Edator

This is a python package that performs exploratory data analysis for users. It takes in a csv file and generates 3 documents that comprise of a text report containing a descriptive summary, a series of plots and a cleaned csv output.

Set up

Dependencies

• Python 3.8x
• matplotlib==3.1.2
• numpy==1.18.1
• pandas==1.0.0
• PySimpleGUI==4.19.0
• scikit-learn==0.22.1
• scipy==1.4.1
• seaborn==0.10.0
• statsmodels==0.11.1
• more-itertools==8.3.0

How to set up? (Important!)

1. You can clone or download my package.
2. Using terminal, move to the directory.
   • Example for Mac OS users:

   $ cd Downloads/Edator

3. Install the required packages using:

   pip install -r requirements.txt

4. After that, change directory into the Script folder using:

   $ cd Script

5. Now, execute the main.py file by:

   $ python main.py

6. You should see the following:

7. Choose the format of the file (csv or xls), the path to the file and the paths to export the plots, the report and the cleaned csv file to.

8. Done!

The concept behind Edator

Dealing with NaN values and zeros

How I deal with NaN value is that I only remove the affected rows when the percentage of NaN within that column is less than 5%. This applies to both numerical and categorical values. For anything above 5%, I replace the NaN values with median. For categorical values, the NaN values will be replace by mode.

Dealing with zeros is much harder as it is challenging to differentiate between a zero that is meaningful (has a purpose and should not be removed) and a zero that serves no purpose and can potentially add more noise to the dataset. Hence, I decided to inform the user about the percentage of zeros in the dataset.

Processing outliers

I use Z-score to detect outliers. If a Z-score is 0, it indicates that the data point’s score is identical to the mean score. A Z-score of 1.0 would indicate a value that is one standard deviation from the mean. Z-scores may be positive or negative, with a positive value indicating the score is above the mean and a negative score indicating it is below the mean.

In most cases, a threshold of 3 or -3 is used to filter off outliers and I have used this approach for all of my analysis.

Correlation

For correlation, I included:

1. Pearson and Spearman correlation for numerical-numerical variables.
2. One Way ANOVA for numerical-categorical variables
3. Chi-Square test for categorical-categorical variables

Using itertools.combinations, I identify every possible combinations among numerical-numerical variables, numerical-categorical variables and categorical-categorical variables. I then apply the correlation test based on the criteria I have set above.

Plots

For plots, I created:

1. Scatterplot for numerical variables
2. Countplot for categorical variables
3. Boxplot for numerical-categorical variables

Similar to correlation, I used itertools.combinations to create every possible plot. I have also added the hue feature to each scatterplot. I will only do so when the categorical variable has less than 5 unique values. Example, if hue = "fruits", I should only see 4 types of fruits.

Upcoming changes for version 0.3

1. Take in more file outputs beyond CSV and Excel
2. Gathering user input, I will increase the variety of plots beyond scatterplots, barplots and boxplots.
3. Report generated will be in HTML format.