nelson-gon / manymodelr

Build and Tune Several Models

• 🖤 Building Impactful Software.

• 💻 Simplicity focused open source advocate. Author of several open source R and python packages like cytounet, urlfix, manymodelr, pyfdc, mde. A full list is available at https://nelson-gon.github.io/projects.

• 📫 Please contact me via LinkedIn at https://www.linkedin.com/in/nelsongon/

Thank you 🖤

Keep Building 🏗

manymodelr's Issues

Moving to dplyr 1.0.0

With dplyr 1.0.0, functions such as *_ifhave been superseded. There is therefore a need to move to the neweracross` and also to check for any breaking changes.

Support multiple model types in fit_models

I would like to for example do:

fit_models(df=iris,yname=c("Sepal.Length","Sepal.Width"), xname="Petal.Length + Petal.Width",modeltype=c("lm", "glm","lmer"))

Fix typos in vignettes and warnings

Null output for attribute extraction.

extract_model_info returns NULL when p values are requested yet p values actually exist.

Rethink argument matching in `multi_model_`.

Given several methods, multi_model_1 should be able to match arguments only for the method in question and not throw warnings.

Something like:

[multi_model_1(train,"dist","speed",method = c("glm","lm"),metric = "rmse",control = ctrl,new_data = test,family="poisson")

should not throw any warnings.

Similarly, perhaps have several metrics eg c("accuracy","rmse") and match these to the relevant model?

Let get_var_corr handle multiple combinations

get_var_corr currently only allows setting a single comparison_var. Extend that to handle more than a single variable, return a matrix-like object perhaps?
See ref.

Speed up get_mode

get_mode is currently slow, is there a better way to rewrite it while achieving the same results?

Improve coverage

Tidy up error handling

To ease error handling and make it more uniform and/or improve readability.

Extend fit_models output

Perhaps add an option to allow control over what should be returned. Perhaps something like add?

Support simple model reports

Description

I would like to make a simple report to explain what a model's output means.

Similar Features

Some packages support this but I found it a bit less ideal.

Feature Details

Given a model, produce a table with estimates and what they actually mean.

Proposed Implementation

As above.

agg_by_group

In agg_by_group, the length of the grouping is incorrect for groups >1. See for instance:

head(agg_by_group(mtcars,cyl~hp+vs,sum))
#> Grouped By[1]:   hp vs 

That should be 2, not one.

Support Type Exclusion in fit_model(s)

Say I wanted to fit a linear model, it would be great if I could filter out non numeric columns.

Fix issues with coverage and cmd check

In coverage, workflows do not work for some reason(possibly lack of webhooks for codecov).
CMD checks fail due to dplyr 0.8.5 being loaded yet we need a higher version of dplyr.

Fix CRAN checks

Fix as suggested at r-lib/testthat#1051

Drop plot_corr

This function heavily relies on ggplot2 and does somewhat go against the package's design philosophy. Perhaps move it to a different standalone package or drop it altogether? Better alternatives exist too making it really less useful in the long run.

Improve multi_model_1

multi_model_1 seems to go through several layers of "recursion" to do something relatively simple. Could this be simplified and/or sped up?

Builds failing: Issues with length of logical arguments

Currently builds seem to fail on R devel mainly due to issues in extract_model_info especially regarding length of the output of logical operations.