diagnoserjl

Robert Schnitman
2018-03-16
Recommended Citation:
       Schnitman, Robert (2018). diagnoserjl v0.0.1.0. https://github.com/robertschnitman/diagnoserjl

Outline

0. Installation (under construction)
1. Introduction
2. diagnose()
3. fitres()
4. modeldf()
5. validate()
6. Conclusion
7. References

0. Installation

# Julia ≥ 0.6.2.
# Package imports
#   DataFrames ≥ 0.10.1
#   GLM        ≥ 0.8.1
#   Plots      ≥ 0.15.1
  
Pkg.clone("https://github.com/robertschnitman/diagnoserjl.git")

1. Introduction

Based on the original R library, diagnoser (https://github.com/robertschnitman/diagnoser).

Motivation: make an equivalent package for Julia for pedagogical purposes and to take advantage of its capabilities.

Tables in the sections below were created with a Markdown table converter (Donat Studios 2017).

2. diagnose()

file = "stata_auto.csv"
auto = readtable(file)

model_lm = fit(LinearModel, @formula(price ~ mpg + weight), auto)

diagnose(model_lm)

3. modeldf()

modeldf(model_lm, 0.95) # default confidence interval is 0.95.

Row	terms	beta	se	moe	ci_lower	ci_upper	t	p
1	(Intercept)	1946.07	3597.05	7172.31	-5226.24	9118.38	0.541018	0.5902
2	mpg	-49.5122	86.156	171.79	-221.302	122.278	-0.574681	0.5673
3	weight	1.74656	0.641354	1.27882	0.467736	3.02538	2.72324	0.0081

4. fitres()

fitres(model_lm) # Outputs a dataframe: fitted values, residuals, and residuals %.

Row	fit	residual	residual_pct
1	5974.22	-1875.22	-0.457482
2	6955.33	-2206.33	-0.464589
3	5467.72	-1668.72	-0.439251
4	6632.14	-1816.14	-0.377106
5	8329.35	-502.347	-0.0641813
6	7464.72	-1676.72	-0.289689
7	4553.58	-100.578	-0.0225865
8	6684.54	-1495.54	-0.288213
9	7930.52	2441.48	0.235391
10	6943.64	-2861.64	-0.701038
11	8815.5	2569.5	0.225692
…
63	3918.89	76.1108	0.0190515
64	7226.13	5763.87	0.443716
65	3854.95	40.0458	0.0102813
66	3793.59	4.41278	0.00116187
67	5264.06	634.944	0.107636
68	4253.62	-505.62	-0.134904
69	5718.16	0.838351	0.000146591
70	4579.86	2560.14	0.358564
71	3479.05	1917.95	0.355374
72	4079.12	617.878	0.131547
73	4183.92	2666.08	0.389209
74	6640.95	5354.05	0.446357

fr = fitres(model_lm, auto) # If the original dataset from the model is specified,
                            #   then the fitted values & residuals are merged with it.
                            
fr[1:10, [:price, :mpg, :weight, :fit, :residual, :residual_pct]]

Row	price	mpg	weight	fit	residual	residual_pct
1	4099	22	2930	5974.22	-1875.22	-0.457482
2	4749	17	3350	6955.33	-2206.33	-0.464589
3	3799	22	2640	5467.72	-1668.72	-0.439251
4	4816	20	3250	6632.14	-1816.14	-0.377106
5	7827	15	4080	8329.35	-502.347	-0.0641813
6	5788	18	3670	7464.72	-1676.72	-0.289689
7	4453	26	2230	4553.58	-100.578	-0.0225865
8	5189	20	3280	6684.54	-1495.54	-0.288213
9	10372	16	3880	7930.52	2441.48	0.235391
10	4082	19	3400	6943.64	-2861.64	-0.701038

5. validate()

Case 1: OLS

validate(model_lm, false) # By default, the output (dataframe = false) returns an array. 
                          # Set to "true" for a dataframe.
                          # See help documentation (?validate) for statistics definitions.

:n	[74.0]
:ar2	[0.273485]
:r2	[0.293389]
:rmse	[2462.54]
:mad	[1389.76]
:mae	[1966.89]
:medianpe	[-0.105916]
:mpe	[-0.11274]
:sdpe	[0.358649]
:sepe	[0.041692]
:residual_mean	[-0.0]
:residual_median	[-503.983]
:residual_sd	[2479.35]
:residual_se	[288.219]

Case 2: GLM (logit)

testf = function(x)      # Setting up a binary variable for a logistic regression.
    if (x == "Domestic")
		0
    else
		1
    end
end

auto[:foreign2] = testf.(auto[:foreign])
model_glm = fit(GeneralizedLinearModel, @formula(foreign2 ~ mpg + weight), auto, Binomial())

validate(model_glm, true) # dataframe = true --> outputs dataframe

Row	statistic	value
1	n	[74.0]
2	deviance_residual	[54.3503]
3	aer	[0.013514]

6. Conclusion

The hope of this library is to (1) minimize the programming tedium in statistical reporting; (2) assist people in diagnosing the validity of their results; and (3) inspire developers and end-users alike to apply Julia in their work.

7. References

Schnitman, Robert (2017). diagnoser v0.0.2.5. https://github.com/robertschnitman/diagnoser

Donat Studios (2017). CSV To Markdown Table Generator. https://donatstudios.com/CsvToMarkdownTable

End of Document