-
Notifications
You must be signed in to change notification settings - Fork 290
/
12-tuning-parameters.Rmd
639 lines (470 loc) · 37.9 KB
/
12-tuning-parameters.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
```{r tuning-setup, include = FALSE}
knitr::opts_chunk$set(fig.path = "figures/")
library(tidymodels)
library(patchwork)
library(ggforce)
library(doMC)
registerDoMC(cores = parallel::detectCores())
tidymodels_prefer()
## -----------------------------------------------------------------------------
source("ames_snippets.R")
## -----------------------------------------------------------------------------
data(two_class_dat)
set.seed(91)
split <- initial_split(two_class_dat)
training_set <- training(split)
testing_set <- testing(split)
data_grid <- crossing(A = seq(0.4, 4, length = 200), B = seq(.14, 3.9, length = 200))
## -----------------------------------------------------------------------------
load("RData/search_examples.RData")
```
# Model Tuning and the Dangers of Overfitting {#tuning}
In order to use a model for prediction, the parameters for that model must be estimated. Some of these parameters can be estimated directly from the training data, but other parameters, called _tuning parameters_ or _hyperparameters_, must be specified ahead of time and can't be directly found from training data. These are unknown structural or other kind of values that have significant impact on the model but cannot be directly estimated from the data. This chapter will provide examples of tuning parameters and show how we use tidymodels functions to create and handle tuning parameters. We'll also demonstrate how poor choices of these values lead to overfitting and introduce several tactics for finding optimal tuning parameters values. Chapters \@ref(grid-search) and \@ref(iterative-search) go into more detail on specific optimization methods for tuning.
## Model Parameters
In ordinary linear regression, there are two parameters $\beta_0$ and $\beta_1$ of the model:
$$ y_i = \beta_0 + \beta_1 x_i + \epsilon_i$$
When we have the outcome ($y$) and predictor ($x$) data, we can estimate the two parameters $\beta_0$ and $\beta_1$:
$$\hat \beta_1 = \frac{\sum_i (y_i-\bar{y})(x_i-\bar{x})}{\sum_i(x_i-\bar{x})^2}$$
and
$$\hat \beta_0 = \bar{y}-\hat \beta_1 \bar{x}.$$
We can directly estimate these values from the data for this example model because they are analytically tractable; if we have the data, then we can estimate these model parameters.
:::rmdnote
There are many situations where a model has parameters that _can't_ be directly estimated from the data.
:::
For the KNN model, the prediction equation for a new value $x_0$ is
$$\hat y = \frac{1}{K}\sum_{\ell = 1}^K x_\ell^*$$
where $K$ is the number of neighbors and the $x_\ell^*$ are the $K$ closest values to $x_0$ in the training set. The model itself is not defined by a model equation; the previous prediction equation instead defines it. This characteristic, along with the possible intractability of the distance measure, makes it impossible to create a set of equations that can be solved for $K$ (iteratively or otherwise). The number of neighbors has a profound impact on the model; it governs the flexibility of the class boundary. For small values of $K$, the boundary is very elaborate while for large values, it might be quite smooth.
The number of nearest neighbors is a good example of a tuning parameter or hyperparameter that cannot be directly estimated from the data.
## Tuning Parameters for Different Types of Models {#tuning-parameter-examples}
There are many examples of tuning parameters or hyperparameters in different statistical and machine learning models:
* Boosting is an ensemble method that combines a series of base models, each of which is created sequentially and depends on the previous models. The number of boosting iterations is an important tuning parameter that usually requires optimization.
* In the classic single-layer artificial neural network (a.k.a. the multilayer perceptron), the predictors are combined using two or more hidden units. The hidden units are linear combinations of the predictors that are captured in an _activation function_ (typically a nonlinear function, such as a sigmoid). The hidden units are then connected to the outcome units; one outcome unit is used for regression models, and multiple outcome units are required for classification. The number of hidden units and the type of activation function are important structural tuning parameters.
* Modern gradient descent methods are improved by finding the right optimization parameters. Examples of such hyperparameters are learning rates, momentum, and the number of optimization iterations/epochs [@Goodfellow]. Neural networks and some ensemble models use gradient descent to estimate the model parameters. While the tuning parameters associated with gradient descent are not structural parameters, they often require tuning.
In some cases, preprocessing techniques require tuning:
* In principal component analysis, or its supervised cousin called partial least squares, the predictors are replaced with new, artificial features that have better properties related to collinearity. The number of extracted components can be tuned.
* Imputation methods estimate missing predictor values using the complete values of one or more predictors. One effective imputation tool uses $K$-nearest neighbors of the complete columns to predict the missing value. The number of neighbors modulates the amount of averaging and can be tuned.
Some classical statistical models also have structural parameters:
* In binary regression, the logit link is commonly used (i.e., logistic regression). Other link functions, such as the probit and complementary log-log, are also available [@Dobson99]. This example is described in more detail in the Section \@ref(what-to-optimize).
* Non-Bayesian longitudinal and repeated measures models require a specification for the covariance or correlation structure of the data. Options include compound symmetric (a.k.a. exchangeable), autoregressive, Toeplitz, and others [@littell2000modelling].
A counterexample where it is inappropriate to tune a parameter is the prior distribution required for Bayesian analysis. The prior encapsulates the analyst's belief about the distribution of a quantity before evidence or data are taken into account. For example, in Section \@ref(tidyposterior), we used a Bayesian ANOVA model, and we were unclear about what the prior should be for the regression parameters (beyond being a symmetric distribution). We chose a t-distribution with one degree of freedom for the prior since it has heavier tails; this reflects our added uncertainty. Our prior beliefs should not be subject to optimization. Tuning parameters are typically optimized for performance whereas priors should not be tweaked to get "the right results."
:::rmdwarning
Another (perhaps more debatable) counterexample of a parameter that does _not_ need to be tuned is the number of trees in a random forest or bagging model. This value should instead be chosen to be large enough to ensure numerical stability in the results; tuning it cannot improve performance as long as the value is large enough to produce reliable results. For random forests, this value is typically in the thousands while the number of trees needed for bagging is around 50 to 100.
:::
## What do we Optimize? {#what-to-optimize}
How should we evaluate models when we optimize tuning parameters? It depends on the model and the purpose of the model.
For cases where the statistical properties of the tuning parameter are tractable, common statistical properties can be used as the objective function. For example, in the case of binary logistic regression, the link function can be chosen by maximizing the likelihood or information criteria. However, these statistical properties may not align with the results achieved using accuracy-oriented properties. As an example, @FriedmanGFA optimized the number of trees in a boosted tree ensemble and found different results when maximizing the likelihood and accuracy:
> degrading the likelihood by overfitting actually improves misclassification error rate. Although perhaps counterintuitive, this is not a contradiction; likelihood and error rate measure different aspects of fit quality.
To demonstrate, consider the classification data shown in Figure \@ref(fig:two-class-dat) with two predictors, two classes, and a training set of `r nrow(training_set)` data points.
```{r two-class-dat}
#| echo = FALSE,
#| fig.cap = "An example two-class classification data set with two predictors",
#| fig.alt = "An example two-class classification data set with two predictors. The two predictors have a moderate correlation and there is some locations of separation between the classes."
ggplot(training_set, aes(x = A, y = B, color = Class, pch = Class)) +
geom_point(alpha = 0.7) +
coord_equal() +
labs(x = "Predictor A", y = "Predictor B", color = NULL, pch = NULL) +
scale_color_manual(values = c("#CC6677", "#88CCEE"))
```
We could start by fitting a linear class boundary to these data. The most common method for doing this is to use a generalized linear model in the form of _logistic regression_. This model relates the _log odds_ of a sample being Class 1 using the _logit_ transformation:
$$ \log\left(\frac{\pi}{1 - \pi}\right) = \beta_0 + \beta_1x_1 + \ldots + \beta_px_p$$
In the context of generalized linear models, the logit function is the _link function_ between the outcome ($\pi$) and the predictors. There are other link functions that include the _probit_ model:
$$\Phi^{-1}(\pi) = \beta_0 + \beta_1x_1 + \ldots + \beta_px_p$$
where $\Phi$ is the cumulative standard normal function, as well as the _complementary log-log_ model:
$$\log(-\log(1-\pi)) = \beta_0 + \beta_1x_1 + \ldots + \beta_px_p$$
Each of these models results in linear class boundaries. Which one should we use? Since, for these data, the number of model parameters does not vary, the statistical approach is to compute the (log) likelihood for each model and determine the model with the largest value. Traditionally, the likelihood is computed using the same data that were used to estimate the parameters, not using approaches like data splitting or resampling from Chapters \@ref(splitting) and \@ref(resampling).
For a data frame `training_set`, let's create a function to compute the different models and extract the likelihood statistics for the training set (using `broom::glance()`):
```{r tuning-likelihood}
library(tidymodels)
tidymodels_prefer()
llhood <- function(...) {
logistic_reg() %>%
set_engine("glm", ...) %>%
fit(Class ~ ., data = training_set) %>%
glance() %>%
select(logLik)
}
bind_rows(
llhood(),
llhood(family = binomial(link = "probit")),
llhood(family = binomial(link = "cloglog"))
) %>%
mutate(link = c("logit", "probit", "c-log-log")) %>%
arrange(desc(logLik))
```
According to these results, the logistic model has the best statistical properties.
From the scale of the log-likelihood values, it is difficult to understand if these differences are important or negligible. One way of improving this analysis is to resample the statistics and separate the modeling data from the data used for performance estimation. With this small data set, repeated 10-fold cross-validation is a good choice for resampling. In the `r pkg(yardstick)` package, the `mn_log_loss()` function is used to estimate the negative log-likelihood, with our results shown in Figure \@ref(fig:resampled-log-lhood).
```{r tuning-resampled-log-lhood, eval=FALSE}
set.seed(1201)
rs <- vfold_cv(training_set, repeats = 10)
# Return the individual resampled performance estimates:
lloss <- function(...) {
perf_meas <- metric_set(roc_auc, mn_log_loss)
logistic_reg() %>%
set_engine("glm", ...) %>%
fit_resamples(Class ~ A + B, rs, metrics = perf_meas) %>%
collect_metrics(summarize = FALSE) %>%
select(id, id2, .metric, .estimate)
}
resampled_res <-
bind_rows(
lloss() %>% mutate(model = "logistic"),
lloss(family = binomial(link = "probit")) %>% mutate(model = "probit"),
lloss(family = binomial(link = "cloglog")) %>% mutate(model = "c-log-log")
) %>%
# Convert log-loss to log-likelihood:
mutate(.estimate = ifelse(.metric == "mn_log_loss", -.estimate, .estimate)) %>%
group_by(model, .metric) %>%
summarize(
mean = mean(.estimate, na.rm = TRUE),
std_err = sd(.estimate, na.rm = TRUE) / sqrt(n()),
.groups = "drop"
)
resampled_res %>%
filter(.metric == "mn_log_loss") %>%
ggplot(aes(x = mean, y = model)) +
geom_point() +
geom_errorbar(aes(xmin = mean - 1.64 * std_err, xmax = mean + 1.64 * std_err),
width = .1) +
labs(y = NULL, x = "log-likelihood")
```
```{r resampled-log-lhood, ref.label = "tuning-resampled-log-lhood"}
#| echo = FALSE,
#| fig.height = 3,
#| fig.cap = "Means and approximate 90% confidence intervals for the resampled binomial log-likelihood with three different link functions",
#| fig.alt = "Means and approximate 90% confidence intervals for the resampled binomial log-likelihood with three different link functions. The logit link has the largest value, followed by the probit link. The complementary log log link has far lower values."
```
:::rmdnote
The scale of these values is different than the previous values since they are computed on a smaller data set; the value produced by `broom::glance()` is a sum while `yardstick::mn_log_loss()` is an average.
:::
These results exhibit evidence that the choice of the link function matters somewhat. Although there is an overlap in the confidence intervals, the logistic model has the best results.
What about a different metric? We also calculated the area under the ROC curve for each resample. These results, which reflect the discriminative ability of the models across numerous probability thresholds, show a lack of difference in Figure \@ref(fig:resampled-roc).
```{r resampled-roc}
#| echo = FALSE,
#| fig.height = 3,
#| fig.cap = "Means and approximate 90% confidence intervals for the resampled area under the ROC curve with three different link functions",
#| fig.alt = "Means and approximate 90% confidence intervals for the resampled area under the ROC curve with three different link functions. The logit link has the largest value, followed by the probit link. The confidence intervals show a large amount of overlap between the two methods."
resampled_res %>%
filter(.metric == "roc_auc") %>%
ggplot(aes(x = mean, y = model)) +
geom_point() +
geom_errorbar(aes(xmin = mean - 1.64 * std_err, xmax = mean+ 1.64 * std_err),
width = .1) +
labs(y = NULL, x = "area under the ROC curve")
```
Given the overlap of the intervals, as well as the scale of the x-axis, any of these options could be used. We see this again when the class boundaries for the three models are overlaid on the test set of `r nrow(testing_set)` data points in Figure \@ref(fig:three-link-fits).
```{r three-link-fits}
#| echo = FALSE,
#| fig.cap = "The linear class boundary fits for three link functions",
#| fig.alt = "The linear class boundary fits for three link functions. The lines have very similar slopes with the complementary log log having a slightly different intercept than the other two links."
logit_pred <-
logistic_reg() %>%
set_engine("glm") %>%
fit(Class ~ A + B, data = training_set) %>%
predict(data_grid, type = "prob") %>%
bind_cols(data_grid) %>%
mutate(link = "logit")
probit_pred <-
logistic_reg() %>%
set_engine("glm", family = binomial(link = "probit")) %>%
fit(Class ~ A + B, data = training_set) %>%
predict(data_grid, type = "prob") %>%
bind_cols(data_grid) %>%
mutate(link = "probit")
cloglog_pred <-
logistic_reg() %>%
set_engine("glm", family = binomial(link = "cloglog")) %>%
fit(Class ~ A + B, data = training_set) %>%
predict(data_grid, type = "prob") %>%
bind_cols(data_grid) %>%
mutate(link = "c-log-log")
link_grids <-
bind_rows(logit_pred, probit_pred, cloglog_pred) %>%
mutate(link = factor(link, levels = c("logit", "probit", "c-log-log")))
link_grids %>%
ggplot(aes(x = A, y = B)) +
geom_point(data = testing_set, aes(color = Class, pch = Class),
alpha = 0.7, show.legend = FALSE) +
geom_contour(aes( z = .pred_Class1, lty = link), breaks = 0.5, color = "black") +
scale_color_manual(values = c("#CC6677", "#88CCEE")) +
coord_equal() +
labs(x = "Predictor A", y = "Predictor B")
```
:::rmdwarning
This exercise emphasizes that different metrics might lead to different decisions about the choice of tuning parameter values. In this case, one metric indicates the models are somewhat different while another metric shows no difference at all.
:::
Metric optimization is thoroughly discussed by @thomas2020problem who explore several issues, including the gaming of metrics. They warn that:
> The unreasonable effectiveness of metric optimization in current AI approaches is a fundamental challenge to the field, and yields an inherent contradiction: solely optimizing metrics leads to far from optimal outcomes.
## The consequences of poor parameter estimates {#overfitting-bad}
Many tuning parameters modulate the amount of model complexity. More complexity often implies more malleability in the patterns that a model can emulate. For example, as shown in Section \@ref(spline-functions), adding degrees of freedom in a spline function increases the intricacy of the prediction equation. While this is an advantage when the underlying motifs in the data are complex, it can also lead to overinterpretation of chance patterns that would not reproduce in new data. _Overfitting_ is the situation where a model adapts too much to the training data; it performs well for the data used to build the model but poorly for new data.
:::rmdwarning
Since tuning model parameters can increase model complexity, poor choices can lead to overfitting.
:::
Recall the single layer neural network model described in Section \@ref(tuning-parameter-examples). With a single hidden unit and sigmoidal activation functions, a neural network for classification is, for all intents and purposes, just logistic regression. However, as the number of hidden units increases, so does the complexity of the model. In fact, when the network model uses sigmoidal activation units, @cybenko1989approximation showed that the model is a universal function approximator as long as there are enough hidden units.
We fit neural network classification models to the same two-class data from the previous section, varying the number of hidden units. Using the area under the ROC curve as a performance metric, the effectiveness of the model on the training set increases as more hidden units are added. The network model thoroughly and meticulously learns the training set. If the model judges itself on the training set ROC value, it prefers many hidden units so that it can nearly eliminate errors.
Chapters \@ref(splitting) and \@ref(resampling) demonstrated that simply repredicting the training set is a poor approach to model evaluation. Here, the neural network very quickly begins to overinterpret patterns that it sees in the training set. Compare these three example class boundaries (developed with the training set) overlaid on training and test sets in Figure \@ref(fig:two-class-boundaries).
```{r tuning-two-class-comps, include = FALSE}
two_class_rec <-
recipe(Class ~ ., data = two_class_dat) %>%
step_normalize(all_numeric_predictors())
mlp_mod <-
mlp(hidden_units = tune(), epochs = 1000) %>%
set_engine("nnet") %>%
set_mode("classification")
mlp_wflow <-
workflow() %>%
add_recipe(two_class_rec) %>%
add_model(mlp_mod)
mlp_res <-
tibble(
hidden_units = 1:20,
train = NA_real_,
test = NA_real_,
model = vector(mode = "list", length = 20)
)
for(i in 1:nrow(mlp_res)) {
set.seed(27)
tmp_mod <-
mlp_wflow %>% finalize_workflow(mlp_res %>% slice(i) %>% select(hidden_units)) %>%
fit(training_set)
mlp_res$train[i] <-
roc_auc_vec(training_set$Class, predict(tmp_mod, training_set, type = "prob")$.pred_Class1)
mlp_res$test[i] <-
roc_auc_vec(testing_set$Class, predict(tmp_mod, testing_set, type = "prob")$.pred_Class1)
mlp_res$model[[i]] <- tmp_mod
}
```
```{r two-class-boundaries}
#| echo = FALSE,
#| fig.height = 8,
#| fig.cap = "Class boundaries for three models with increasing numbers of hidden units. The boundaries are fit on the training set and shown for the training and test sets.",
#| fig.alt = "Class boundaries for three models with increasing numbers of hidden units. The boundaries are fit on the training set and shown for the training and test sets. After a single hidden unit, the boundaries become wildly complex. The test set plots show that the more complex models do not conform to the data that was not used to fit the model."
te_plot <-
mlp_res %>%
slice(c(1, 4, 20)) %>%
mutate(
probs = map(model, ~ bind_cols(data_grid, predict(.x, data_grid, type = "prob")))
) %>%
dplyr::select(hidden_units, probs) %>%
unnest(cols = c(probs)) %>%
mutate(
label = paste(format(hidden_units), "units"),
label = ifelse(label == " 1 units", " 1 unit", label)
) %>%
ggplot(aes(x = A, y = B)) +
geom_point(data = testing_set, aes(color = Class, pch = Class),
alpha = 0.5, show.legend = FALSE) +
geom_contour(aes( z = .pred_Class1), breaks = 0.5, color = "black") +
scale_color_manual(values = c("#CC6677", "#88CCEE")) +
facet_wrap(~ label, nrow = 1) +
coord_equal() +
ggtitle("Test Set") +
labs(x = "Predictor A", y = "Predictor B")
tr_plot <-
mlp_res %>%
slice(c(1, 4, 20)) %>%
mutate(
probs = map(model, ~ bind_cols(data_grid, predict(.x, data_grid, type = "prob")))
) %>%
dplyr::select(hidden_units, probs) %>%
unnest(cols = c(probs)) %>%
mutate(
label = paste(format(hidden_units), "units"),
label = ifelse(label == " 1 units", " 1 unit", label)
) %>%
ggplot(aes(x = A, y = B)) +
geom_point(data = training_set, aes(color = Class, pch = Class),
alpha = 0.5, show.legend = FALSE) +
geom_contour(aes( z = .pred_Class1), breaks = 0.5, color = "black") +
scale_color_manual(values = c("#CC6677", "#88CCEE")) +
facet_wrap(~ label, nrow = 1) +
coord_equal() +
ggtitle("Training Set") +
labs(x = "Predictor A", y = "Predictor B")
tr_plot / te_plot
```
The single unit model does not adapt very flexibly to the data (since it is constrained to be linear). A model with four hidden units begins to show signs of overfitting with an unrealistic boundary for values away from the data mainstream. This is caused by a single data point from the first class in the upper-right corner of the data. By 20 hidden units, the model is beginning to memorize the training set, creating small islands around those data to minimize the resubstitution error rate. These patterns do not repeat in the test set. This last panel is the best illustration of how tuning parameters that control complexity must be modulated so that the model is effective. For a 20-unit model, the training set ROC AUC is `r round(mlp_res$train[20], 3)` but the test set value is `r round(mlp_res$test[20], 3)`.
This occurrence of overfitting is obvious with two predictors that we can plot. However, in general, we must use a quantitative approach for detecting overfitting.
:::rmdnote
The solution for detecting when a model is overemphasizing the training set is using out-of-sample data.
:::
Rather than using the test set, some form of resampling is required. This could mean an iterative approach (e.g., 10-fold cross-validation) or a single data source (e.g., a validation set).
## Two general strategies for optimization
Tuning parameter optimization usually falls into one of two categories: grid search and iterative search.
_Grid search_ is when we predefine a set of parameter values to evaluate. The main choices involved in grid search are how to make the grid and how many parameter combinations to evaluate. Grid search is often judged as inefficient since the number of grid points required to cover the parameter space can become unmanageable with the curse of dimensionality. There is truth to this concern, but it is most true when the process is not optimized. This is discussed more in Chapter \@ref(grid-search).
_Iterative search_ or sequential search is when we sequentially discover new parameter combinations based on previous results. Almost any nonlinear optimization method is appropriate, although some are more efficient than others. In some cases, an initial set of results for one or more parameter combinations is required to start the optimization process. Iterative search is discussed more in Chapter \@ref(iterative-search).
Figure \@ref(fig:tuning-strategies) shows two panels that demonstrate these two approaches for a situation with two tuning parameters that range between zero and one. In each, a set of contours shows the true (simulated) relationship between the parameters and the outcome. The optimal results are in the upper-right-hand corners.
```{r tuning-strategies}
#| echo = FALSE,
#| fig.cap = "Examples of pre-defined grid tuning and an iterative search method. The lines represent contours of a performance metric; it is best in the upper-right-hand side of the plot.",
#| fig.alt = "Examples of pre-defined grid tuning and an iterative search method. The lines represent contours of some performance metric that is best in the upper-right-hand side of the plot. The grid search shows points that cover the space well and has one point near the optimum. The iterative search method has many more points and meanders to the optimum where many points zero in on the best value."
grid_plot <-
ggplot(sfd_grid, aes(x = x, y = y)) +
geom_point() +
lims(x = 0:1, y = 0:1) +
labs(x = "Parameter 1", y = "Parameter 2", title = "Space-Filling Grid") +
geom_contour(data = grid_contours,
aes(z = obj),
alpha = .3,
bins = 12) +
coord_equal() +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank())
search_plot <-
ggplot(nm_res, aes(x = x, y = y)) +
geom_point(size = .7) +
lims(x = 0:1, y = 0:1) +
labs(x = "Parameter 1", y = "Parameter 2", title = "Global Search") +
coord_equal() +
geom_contour(data = grid_contours,
aes(x = x, y = y, z = obj),
alpha = .3,
bins = 12) +
theme_bw() +
theme(panel.grid.major = element_blank(),
panel.grid.minor = element_blank())
grid_plot + search_plot
```
The left-hand panel of Figure \@ref(fig:tuning-strategies) shows a type of grid called a space-filling design. This is a type of experimental design devised for covering the parameter space such that tuning parameter combinations are not close to one another. The results for this design do not place any points exactly at the truly optimal location. However, one point is in the general vicinity and would probably have performance metric results that are within the noise of the most optimal value.
The right-hand panel of Figure \@ref(fig:tuning-strategies) illustrates the results of a global search method: the Nelder-Mead simplex method [@Olsson:1975p3609]. The starting point is in the lower-left part of the parameter space. The search meanders across the space until it reaches the optimum location, where it strives to come as close as possible to the numerically best value. This particular search method, while effective, is not known for its efficiency; it requires many function evaluations, especially near the optimal values. Chapter \@ref(iterative-search) discusses more efficient search algorithms.
:::rmdnote
Hybrid strategies are also an option and can work well. After an initial grid search, a sequential optimization can start from the best grid combination.
:::
Examples of these strategies are discussed in detail in the next two chapters. Before moving on, let's learn how to work with tuning parameter objects in tidymodels, using the `r pkg(dials)` package.
## Tuning Parameters in tidymodels {#tuning-params-tidymodels}
We've already dealt with quite a number of arguments that correspond to tuning parameters for recipe and model specifications in previous chapters. It is possible to tune:
* the threshold for combining neighborhoods into an "other" category (with argument name `threshold`) discussed in Section \@ref(dummies)
* the number of degrees of freedom in a natural spline (`deg_free`, Section \@ref(spline-functions))
* the number of data points required to execute a split in a tree-based model (`min_n`, Section \@ref(create-a-model))
* the amount of regularization in penalized models (`penalty`, Section \@ref(create-a-model))
For `r pkg(parsnip)` model specifications, there are two kinds of parameter arguments. *Main arguments* are those that are most often optimized for performance and are available in multiple engines. The main tuning parameters are top-level arguments to the model specification function. For example, the `rand_forest()` function has main arguments `trees`, `min_n`, and `mtry` since these are most frequently specified or optimized.
A secondary set of tuning parameters are *engine specific*. These are either infrequently optimized or are specific only to certain engines. Again using random forests as an example, the `r pkg(ranger)` package contains some arguments that are not used by other packages. One example is gain penalization, which regularizes the predictor selection in the tree induction process. This parameter can help modulate the trade-off between the number of predictors used in the ensemble and performance [@wundervald2020generalizing]. The name of this argument in `ranger()` is `regularization.factor`. To specify a value via a `r pkg(parsnip)` model specification, it is added as a supplemental argument to `set_engine()`:
```{r tuning-args, eval = FALSE}
rand_forest(trees = 2000, min_n = 10) %>% # <- main arguments
set_engine("ranger", regularization.factor = 0.5) # <- engine-specific
```
:::rmdwarning
The main arguments use a harmonized naming system to remove inconsistencies across engines while engine-specific arguments do not.
:::
How can we signal to tidymodels functions which arguments should be optimized? Parameters are marked for tuning by assigning them a value of `tune()`. For the single layer neural network used in Section \@ref(overfitting-bad), the number of hidden units is designated for tuning using:
```{r tuning-mlp-units}
neural_net_spec <-
mlp(hidden_units = tune()) %>%
set_mode("regression") %>%
set_engine("keras")
```
The `tune()` function doesn't execute any particular parameter value; it only returns an expression:
```{r tuning-tune-exprs}
tune()
```
Embedding this `tune()` value in an argument will tag the parameter for optimization. The model tuning functions shown in the next two chapters parse the model specification and/or recipe to discover the tagged parameters. These functions can automatically configure and process these parameters since they understand their characteristics (e.g., the range of possible values, etc.).
To enumerate the tuning parameters for an object, use the `extract_parameter_set_dials()` function:
```{r tuning-mlp-units-param}
extract_parameter_set_dials(neural_net_spec)
```
The results show a value of `nparam[+]`, indicating that the number of hidden units is a numeric parameter.
There is an optional identification argument that associates a name with the parameters. This can come in handy when the same kind of parameter is being tuned in different places. For example, with the Ames housing data from Section \@ref(resampling-summary), the recipe encoded both longitude and latitude with spline functions. If we want to tune the two spline functions to potentially have different levels of smoothness, we call `step_ns()` twice, once for each predictor. To make the parameters identifiable, the identification argument can take any character string:
```{r tuning-id}
ames_rec <-
recipe(Sale_Price ~ Neighborhood + Gr_Liv_Area + Year_Built + Bldg_Type +
Latitude + Longitude, data = ames_train) %>%
step_log(Gr_Liv_Area, base = 10) %>%
step_other(Neighborhood, threshold = tune()) %>%
step_dummy(all_nominal_predictors()) %>%
step_interact( ~ Gr_Liv_Area:starts_with("Bldg_Type_") ) %>%
step_ns(Longitude, deg_free = tune("longitude df")) %>%
step_ns(Latitude, deg_free = tune("latitude df"))
recipes_param <- extract_parameter_set_dials(ames_rec)
recipes_param
```
Note that the `identifier` and `type` columns are not the same for both of the spline parameters.
When a recipe and model specification are combined using a workflow, both sets of parameters are shown:
```{r tuning-mlp-rec}
wflow_param <-
workflow() %>%
add_recipe(ames_rec) %>%
add_model(neural_net_spec) %>%
extract_parameter_set_dials()
wflow_param
```
:::rmdwarning
Neural networks are exquisitely capable of emulating nonlinear patterns. Adding spline terms to this type of model is unnecessary; we combined this model and recipe for illustration only.
:::
Each tuning parameter argument has a corresponding function in the `r pkg(dials)` package. In the vast majority of the cases, the function has the same name as the parameter argument:
```{r tuning-dials-unit}
hidden_units()
threshold()
```
The `deg_free` parameter is a counterexample; the notion of degrees of freedom comes up in a variety of different contexts. When used with splines, there is a specialized `r pkg(dials)` function called `spline_degree()` that is, by default, invoked for splines:
```{r tuning-spline-df}
spline_degree()
```
The `r pkg(dials)` package also has a convenience function for extracting a particular parameter object:
```{r tuning-extract}
# identify the parameter using the id value:
wflow_param %>% extract_parameter_dials("threshold")
```
Inside the parameter set, the range of the parameters can also be updated in place:
```{r tuning-update}
extract_parameter_set_dials(ames_rec) %>%
update(threshold = threshold(c(0.8, 1.0)))
```
The _parameter sets_ created by `extract_parameter_set_dials()` are consumed by the tidymodels tuning functions (when needed). If the defaults for the tuning parameter objects require modification, a modified parameter set is passed to the appropriate tuning function.
:::rmdnote
Some tuning parameters depend on the dimensions of the data. For example, the number of nearest neighbors must be between one and the number of rows in the data.
:::
In some cases, it is easy to have reasonable defaults for the range of possible values. In other cases, the parameter range is critical and cannot be assumed. The primary tuning parameter for random forest models is the number of predictor columns that are randomly sampled for each split in the tree, usually denoted as `mtry()`. Without knowing the number of predictors, this parameter range cannot be preconfigured and requires finalization.
```{r tuning-rf}
rf_spec <-
rand_forest(mtry = tune()) %>%
set_engine("ranger", regularization.factor = tune("regularization")) %>%
set_mode("regression")
rf_param <- extract_parameter_set_dials(rf_spec)
rf_param
```
Complete parameter objects have `[+]` in their summary; a value of `[?]` indicates that at least one end of the possible range is missing. There are two methods for handling this. The first is to use `update()`, to add a range based on what you know about the data dimensions:
```{r tuning-rfupdate}
rf_param %>%
update(mtry = mtry(c(1, 70)))
```
However, this approach might not work if a recipe is attached to a workflow that uses steps that either add or subtract columns. If those steps are not slated for tuning, the `finalize()` function can execute the recipe once to obtain the dimensions:
```{r tuning-finalize}
pca_rec <-
recipe(Sale_Price ~ ., data = ames_train) %>%
# Select the square-footage predictors and extract their PCA components:
step_normalize(contains("SF")) %>%
# Select the number of components needed to capture 95% of
# the variance in the predictors.
step_pca(contains("SF"), threshold = .95)
updated_param <-
workflow() %>%
add_model(rf_spec) %>%
add_recipe(pca_rec) %>%
extract_parameter_set_dials() %>%
finalize(ames_train)
updated_param
updated_param %>% extract_parameter_dials("mtry")
```
When the recipe is prepared, the `finalize()` function learns to set the upper range of `mtry` to `r updated_param %>% extract_parameter_dials("mtry") %>% range_get() %>% pluck("upper")` predictors.
Additionally, the results of `extract_parameter_set_dials()` will include engine-specific parameters (if any). They are discovered in the same way as the main arguments and included in the parameter set. The `r pkg(dials)` package contains parameter functions for all potentially tunable engine-specific parameters:
```{r tuning-engine-param-set}
rf_param
regularization_factor()
```
Finally, some tuning parameters are best associated with transformations. A good example of this is the penalty parameter associated with many regularized regression models. This parameter is nonnegative and it is common to vary its values in log units. The primary `r pkg(dials)` parameter object indicates that a transformation is used by default:
```{r tuning-penalty}
penalty()
```
This is important to know, especially when altering the range. New range values must be in the transformed units:
```{r tuning-penalty-set}
# correct method to have penalty values between 0.1 and 1.0
penalty(c(-1, 0)) %>% value_sample(1000) %>% summary()
# incorrect:
penalty(c(0.1, 1.0)) %>% value_sample(1000) %>% summary()
```
The scale can be changed if desired with the `trans` argument. You can use natural units but the same range:
```{r tuning-penalty-natural}
penalty(trans = NULL, range = 10^c(-10, 0))
```
## Chapter Summary
This chapter introduced the process of tuning model hyperparameters that cannot be directly estimated from the data. Tuning such parameters can lead to overfitting, often by allowing a model to grow overly complex, so using resampled data sets together with appropriate metrics for evaluation is important. There are two general strategies for determining the right values, grid search and iterative search, which we will explore in depth in the next two chapters. In tidymodels, the `tune()` function is used to identify parameters for optimization, and functions from the `r pkg(dials)` package can extract and interact with tuning parameters objects.