the same task.
• An easy interface to specify the task, pipeline, and generate results and summarize them.
• A production ready, deployable pipeline.
• An easy interface to tune pipelines using Bayesian Tuning and Bandits library.
• A community oriented infrastructure to incorporate new pipelines.
• A robust continuous integration and testing infrastructure.
• A learning database recording all past outcomes --> tasks, pipelines, outcomes.

Data Format

In order to be able to use the GreenGuard Pipelines to make predictions over you time Series data, you will need to following tables, formatted as CSV files:

• A Readings table that contains:
  • turbine_id: Unique identifier of the turbine which this reading comes from.
  • signal_id: Unique identifier of the signal which this reading comes from.
  • timestamp: Time where the reading took place, as an ISO formatted datetime.
  • value: Numeric value of this reading.

	turbine_id	signal_id	timestamp	value
0	T1	S1	2001-01-01 00:00:00	1
1	T1	S1	2001-01-01 12:00:00	2
2	T1	S1	2001-01-02 00:00:00	3
3	T1	S1	2001-01-02 12:00:00	4
4	T1	S1	2001-01-03 00:00:00	5
5	T1	S1	2001-01-03 12:00:00	6
6	T1	S2	2001-01-01 00:00:00	7
7	T1	S2	2001-01-01 12:00:00	8
8	T1	S2	2001-01-02 00:00:00	9
9	T1	S2	2001-01-02 12:00:00	10
10	T1	S2	2001-01-03 00:00:00	11
11	T1	S2	2001-01-03 12:00:00	12

• A Target times table that contains:
  • turbine_id: Unique identifier of the turbine which this label corresponds to.
  • cutoff_time: Time associated with this target
  • target: The value that we want to predict. This can either be a numerical value or a categorical label. This column can also be skipped when preparing data that will be used only to make predictions and not to fit any pipeline.

	turbine_id	cutoff_time	target
0	T1	2001-01-02 00:00:00	0
1	T1	2001-01-03 00:00:00	1
2	T1	2001-01-04 00:00:00	0

Additionally, if available, two more tables can be passed alongside the previous ones in order to provide additional information about the turbines and signals.

• A Turbines table that contains a turbine_id and additional properties about each turbine

	turbine_id	latitude	longitude	height	manufacturer
0	T1	49.8729	-6.44571	23.435	M1
1	T2	49.8729	-6.4457	24.522	M1
2	T3	49.8729	-6.44565	23.732	M2

• A Signals table that contains a signal_id and additional properties about each signal

	signal_id	sensor_type	sensor_brand	sensitivity
0	S1	t1	b1	200
1	S2	t2	b2	500

Demo Dataset

For development and demonstration purposes, we include a dataset with data from several telemetry signals associated with one wind energy production turbine.

This data, which has been already formatted as expected by the GreenGuard Pipelines, can be browsed and downloaded directly from the d3-ai-greenguard AWS S3 Bucket.

This dataset is adapted from the one used in the project by Cohen, Elliot J., "Wind Analysis." Joint Initiative of the ECOWAS Centre for Renewable Energy and Energy Efficiency (ECREEE), The United Nations Industrial Development Organization (UNIDO) and the Sustainable Engineering Lab (SEL). Columbia University, 22 Aug. 2014. Available online here

The complete list of manipulations performed on the original dataset to convert it into the demo one that we are using here is exhaustively shown and explained in the Green Guard Demo Data notebook.

Concepts

Before diving into the software usage, we briefly explain some concepts and terminology.

Primitive

We call the smallest computational blocks used in a Machine Learning process primitives, which:

• Can be either classes or functions.
• Have some initialization arguments, which MLBlocks calls init_params.
• Have some tunable hyperparameters, which have types and a list or range of valid values.

Template

Primitives can be combined to form what we call Templates, which:

• Have a list of primitives.
• Have some initialization arguments, which correspond to the initialization arguments of their primitives.
• Have some tunable hyperparameters, which correspond to the tunable hyperparameters of their primitives.

Pipeline

Templates can be used to build Pipelines by taking and fixing a set of valid hyperparameters for a Template. Hence, Pipelines:

• Have a list of primitives, which corresponds to the list of primitives of their template.
• Have some initialization arguments, which correspond to the initialization arguments of their template.
• Have some hyperparameter values, which fall within the ranges of valid tunable hyperparameters of their template.

A pipeline can be fitted and evaluated using the MLPipeline API in MLBlocks.

Tuning

We call tuning the process of, given a dataset and a template, find the pipeline derived from the given template that gets the best possible score on the given dataset.

This process usually involves fitting and evaluating multiple pipelines with different hyperparameter values on the same data while using optimization algorithms to deduce which hyperparameters are more likely to get the best results in the next iterations.

We call each one of these tries a tuning iteration.

Current tasks and pipelines

In our current phase, we are addressing two tasks - time series classification and time series regression. To provide solutions for these two tasks we have two components.

GreenGuardPipeline

This class is the one in charge of learning from the data and making predictions by building MLBlocks pipelines and later on tuning them using BTB

GreenGuardLoader

A class responsible for loading the time series data from CSV files, and return it in the format ready to be used by the GreenGuardPipeline.

Install

Requirements

GreenGuard has been developed and runs on Python 3.5, 3.6 and 3.7.

Also, although it is not strictly required, the usage of a virtualenv is highly recommended in order to avoid interfering with other software installed in the system where you are trying to run GreenGuard.

Installation

The simplest and recommended way to install GreenGuard is using pip:

pip install greenguard

For development, you can also clone the repository and install it from sources

git clone [email protected]:D3-AI/GreenGuard.git
cd GreenGuard
make install-develop

Quickstart

In this example we will load some demo data using the GreenGuardLoader and fetch it to the GreenGuardPipeline for it to find the best possible pipeline, fit it using the given data and then make predictions from it.

1. Load and explore the data

The first step is to load the demo data.

For this, we will import and call the greenguard.loader.load_demo function without any arguments:

from greenguard.loader import load_demo

X, y, readings = load_demo()

The returned objects are:

X: A pandas.DataFrame with the target_times table data without the target column.

   turbine_id  timestamp
0          T1 2013-01-01
1          T1 2013-01-02
2          T1 2013-01-03
3          T1 2013-01-04
4          T1 2013-01-05

y: A pandas.Series with the target column from the target_times table.

0    0.0
1    0.0
2    0.0
3    0.0
4    0.0
Name: target, dtype: float64

readings: A pandas.DataFrame containing the time series data in the format explained above.

   turbine_id  signal_id  timestamp  value
0  T1          S1        2013-01-01  817.0
1  T1          S2        2013-01-01  805.0
2  T1          S3        2013-01-01  786.0
3  T1          S4        2013-01-01  809.0
4  T1          S5        2013-01-01  755.0

2. Split the data

If we want to split the data in train and test subsets, we can do so by splitting the X and y variables with any suitable tool.

In this case, we will do it using the train_test_split function from scikit-learn.

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, random_state=0)

3. Finding a Pipeline

Once we have the data ready, we need to find a suitable pipeline.

The list of available GreenGuard Pipelines can be obtained using the greenguard.get_pipelines function.

from greenguard import get_pipelines

pipelines = get_pipelines()

The returned pipeline variable will be dict containing the names of all the pipelines available and their paths:

'greenguard_classification'
'greenguard_regression'

3. Finding the best Pipeline

Once we have loaded the data, we create a GreenGuardPipeline instance by passing:

• template (string): the name of a template or the path to a template json file.
• metric (string or function): The name of the metric to use or a metric function to use.
• cost (bool): Whether the metric is a cost function to be minimized or a score to be maximized.

Optionally, we can also pass defails about the cross validation configuration:

• stratify
• cv_splits
• shuffle
• random_state

In this case, we will be loading the greenguard_classification pipeline, using the accuracy metric, and using only 2 cross validation splits:

from greenguard.pipeline import GreenGuardPipeline

pipeline = GreenGuardPipeline(
    template='greenguard_classification',
    metric='f1_macro',
    cv_splits=5
)

Once we have created the pipeline, we can call its tune method to find the best possible hyperparameters for our data, passing the X, y, and readings variables returned by the loader, as well as an indication of the number of tuning iterations that we want to perform.

pipeline.tune(X_train, y_train, readings, iterations=10)

After the tuning process has finished, the hyperparameters have been already set in the classifier.

We can see the found hyperparameters by calling the get_hyperparameters method,

pipeline.get_hyperparameters()

which will return a dictionary with the best hyperparameters found so far:

{
    "pandas.DataFrame.resample#1": {
        "rule": "1D",
        "time_index": "timestamp",
        "groupby": [
            "turbine_id",
            "signal_id"
        ],
        "aggregation": "mean"
    },
    "pandas.DataFrame.unstack#1": {
        "level": "signal_id",
        "reset_index": true
    },
    ...

as well as the obtained cross validation score by looking at the score attribute of the pipeline object:

pipeline.score  # -> 0.6447509660798626

NOTE: If the score is not good enough, we can call the tune method again as many times as needed and the pipeline will continue its tuning process every time based on the previous results!

4. Fitting the pipeline

Once we are satisfied with the obtained cross validation score, we can proceed to call the fit method passing again the same data elements.

This will fit the pipeline with all the training data available using the best hyperparameters found during the tuning process:

pipeline.fit(X_train, y_train, readings)

5. Use the fitted pipeline

After fitting the pipeline, we are ready to make predictions on new data:

predictions = pipeline.predict(X_test, readings)

And evaluate its prediction performance:

from sklearn.metrics import accuracy_score

accuracy_score(y_test, predictions) # -> 0.6413043478260869

6. Save and load the pipeline

Since the tuning and fitting process takes time to execute and requires a lot of data, you will probably want to save a fitted instance and load it later to analyze new signals instead of fitting pipelines over and over again.

This can be done by using the save and load methods from the GreenGuardPipeline.

In order to save an instance, call its save method passing it the path and filename where the model should be saved.

path = 'my_pipeline.pkl'

pipeline.save(path)

Once the pipeline is saved, it can be loaded back as a new GreenGuardPipeline by using the GreenGuardPipeline.load method:

new_pipeline = GreenGuardPipeline.load(path)

Once loaded, it can be directly used to make predictions on new data.

new_pipeline.predict(X_test, readings)

Use your own Dataset

Once you are familiar with the GreenGuardPipeline usage, you will probably want to run it on your own dataset.

Here are the necessary steps:

1. Prepare the data

Firt of all, you will need to prepare your data as 4 CSV files like the ones described in the data format section above.

2. Create a GreenGuardLoader

Once you have the CSV files ready, you will need to import the greenguard.loader.GreenGuardLoader class and create an instance passing:

• path - str: The path to the folder where the 4 CSV files are
• target_times - str, gptional: The name of the target table. Defaults to target_times.
• target_column - str, optional: The name of the target column. Defaults to target.
• readings - str, optional: The name of the readings table. Defaults to readings.
• turbines - str, optional: The name of the turbines table. Defaults to None.
• signals - str, optional: The name of the signals table. Defaults to None.
• gzip - bool, optional: Set to True if the CSV files are gzipped. Defaults to False.

For example, here we will be loading a custom dataset which has been sorted in gzip format inside the my_dataset folder, and for which the target table has a different name:

from greenguard.loader import GreenGuardLoader

loader = GreenGuardLoader(path='my_dataset', target='labels', gzip=True)

3. Call the loader.load method.

Once the loader instance has been created, we can call its load method:

X, y, tables = loader.load()

Optionally, if the dataset contains only data to make predictions and the target column does not exist, we can pass it the argument False to skip it:

X, readings = loader.load(target=False)

Docker Usage

GreenGuard comes configured and ready to be distributed and run as a docker image which starts a jupyter notebook already configured to use greenguard, with all the required dependencies already installed.

For more details about how to run GreenGuard over docker, please check the DOCKER.md documentation.

What's next?

For more details about GreenGuard and all its possibilities and features, please check the project documentation site!