gramener / eda Goto Github PK

• Create a data/samples/ directory with a sample.csv and sample.xlsx file
• Follow Hadley Wickham's instructions to create a test case structure
• Write a test case for read_eda for each file in the data/samples/ directory. Ensure that the file is read without errors and that the results have the correct number of rows, column and column names
• Write a test case for eda_metadata for each file in the data/samples/ directory. Compare the results against a pre-defined data structure stored as a separate file in the data/samples/ directory as sample.csv.meta.yaml and sample.xslx.meta.yaml. Use the attached meta.yaml as a sample
• Ensure that the tests pass

meta.zip

Structure the API so that the following works:

install.packages('eda')
library(eda)

# metadata returns an RC class.
# See http://adv-r.had.co.nz/OO-essentials.html
meta = metadata('path-to-file.xlsx', ...)

# All top level meta variables should be available as attributes
# The variable name is the same as the label in lowercase, with space -> underscore
meta$filename
meta$description
meta$source
meta$format
meta$file_size
meta$encoding
meta$row_count
meta$column_count
meta$row_description
meta$sampling_method
meta$prepared_on
meta$modified_on
meta$prepared_by

# All column info must be as a DataFrame
meta$columns
# The library user should be able to change the above.
# For example, they may override a column type from boolean to discrete

# Also store the data
meta$data

# We run methods on this class
meta$save(path='path-to-file.xlsx', sheet='Sheetname', format='xlsx')
# Format can only be XLSX for now, i.e. ignore it. We can add it later

# Apply univariate analysis on the (potentially modified) metadata
# Return ann RC object
uni = univariate(meta)

# This can be saved
uni$save(path='path-to-file.xlsx', sheet='Sheetname', format='xlsx')
uni$saveplot(path='path-to-file.xlsx', sheet='Sheetname', format='xlsx')
# Format can only be XLSX for now, i.e. ignore it. We can add it later

• Metadata: show file size in natural units. Or at least add commas
• Univariate: allow log plots and log-log plots. Users should be able to pick which plots have what scale
• Box plots: not as useful as histogram - other than to spot outliers. Can we add outliers to histogram and skip this?
• Rank frequency: less useful than histogram. Make this optional
• Bivariate cat-vs-cat: show values as proportions of column total
• Bivariate outliers: show exceptions when relationship is strong / clear

• Documentation - Add the documentation on usage.
• Samples - Add an examples folder containing 3 R inbuilt datasets, run the analysis and show the output.