Otpsy is a toolkit for detecting and managing outliers in your dataset with ease. Outliers, those data points that deviate significantly from the majority, can have a substantial impact on statistical analyses. My package gather various outlier detection methods, making it convenient for psychologists and data analysts to identify and handle outliers in a straightforward manner. Click here for a glossary and here for a detailed explanation of the structure.
The package provides a range of outlier detection methods, allowing users to choose the most suitable approach for their specific needs. The available methods include:
- IQR (Interquartile Range)
- Tukey
- Standard Deviation (SD)
- Recursive Standard Deviation (rSD)
- Median Absolute Deviation (MAD)
-
$S_n$ Method - Percentile
- Cut-off (e.g., response time under 80 ms)
- Identical response (for Likert scales or behavioral performance)
For a mathematical description, see Jones et al., (2019)
The Otpsy package enables users to effortlessly and dynamically visualize outliers across multiple columns and methods simultaneously. This feature facilitates a comprehensive understanding of data distribution and threshold through histograms or scatter plots. The visualisation is running on plotly and dash. To do that, use the method visualise() on the sample object (see below or the docs for more details).
One of the major features of this package is its ability to efficiently compute outlier detection for multiple columns simultaneously. This streamlines the process for psychologists dealing with extensive datasets and diverse variables.
To get started, create a Sample object by specifying the DataFrame, the columns to test, and the participant reference column. After the visualisation of the distribution (if wanted), you can apply the desired outlier detection method using a method call, such as sample.methodIQR(). You apply to your "sample" the method IQR. This will generate an outlier object containing the identified outliers.
After detection, the package allows for comprehensive inspection of the outliers using print(outliers). It prints detailed information as we can usually see in R summary. Furthermore, with outliers.inspect() method, you obtains a DataFrame to inspect the value of outliers.
It is possible to concat outliers object .concat(), add manually specific index outliers.add() or to remove an index outliers.remove(). Finally, outliers can be managed using the outliers.manage() method, providing options to remove, winsorize, or replace them with NaN values.
Otpsy is a python package. To install it, you should use pip :
pip install otpsyFor now, the package only support pandas Dataframe and numpy array. Imagine that you want to delete participants that have at least one aberrant values on the column 1, the column 2 or the column 3.
import otpsy as ot
import pandas as pd
# Import dataframe
df = pd.read_csv("example.csv")
# Select what has to be tested
sample = ot.Sample(df, ["Col1", "Col2", "Col3"], "participant_name")
# Apply a specific method (mad) and remove outliers
clean_df = sample.methodMAD(distance = 2.5).manage("delete")For a more exhaustive presentation of functionnality, check the jupyter file exhaustive_example.ipynb and the folder "docs".
About outlier detection methods :
- Add self.position to multi and update parameters
- Check print output for each method
- Deactivate the conversion to numeric if user wants
About visualisation :
- Hover text on shape or annotation or legend to understand which color corresponds to which method.
- Add summary to the left.
- Add the possibility to add user-input value (and thus percentile method)
About the structure :
- Create automated test (in progress)
- Create tutorial
- Create data simulation
- Create a clear documentation (in progress)
- Create an export function to keep the information