UseMethod("predict") : no applicable method for 'predict' applied to an object of class "train"

Question

I have a model (fit), based on historic information until last month. Now I would like to predict using my model for the current month. When I try to invoke the following code:

predicted <- predict(fit, testData[-$Readmit])

I get the following error:

Error in UseMethod("predict") : no applicable method for 'predict'
    applied to an object of class "train"

Notes:

1. The fit model was created via: train function from caret package, using random forest algorithm

2. The predict is a generic function that will invoke the specific predict function based on the first input argument. In my case it will be:

   >fit$modelInfo$label

   [1] "Random Forest"

Therefore the predict method invoked will be: predict.randomForest. See [caret documentation][3] for more info.

Here the summary source code for generating the model and invoking it:

# Script-1: create a model:
fit <- train(testData[-$Readmit], testData$Readmit)
saveRDS(fit, modelFileName) # save the fit object into a file

# Script-2: predict
fit <- readRDS(modelFileName) # Load the model (generated previously)
predicted <- predict(fit, testData[-$Readmit])

Note: The execution time for generating the model is about 3 hours, that is why I save the object for reusing after that.

The data set from the training model as the following structure:

> str(fit$trainingData)
'data.frame':   29955 obs. of  27 variables:
$ Acuity                : Factor w/ 3 levels "Elective  ","Emergency ",..: 2 2 2 1 1 2 2 2 1 1 ...
$ AgeGroup              : Factor w/ 10 levels "100-105","65-70",..: 8 6 9 9 5 4 9 2 3 2 ...
$ IsPriority            : int  0 0 0 0 0 0 0 0 0 0 ...
$ QNXTReferToId         : int  115 1703712 115 3690 1948 115 109 512 481 1785596 ...
$ QNXTReferFromId       : int  1740397 1724801 1711465 1704170 1714272 1731911 1535 1712758 1740614 1760252 ...
$ iscasemanagement      : Factor w/ 2 levels "N","Y": 2 1 1 2 2 1 2 1 2 2 ...
$ iseligible            : Factor w/ 2 levels "N","Y": 2 2 2 2 2 2 2 2 2 2 ...
$ referralservicecode   : Factor w/ 11 levels "12345","278",..: 1 1 1 9 9 1 1 6 9 9 ...
$ IsHighlight           : Factor w/ 2 levels "N","Y": 1 1 1 1 1 1 1 1 1 1 ...
$ admittingdiagnosiscode: num  439 786 785 786 428 ...
$ dischargediagnosiscode: num  439 0 296 786 428 ...
$ RealLengthOfStay      : int  3 1 6 1 2 3 3 7 3 2 ...
$ QNXTPCPId             : int  1740397 1724801 1711465 1704170 1714272 1731911 1535 1712758 1740614 1760252 ...
$ QNXTProgramId         : Factor w/ 3 levels "QMXHPQ0839     ",..: 1 1 1 1 1 1 1 1 1 1 ...
$ physicalzipcode       : int  33054 33712 33010 33809 33010 33013 33142 33030 33161 33055 ...
$ gender                : Factor w/ 2 levels "F","M": 1 1 1 1 2 1 1 2 2 1 ...
$ ethnicitycode         : Factor w/ 4 levels "ETHN0001       ",..: 4 4 4 4 4 4 4 4 4 4 ...
$ dx1                   : num  439 786 296 786 428 ...
$ dx2                   : num  439 292 785 786 428 ...
$ dx3                   : num  402 0 250 0 0 ...
$ svc1                  : int  0 120 120 762 762 120 120 120 762 762 ...
$ svc2                  : int  120 0 0 0 0 0 0 0 0 0 ...
$ svc3                  : int  0 0 0 0 0 0 0 0 0 0 ...
$ Disposition           : Factor w/ 28 levels "0","APPEAL & GRIEVANCE REVIEW                                   ",..: 11 11 16 11 11 11 11 11 11 11 ...
$ AvgIncome             : Factor w/ 10 levels "-1",">100k","0-25k",..: 3 6 3 8 3 4 3 5 4 4 ...
$ CaseManagerNameID     : int  124 1 1 19 20 1 16 1 43 20 ...
$ .outcome              : Factor w/ 2 levels "NO","YES": 1 2 2 1 1 1 2 2 1 1    ...

now the testData will have the following structure:

> str(testData[-$Readmit])
'data.frame':   610 obs. of  26 variables:
$ Acuity                : Factor w/ 4 levels "0","Elective  ",..: 3 2 4 2 2 2 4 3 3 3 ...
$ AgeGroup              : Factor w/ 9 levels "100-105","65-70",..: 4 3 5 4 2 9 4 2 4 6 ...
$ IsPriority            : int  0 0 0 0 0 0 1 1 1 1 ...
$ QNXTReferToId         : int  2140 482 1703785 1941 114 1714905 1703785 98 109 109 ...
$ QNXTReferFromId       : int  1791383 1729375 1718532 1746336 1718267 1718267 1718532 98 109 109 ...
$ iscasemanagement      : Factor w/ 2 levels "N","Y": 2 2 2 2 2 2 1 2 2 1 ...
$ iseligible            : Factor w/ 2 levels "N","Y": 2 2 2 2 2 2 2 2 2 2 ...
$ referralservicecode   : Factor w/ 7 levels "12345","IPMAT          ",..: 5 1 1 1 1 1 1 5 1 5 ...
$ IsHighlight           : Factor w/ 2 levels "N","Y": 1 1 1 1 1 1 1 1 1 1 ...
$ admittingdiagnosiscode: num  11440 11317 11420 11317 1361 ...
$ dischargediagnosiscode: num  11440 11317 11420 11317 1361 ...
$ RealLengthOfStay      : int  1 2 4 3 1 1 16 1 1 3 ...
$ QNXTPCPId             : int  3212 1713678 1738430 1713671 1720569 1791640 1725962 1148 1703290 1705009 ...
$ QNXTProgramId         : Factor w/ 2 levels "QMXHPQ0839     ",..: 1 1 1 1 1 1 1 1 1 1 ...
$ physicalzipcode       : int  34744 33175 33844 33178 33010 33010 33897 33126 33127 33125 ...
$ gender                : Factor w/ 2 levels "F","M": 2 1 2 1 2 2 2 1 1 2 ...
$ ethnicitycode         : Factor w/ 1 level "No Ethnicity   ": 1 1 1 1 1 1 1 1 1 1 ...
$ dx1                   : num  11440 11317 11420 11317 1361 ...
$ dx2                   : num  11440 11317 11420 11317 1361 ...
$ dx3                   : num  0 1465 0 11326 0 ...
$ svc1                  : int  52648 27447 50040 27447 55866 55866 51595 0 99221 300616 ...
$ svc2                  : int  76872 120 50391 120 120 38571 120 762 120 0 ...
$ svc3                  : int  762 0 120 0 0 51999 0 0 0 762 ...
$ Disposition           : Factor w/ 14 levels "0","DENIED- Not Medically Necessary                             ",..: 3 5 3 4 3 3 5 3 3 5 ...
$ AvgIncome             : Factor w/ 10 levels "-1",">100k","0-25k",..: 6 7 5 9 3 3 6 4 3 4 ...
$ CaseManagerNameID     : int  1 2 3 4 5 6 7 8 9 7 ...

The variable structure is the same, just that some factor variables has different levels because some variable has new values. For example: Acuity in the model has 3-levels and in the testing data 4-levels.

I don't have from upfront a way to know all possible level for all variables.

Any advice, please...

Thanks in advance,

David

Answer 1

I think I found why this happened...The predict is a generic function from: stats package. I use the namespace ::-notation for invoking the functions from the caret package (that is the recommendation for creating a user packages) and the equivalent predict function from caret package is: predict.train, that is an internal function, that cannot be invoked by an external application. The only way to invoke this function, is using the generic predict function from stats package, then based on the class of the first input argument: predicted <- predict(fit, testData[-$Readmit]) it identifies the particular predict function will be invoked.

For this particular case the class of this function is train, so it would call actually the function: train.predict from caret package. This function also handles the particular function requested for prediction based on the algorithm (method) used, for example: predict.gbm or predict.glm, etc. It is explained, in detail, in the caret documentation section: "5.7 Extracting Predictions and Class Probabilities".

Therefore the ::-notation works well for other functions in the package, such as: caret.train for example, but not for this particular one: predict. In such cases it is necessary to explicitly load the library, so it internally can invoke predict.train function.

In short, the solution is just adding the following line before invoking the predict function:

library(caret)

Then error disappears.

Answer 2

Based on the answer from @David Leal, I tried loading library(caret) before calling the predict function but it did not help.

After trying a bit, I realized that I had to load the library that contains the model itself. In my case, I had to call library(kenlab) for Support Vectors.