SageMaker Training
================

## SageMaker Training

Clear the workspace

``` r
rm(list = ls())
```

``` bash
python -m pip install -U sagemaker
```

## Imports

``` r
suppressWarnings(library(reticulate))
sagemaker <- import('sagemaker')

role = sagemaker$get_execution_role()
session = sagemaker$Session()
s3_output = paste("s3://", session$default_bucket(), sep="")
bucket <- session$default_bucket()

account_id <- session$account_id()
region <- session$boto_region_name
```

``` r
sagemaker$s3$S3Downloader$download("s3://sagemaker-sample-files/datasets/tabular/iris/iris.data","dataset")

data <- civit_gps <- read.csv(file="dataset/iris.data",head=FALSE,sep=",")
newheaders <- c("Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width", "Species")
colnames(data) <- newheaders
write.csv(data, "dataset/iris.data", row.names = FALSE)


s3_train <- session$upload_data(path = 'dataset/iris.data',
                                bucket = bucket,
                                key_prefix = 'data')
```

``` r
head(data)
```

    ##   Sepal.Length Sepal.Width Petal.Length Petal.Width     Species
    ## 1          5.1         3.5          1.4         0.2 Iris-setosa
    ## 2          4.9         3.0          1.4         0.2 Iris-setosa
    ## 3          4.7         3.2          1.3         0.2 Iris-setosa
    ## 4          4.6         3.1          1.5         0.2 Iris-setosa
    ## 5          5.0         3.6          1.4         0.2 Iris-setosa
    ## 6          5.4         3.9          1.7         0.4 Iris-setosa

``` r
container_uri <- paste(account_id, "dkr.ecr", region, "amazonaws.com/sagemaker-r-training:1.0", sep=".")

# https://docs.aws.amazon.com/sagemaker/latest/dg/your-algorithms-training-algo-dockerfile.html
# Estimator
estimator <- sagemaker$estimator$Estimator(image_uri = container_uri,
                                           base_job_name="train-iris-mars",
                                           entry_point = "training.R",
                                           role = role,
                                           metric_definitions = list(list("Name" = "mse", "Regex" = "mse: ([0-9\\.]+)")),
                                           instance_count = 1L,
                                           instance_type = 'ml.m5.4xlarge',
                                           volume_size = 30L,
                                           max_run = 3600L,
                                           input_mode = 'File')
```

``` r
# Train the estimator
s3_train_input <- sagemaker$inputs$TrainingInput(s3_data = s3_train,
                                     content_type = 'csv')
estimator$fit(inputs=list("train" = s3_train_input), logs=TRUE)
```

``` r
hyperparameter_ranges = list('thresh' = sagemaker$parameter$ContinuousParameter(0.001, 0.01),
                             'prune'= sagemaker$parameter$CategoricalParameter(list(TRUE, FALSE)))

objective_metric_name = "mse"
metric_definitions = list(list("Name" = "mse", "Regex" = "mse: ([0-9\\.]+)"))

# Create a hyperparameter tuner
tuner <- sagemaker$tuner$HyperparameterTuner(estimator,
                                             objective_metric_name,
                                             hyperparameter_ranges,
                                             metric_definitions,
                                             strategy='Bayesian',
                                             objective_type='Minimize',
                                             max_jobs=4L,
                                             max_parallel_jobs=2L)

job_name <- paste('tune-mars', format(Sys.time(), '%Y%m%d-%H-%M-%S'), sep = '-')

tuner$fit(inputs=list("train" = s3_train_input), wait=TRUE, job_name = job_name)
```

``` r
best_estimator <- tuner$best_estimator()
```

``` r
inference_contianer_uri <- paste(account_id, "dkr.ecr", region, "amazonaws.com/sagemaker-r-inference:1.0", sep=".")

trained_model <- best_estimator$create_model(name='r-iris-model', role=role, image_uri = container_uri)
endpoint  <- trained_model$deploy(initial_instance_count = 1L,
                                  instance_type = "ml.t2.medium",
                                  serializer = sagemaker$serializers$CSVSerializer(content_type='text/csv'),
                                  deserializer = sagemaker$deserializers$JSONDeserializer())
```

Test our endpoint

``` r
test_sample <- unlist(data[1,])
probabilities <- endpoint$predict(test_sample)
predicted_class <- which.max(as.numeric(unlist(strsplit(probabilities, ","))))

print(probabilities)
```

    ## [1] "0.506412021025866,0.950153510028136,-0.500864199734607,-0.500864199734607"

``` r
print(predicted_class)
```

    ## [1] 2

Delete the endpoint when done

``` r
endpoint$delete_endpoint(delete_endpoint_config=TRUE)
```