The objective of this project is to convert observations of time-stamped sensor readings and/or cycling device operations from text files into indexed time series in pandas and NumPy.

For example, it can convert observations of temperatures and thermostat-driven ON/OFF cycles in cooling or heating modes, or batteries’ charging and discharging cycles.

The package will format raw data and match the results across the sources based on metadata such as device ID’s and/or location ID’s and time, in order to form multi-dimensional time series.

It automatically detects the type of data in each column of a text file, based on the data itself and based on column labels. The detection allows for any ordering of columns in the input data.

Note the need for metadata files for both thermostats and zip codes, if matching data with devices based on geographic location. Example files are also in the data folder.

This project’s intent is to accelerate analysis of time-stamped data at the device level as well as associated data from other sources if applicable. It does this by putting observations into an indexed form that can be summarized in aggregated form and at the device level. It supports visualization in time series form.

It may be used for general scientific research that aims to 1) index time stamped data in general from large text files, and 2) use pandas and NumPy.

It provides the user an API that abstracts away some of the mundane details of

• reading raw text data into a structured form in Python
• putting the Python variables into pandas DataFrames, and
• selecting devices and time ranges, and otherwise summarizing the data,

while still providing full usage of pure pandas functionality.

The package may be installed using pip or conda.

• 2.7
• 3.4
• 3.5

• configparser (2.7, 3.4)
• future
• pandas
• NumPy

Assuming no version is specified, the latest version of caar will be installed.

pip install caar

conda install -c nickpowersys caar

The example notebook demonstrates how to work with primary time-stamped data from these sources:

• cooling and heating cycles within buildings
• indoor temperature data, and
• outdoor temperature data

https://anaconda.org/nickpowersys/caarexamples/notebook

http://caar.readthedocs.io/en/latest/

Sample input files are in the data directory at https://github.com/nickpowersys/CaaR.

CaaR can be used to read delimited text files and (optionally) save the data in Python pickle files for fast access.

Common delimited text file formats including commas, tabs, pipes and spaces are detected in that order within the first row and the first delimiter detected is used. In all cases, rows are only used if the number of values match the number of column labels in the first row.

Each input file is expected to have (at least) columns representing ID's, time stamps (or starting and ending time stamps for cycles), and (if not cycles) corresponding observations.

To use the automatic column detection functionality, use the keyword argument 'auto' within the pickle_from_file() or dict_from_file() function (see the notebook example or API documentation) and assign it one of the values: 'cycles', 'inside', or 'outside' (for example, auto='inside').

The IDs should contain digits, and may also contain letters (leading zeroes are also allowed in place of letters). Having the string 'id', 'Id' or 'ID' in the column heading will cause that column to be the ID index within the combined ID-time stamp index for a given input file. If there is no such label, the leftmost column with alphanumeric strings (for example, 'T12' or '0123') will be taken as the ID.

Next, CaaR can create pandas DataFrames. CaaR and the pandas library offer many functions for summarizing and analyzing the data.

CaaR can convert DataFrames into NumPy time series arrays, for plotting/visualization and deeper data analysis.

I would welcome any feedback on features that would be useful. The project is a work in progress.

This project is licensed under the terms of the BSD 3-Clause License.