Support for glmnet with an external family

hnagaty · May 2, 2021, 9:02am

Apologies for this long question, it's because I tried different scenarios for my inquiry and I'm trying to document them in the question.

The glmnet help pages states that glmnet supports any family object as used by stats:glm().

From version 4.0 onwards, glmnet supports both the original built-in families, as well as any family object as used by stats:glm(). This opens the door to a wide variety of additional models. For example family=binomial(link=cloglog) or family=negative.binomial(theta=1.5) (from the MASS library). Note that the code runs faster for the built-in families.

Referring to the help page of family {stats}, I understand that using an external family is specified via a function, for example family = quasipoisson().

I don't know if using external families is supported in tidymodels or not. I tried to use a quasipoisson() family with a poisson model, but I couldn't. Making the model specification and fitting works with no problems, the problems do occur when I use predict.

Using predict() without specifying a penalty makes predictions for all lambda values. The output is a long tibble (number of rows = no. of observations x no. of lambda values).
Using predict() with specifying a penalty gives an error Error: The ellipses are not used to pass args to the model function's predict function. These arguments cannot be used: 'penalty'.
Using multi_predict() gives an error Error in class() : 0 arguments passed to 'class' which requires 1

Extracting the glmnet fitted model object and then using predict.glmnet() method works with no problem. But this is not a suitable workaround, because I'm still unable to use the tidymodels tune functions, such as tune_grid() and fit_resamples().

Using fit_resamples() produces an error related to incorrect matrix dimensions.
Using tune_grid() works smoothly with no problems.
Using tune_grid() with a custom grid yields an error message: Error in class() : 0 arguments passed to 'class' which requires 1

Below is a reprex

library(tidymodels)
library(poissonreg)

Create some data to play with

n <- 2e3
set.seed(123)
x1 <- MASS::rnegbin(n, mu = 28, theta = 0.6)
x2 <- rnorm(n, 5, 10)
x3 <- as.factor(rnorm(n) < 0)
y <- as.integer(x1 + x2 + as.numeric(x3))
y[y < 0] <- 0
data <- tibble(y, x1, x2, x3)

Create a recipe, model spec and a workflow

# recipe
rec <- recipe(y ~ ., data = data) %>% 
  step_dummy(all_nominal()) %>% 
  step_normalize(all_predictors())

First, I try a spec for `glmnet` with the default family

pois_spec <- poisson_reg() %>% 
  set_engine("glmnet")

wflow <- workflow() %>% 
  add_recipe(rec) %>% 
  add_model(pois_spec)

Fit the workflow

wflow_fit <- wflow %>% 
  fit(data = data)

Using the predict() method
I should specify a penatly in a glmnet model

preds <- predict(wflow_fit, data, penalty = 0.5)
str(preds)
#> tibble [2,000 × 1] (S3: tbl_df/tbl/data.frame)
#>  $ .pred: num [1:2000] 21.7 86.5 14 25 22.8 ...

As predict without a penalty gives an error

preds <- predict(wflow_fit, data)
#> Error: `penalty` should be a single numeric value. `multi_predict()` can be used to get multiple predictions per row of data.

Also multi_predict() gives an error

preds <- multi_predict(wflow_fit, data, penalty = c(0.550, 0.604))
#> Error in class(): 0 arguments passed to 'class' which requires 1

Now I use a `quasipoisson()` family

pois_spec <- poisson_reg() %>% 
  set_engine("glmnet", family = quasipoisson())

wflow <- workflow() %>% 
  add_recipe(rec) %>% 
  add_model(pois_spec)

And fit thw workflow

wflow_fit <- wflow %>% 
  fit(data = data)

predict() works with no errors, but it produces unexpected output; a long tibble of
predictions at each lambda value

preds <- predict(wflow_fit, data)
str(preds)
#> tibble [122,000 × 2] (S3: tbl_df/tbl/data.frame)
#>  $ .pred_values: num [1:122000] 32.9 32.9 32.9 32.9 32.9 ...
#>  $ .pred_lambda: num [1:122000] 34.6 34.6 34.6 34.6 34.6 ...
preds %>% count(.pred_lambda)
#> # A tibble: 61 x 2
#>    .pred_lambda     n
#>  *        <dbl> <int>
#>  1        0.130  2000
#>  2        0.143  2000
#>  3        0.157  2000
#>  4        0.172  2000
#>  5        0.189  2000
#>  6        0.207  2000
#>  7        0.227  2000
#>  8        0.250  2000
#>  9        0.274  2000
#> 10        0.301  2000
#> # … with 51 more rows

And specifying a penalty gives am error

preds <- predict(wflow_fit, data, penalty = 0.150)
#> Error: The ellipses are not used to pass args to the model function's predict function. These arguments cannot be used: `penalty`

Using multi_predict() also gives an error, same as when used with default glmnet

preds <- multi_predict(wflow_fit, data, penalty = c(0.550, 0.604))
#> Error in class(): 0 arguments passed to 'class' which requires 1

Extract the fitted model and use glmnet predict method instead.

This works with no problems

model_fit <- wflow_fit %>% 
  chuck("fit", "fit", "fit")
class(model_fit)
#> [1] "glmnetfit" "glmnet"
data_X <- cbind(x1, x2, x3)
preds_glmnet <- predict(model_fit, data_X, s = 0.5)
str(preds_glmnet)
#>  num [1:2000, 1] 7.96 59.87 5.59 18.02 7.02 ...
#>  - attr(*, "dimnames")=List of 2
#>   ..$ : NULL
#>   ..$ : chr "1"

Now I’m using `tune` functions

res <- vfold_cv(data, v = 3)

Use fit_resamples()
yields error of incompatible matrix dimensions

resamples <- wflow %>% 
  fit_resamples(res)
#> x Fold1: preprocessor 1/1, model 1/1 (predictions): Error: Problem with `mutate()...
#> x Fold2: preprocessor 1/1, model 1/1 (predictions): Error: Problem with `mutate()...
#> x Fold3: preprocessor 1/1, model 1/1 (predictions): Error: Problem with `mutate()...
#> Warning: All models failed. See the `.notes` column.
resamples %>% 
  chuck(".notes", 1, 1)
#> [1] "preprocessor 1/1, model 1/1 (predictions): Error: Problem with `mutate()` input `.row`.\nx Input `.row` can't be recycled to size 40020.\nℹ Input `.row` is `orig_rows`.\nℹ Input `.row` must be size 40020 or 1, not 667."

Using tune_grid().
No errors and completes smoothly

pois_spec <- poisson_reg(
  penalty = tune(),
  mixture = tune()) %>% 
  set_engine(
    "glmnet", 
    family = quasipoisson())
wflow <- workflow() %>% 
  add_recipe(rec) %>% 
  add_model(pois_spec)
resamples <- wflow %>% 
  tune_grid(res)
resamples %>% 
  chuck(".notes", 1)
#> # A tibble: 0 x 1
#> # … with 1 variable: .notes <chr>

Using tune_grid() with a custom grid
yields an error Error in class() : 0 arguments passed to 'class' which requires 1

param_grid <- grid_regular(
  parameters(pois_spec),
  levels = c(mixture = 2, penalty = 20))
resamples <- wflow %>% 
  tune_grid(
    res,
    grid = param_grid)
#> x Fold1: preprocessor 1/1, model 1/2 (predictions): Error in class(): 0 arguments...
#> x Fold1: preprocessor 1/1, model 2/2 (predictions): Error in class(): 0 arguments...
#> x Fold2: preprocessor 1/1, model 1/2 (predictions): Error in class(): 0 arguments...
#> x Fold2: preprocessor 1/1, model 2/2 (predictions): Error in class(): 0 arguments...
#> x Fold3: preprocessor 1/1, model 1/2 (predictions): Error in class(): 0 arguments...
#> x Fold3: preprocessor 1/1, model 2/2 (predictions): Error in class(): 0 arguments...
#> Warning: All models failed. See the `.notes` column.
resamples %>% 
  chuck(".notes", 1, 1)
#> [1] "preprocessor 1/1, model 1/2 (predictions): Error in class(): 0 arguments passed to 'class' which requires 1"
#> [2] "preprocessor 1/1, model 2/2 (predictions): Error in class(): 0 arguments passed to 'class' which requires 1"

^{Created on 2021-05-01 by the reprex package (v1.0.0)}

Max · May 4, 2021, 10:26pm

Wow, there are a bunch of bugs that were exposed by this.

To start, when calling a glmnet model through parsnip, you should have gotten an error if penalty did not have a value. Even though that model does not require a value at the time of fit, parsnip does. We are already in the process of documenting this better but a discussion about this can be found here.

When calling predict(), you don't need to add the penalty argument (it is ignored; we should check that). It automatically uses the one specified in poisson_reg().

For Poisson regression, we don't (currently) have a multi_predict() method. The error message about Error in class() is a bug; it should say that "No multi_predict method exists for objects with classes ...". I think that I can add that method.

About the quasipoisson family... we can try to support that. It's in the grey aread for bug or not. Technically, it is not a Poisson model anymore since it doesn't use the regular Poisson distribution. That might sound like I'm splitting hairs but it matters for the R bits:

> poisson_reg(penalty = 0.5) %>% 
+     set_engine("glmnet") %>% 
+     fit(y ~ x1 + x2, data = data) %>% 
+     pluck("fit") %>% 
+     class()
[1] "fishnet" "glmnet" 
> 
> # 'fishnet' is their class of Poisson models
> 
> poisson_reg(penalty = 0.5) %>% 
+     set_engine("glmnet", family = quasipoisson()) %>% 
+     fit(y ~ x1 + x2, data = data) %>% 
+     pluck("fit") %>% 
+     class()
[1] "glmnetfit" "glmnet"   
> 
> # Now the class is not Poisson specific, so our methods don't know what to do

I would be happy to support it but that will take some time to figure out if we can. You can track these at:

github.com/tidymodels/parsnip

handle quasi-likelihood glmnet models better

opened 10:19PM - 04 May 21 UTC

closed 10:44AM - 02 Mar 23 UTC

topepo

feature

These don't return an object class specific to their model: ``` r library(t…idymodels) #> Registered S3 method overwritten by 'tune': #> method from #> required_pkgs.model_spec parsnip data("two_class_dat") logistic_reg(penalty = 0.1) %>% set_engine("glmnet") %>% fit(Class ~ ., data = two_class_dat) %>% pluck("fit") %>% class() #> [1] "lognet" "glmnet" logistic_reg(penalty = 0.1) %>% set_engine("glmnet", family = quasibinomial(link = "logit")) %>% fit(Class ~ ., data = two_class_dat) %>% pluck("fit") %>% class() #> [1] "glmnetfit" "glmnet" ``` Created on 2021-05-04 by the [reprex package](https://reprex.tidyverse.org) (v1.0.0.9000) We should add class to the `parsnip` model or restrict this family for this engine.

github.com/tidymodels/poissonreg

handle quasi-likelihood glmnet models better

opened 10:19PM - 04 May 21 UTC

topepo

feature

Similar to tidymodels/parsnip#483 ``` r library(tidymodels) #> Registered S…3 method overwritten by 'tune': #> method from #> required_pkgs.model_spec parsnip library(poissonreg) n <- 2e3 set.seed(123) x1 <- MASS::rnegbin(n, mu = 28, theta = 0.6) x2 <- rnorm(n, 5, 10) x3 <- as.factor(rnorm(n) < 0) y <- as.integer(x1 + x2 + as.numeric(x3)) y[y < 0] <- 0 data <- tibble(y, x1, x2, x3) poisson_reg(penalty = 0.5) %>% set_engine("glmnet") %>% fit(y ~ x1 + x2, data = data) %>% pluck("fit") %>% class() #> [1] "fishnet" "glmnet" # 'fishnet' is their class of Poisson models poisson_reg(penalty = 0.5) %>% set_engine("glmnet", family = quasipoisson()) %>% fit(y ~ x1 + x2, data = data) %>% pluck("fit") %>% class() #> [1] "glmnetfit" "glmnet" # Now the class is not Poisson specific, so our methods don't know what to do ``` Created on 2021-05-04 by the [reprex package](https://reprex.tidyverse.org) (v1.0.0.9000) We should add class to the `parsnip` model or restrict this family for this engine.

github.com/tidymodels/parsnip

For glmnet models, fail if no penalty value is provided.

opened 10:09PM - 04 May 21 UTC

closed 06:14PM - 11 May 21 UTC

topepo

bug

``` r library(tidymodels) data("two_class_dat") # These should both fail in… translate() logistic_reg(penalty = c(0.01, 0.1)) %>% set_engine("glmnet") %>% fit(Class ~ ., data = two_class_dat) #> parsnip model object #> #> Fit time: 12ms #> #> Call: glmnet::glmnet(x = maybe_matrix(x), y = y, family = "binomial") #> #> Df %Dev Lambda #> 1 0 0.00 0.308100 #> 2 1 4.74 0.280700 #> 3 1 8.72 0.255800 #> 4 1 12.08 0.233100 #> 5 1 14.96 0.212400 #> 6 1 17.44 0.193500 #> 7 1 19.57 0.176300 #> 8 1 21.42 0.160600 #> 9 1 23.02 0.146400 #> 10 1 24.41 0.133400 #> 11 1 25.62 0.121500 #> 12 1 26.67 0.110700 #> 13 1 27.58 0.100900 #> 14 1 28.37 0.091930 #> 15 1 29.05 0.083760 #> 16 1 29.64 0.076320 #> 17 1 30.15 0.069540 #> 18 1 30.59 0.063360 #> 19 1 30.97 0.057730 #> 20 1 31.30 0.052610 #> 21 1 31.58 0.047930 #> 22 1 31.82 0.043670 #> 23 1 32.02 0.039790 #> 24 2 32.48 0.036260 #> 25 2 33.27 0.033040 #> 26 2 33.95 0.030100 #> 27 2 34.54 0.027430 #> 28 2 35.05 0.024990 #> 29 2 35.48 0.022770 #> 30 2 35.86 0.020750 #> 31 2 36.18 0.018910 #> 32 2 36.46 0.017230 #> 33 2 36.70 0.015700 #> 34 2 36.90 0.014300 #> 35 2 37.08 0.013030 #> 36 2 37.23 0.011870 #> 37 2 37.35 0.010820 #> 38 2 37.46 0.009857 #> 39 2 37.55 0.008982 #> 40 2 37.63 0.008184 #> 41 2 37.70 0.007457 #> 42 2 37.75 0.006794 #> 43 2 37.80 0.006191 #> 44 2 37.84 0.005641 #> 45 2 37.87 0.005140 #> 46 2 37.90 0.004683 #> 47 2 37.92 0.004267 #> 48 2 37.94 0.003888 #> 49 2 37.96 0.003543 #> 50 2 37.98 0.003228 #> 51 2 37.99 0.002941 #> 52 2 38.00 0.002680 #> 53 2 38.01 0.002442 #> 54 2 38.01 0.002225 #> 55 2 38.02 0.002027 #> 56 2 38.02 0.001847 #> 57 2 38.03 0.001683 #> 58 2 38.03 0.001533 #> 59 2 38.03 0.001397 #> 60 2 38.03 0.001273 #> 61 2 38.04 0.001160 #> 62 2 38.04 0.001057 #> 63 2 38.04 0.000963 #> 64 2 38.04 0.000878 #> 65 2 38.04 0.000800 logistic_reg() %>% set_engine("glmnet") %>% fit(Class ~ ., data = two_class_dat) #> parsnip model object #> #> Fit time: 9ms #> #> Call: glmnet::glmnet(x = maybe_matrix(x), y = y, family = "binomial") #> #> Df %Dev Lambda #> 1 0 0.00 0.308100 #> 2 1 4.74 0.280700 #> 3 1 8.72 0.255800 #> 4 1 12.08 0.233100 #> 5 1 14.96 0.212400 #> 6 1 17.44 0.193500 #> 7 1 19.57 0.176300 #> 8 1 21.42 0.160600 #> 9 1 23.02 0.146400 #> 10 1 24.41 0.133400 #> 11 1 25.62 0.121500 #> 12 1 26.67 0.110700 #> 13 1 27.58 0.100900 #> 14 1 28.37 0.091930 #> 15 1 29.05 0.083760 #> 16 1 29.64 0.076320 #> 17 1 30.15 0.069540 #> 18 1 30.59 0.063360 #> 19 1 30.97 0.057730 #> 20 1 31.30 0.052610 #> 21 1 31.58 0.047930 #> 22 1 31.82 0.043670 #> 23 1 32.02 0.039790 #> 24 2 32.48 0.036260 #> 25 2 33.27 0.033040 #> 26 2 33.95 0.030100 #> 27 2 34.54 0.027430 #> 28 2 35.05 0.024990 #> 29 2 35.48 0.022770 #> 30 2 35.86 0.020750 #> 31 2 36.18 0.018910 #> 32 2 36.46 0.017230 #> 33 2 36.70 0.015700 #> 34 2 36.90 0.014300 #> 35 2 37.08 0.013030 #> 36 2 37.23 0.011870 #> 37 2 37.35 0.010820 #> 38 2 37.46 0.009857 #> 39 2 37.55 0.008982 #> 40 2 37.63 0.008184 #> 41 2 37.70 0.007457 #> 42 2 37.75 0.006794 #> 43 2 37.80 0.006191 #> 44 2 37.84 0.005641 #> 45 2 37.87 0.005140 #> 46 2 37.90 0.004683 #> 47 2 37.92 0.004267 #> 48 2 37.94 0.003888 #> 49 2 37.96 0.003543 #> 50 2 37.98 0.003228 #> 51 2 37.99 0.002941 #> 52 2 38.00 0.002680 #> 53 2 38.01 0.002442 #> 54 2 38.01 0.002225 #> 55 2 38.02 0.002027 #> 56 2 38.02 0.001847 #> 57 2 38.03 0.001683 #> 58 2 38.03 0.001533 #> 59 2 38.03 0.001397 #> 60 2 38.03 0.001273 #> 61 2 38.04 0.001160 #> 62 2 38.04 0.001057 #> 63 2 38.04 0.000963 #> 64 2 38.04 0.000878 #> 65 2 38.04 0.000800 ``` Created on 2021-05-04 by the [reprex package](https://reprex.tidyverse.org) (v1.0.0.9000)

github.com/tidymodels/parsnip

For `poisson_reg(engine = "glmnet")` models, fail if no penalty value is provided.

opened 10:11PM - 04 May 21 UTC

closed 03:22PM - 22 Jan 23 UTC

topepo

bug

Similar to tidymodels/parsnip#481 ``` r library(tidymodels) #> Registered S…3 method overwritten by 'tune': #> method from #> required_pkgs.model_spec parsnip library(poissonreg) # These should both fail in translate() poisson_reg(penalty = c(0.01, 0.1)) %>% set_engine("glmnet") %>% fit(count ~ (.)^2, data = seniors) #> parsnip model object #> #> Fit time: 8ms #> #> Call: glmnet::glmnet(x = maybe_matrix(x), y = y, family = "poisson") #> #> Df %Dev Lambda #> 1 0 0.00 254.000 #> 2 1 10.36 231.500 #> 3 1 18.38 210.900 #> 4 1 24.70 192.200 #> 5 1 29.76 175.100 #> 6 1 33.84 159.500 #> 7 1 37.15 145.400 #> 8 2 39.89 132.500 #> 9 2 43.20 120.700 #> 10 2 46.01 110.000 #> 11 2 48.40 100.200 #> 12 2 50.43 91.300 #> 13 2 52.15 83.180 #> 14 2 53.62 75.790 #> 15 2 54.87 69.060 #> 16 2 55.94 62.930 #> 17 2 56.84 57.340 #> 18 2 57.61 52.240 #> 19 2 58.27 47.600 #> 20 3 58.92 43.370 #> 21 4 61.09 39.520 #> 22 4 63.71 36.010 #> 23 4 66.01 32.810 #> 24 4 68.03 29.900 #> 25 4 69.80 27.240 #> 26 5 73.21 24.820 #> 27 6 76.71 22.610 #> 28 6 79.84 20.610 #> 29 6 82.50 18.770 #> 30 6 84.77 17.110 #> 31 6 86.72 15.590 #> 32 6 88.41 14.200 #> 33 6 89.87 12.940 #> 34 6 91.13 11.790 #> 35 6 92.23 10.740 #> 36 6 93.19 9.789 #> 37 6 94.03 8.920 #> 38 6 94.76 8.127 #> 39 6 95.40 7.405 #> 40 6 95.96 6.747 #> 41 6 96.45 6.148 #> 42 6 96.88 5.602 #> 43 6 97.27 5.104 #> 44 6 97.60 4.651 #> 45 6 97.89 4.238 #> 46 6 98.15 3.861 #> 47 6 98.38 3.518 #> 48 6 98.58 3.206 #> 49 6 98.75 2.921 #> 50 6 98.91 2.661 #> 51 6 99.04 2.425 #> 52 6 99.16 2.209 #> 53 6 99.27 2.013 #> 54 6 99.36 1.834 #> 55 6 99.44 1.671 #> 56 6 99.51 1.523 #> 57 6 99.57 1.388 #> 58 6 99.63 1.264 #> 59 6 99.68 1.152 #> 60 6 99.72 1.050 #> 61 6 99.75 0.956 #> 62 6 99.78 0.872 #> 63 6 99.81 0.794 #> 64 6 99.84 0.724 #> 65 6 99.86 0.659 #> 66 6 99.87 0.601 #> 67 6 99.89 0.547 #> 68 6 99.90 0.499 poisson_reg() %>% set_engine("glmnet") %>% fit(count ~ (.)^2, data = seniors) #> parsnip model object #> #> Fit time: 6ms #> #> Call: glmnet::glmnet(x = maybe_matrix(x), y = y, family = "poisson") #> #> Df %Dev Lambda #> 1 0 0.00 254.000 #> 2 1 10.36 231.500 #> 3 1 18.38 210.900 #> 4 1 24.70 192.200 #> 5 1 29.76 175.100 #> 6 1 33.84 159.500 #> 7 1 37.15 145.400 #> 8 2 39.89 132.500 #> 9 2 43.20 120.700 #> 10 2 46.01 110.000 #> 11 2 48.40 100.200 #> 12 2 50.43 91.300 #> 13 2 52.15 83.180 #> 14 2 53.62 75.790 #> 15 2 54.87 69.060 #> 16 2 55.94 62.930 #> 17 2 56.84 57.340 #> 18 2 57.61 52.240 #> 19 2 58.27 47.600 #> 20 3 58.92 43.370 #> 21 4 61.09 39.520 #> 22 4 63.71 36.010 #> 23 4 66.01 32.810 #> 24 4 68.03 29.900 #> 25 4 69.80 27.240 #> 26 5 73.21 24.820 #> 27 6 76.71 22.610 #> 28 6 79.84 20.610 #> 29 6 82.50 18.770 #> 30 6 84.77 17.110 #> 31 6 86.72 15.590 #> 32 6 88.41 14.200 #> 33 6 89.87 12.940 #> 34 6 91.13 11.790 #> 35 6 92.23 10.740 #> 36 6 93.19 9.789 #> 37 6 94.03 8.920 #> 38 6 94.76 8.127 #> 39 6 95.40 7.405 #> 40 6 95.96 6.747 #> 41 6 96.45 6.148 #> 42 6 96.88 5.602 #> 43 6 97.27 5.104 #> 44 6 97.60 4.651 #> 45 6 97.89 4.238 #> 46 6 98.15 3.861 #> 47 6 98.38 3.518 #> 48 6 98.58 3.206 #> 49 6 98.75 2.921 #> 50 6 98.91 2.661 #> 51 6 99.04 2.425 #> 52 6 99.16 2.209 #> 53 6 99.27 2.013 #> 54 6 99.36 1.834 #> 55 6 99.44 1.671 #> 56 6 99.51 1.523 #> 57 6 99.57 1.388 #> 58 6 99.63 1.264 #> 59 6 99.68 1.152 #> 60 6 99.72 1.050 #> 61 6 99.75 0.956 #> 62 6 99.78 0.872 #> 63 6 99.81 0.794 #> 64 6 99.84 0.724 #> 65 6 99.86 0.659 #> 66 6 99.87 0.601 #> 67 6 99.89 0.547 #> 68 6 99.90 0.499 ``` Created on 2021-05-04 by the [reprex package](https://reprex.tidyverse.org) (v1.0.0.9000)

github.com/tidymodels/parsnip

multi_predict error message has an error

opened 10:13PM - 04 May 21 UTC

closed 12:25AM - 07 May 21 UTC

topepo

bug next release

This code: ```r multi_predict.default <- function(object, ...) rlang::abo…rt( glue::glue( "No `multi_predict` method exists for objects with classes ", glue::glue_collapse(glue::glue("'{class()}'"), sep = ", ") ) ) ``` gives a wrong error message of > `Error in class(): 0 arguments passed to 'class' which requires 1"` It should probably use `"'{class(object)}'"`

system · May 11, 2021, 10:27pm

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.

Support for glmnet with an external family

First, I try a spec for glmnet with the default family

Now I use a quasipoisson() family

Extract the fitted model and use glmnet predict method instead.

Now I’m using tune functions

First, I try a spec for `glmnet` with the default family

Now I use a `quasipoisson()` family

Now I’m using `tune` functions