guyernest

# Read CSV into R
TagsPerMonth <- read.csv(file="../StackOverFlowData.csv", header=TRUE, sep=",")

amazonTags <- c("amazon-web-services", "amazon", "amazon-product-api", "amazon-mws", 
                "amazon-appstore", "amazon-echo", "amazon-fire-tv", "amazon-payments", 
               "amazonica", "amazon-silk", "amazon-marketplace", "amazonads", "login-with-amazon",
               "amazon-echo-show", "amazon-mobile-ads", "amazon-clouddrive", "amazon-firefly",
               "amazon-in", "amazonsellercentral")
#awsPopularTags <- c("amazon-s3", "amazon-ec2")
topTagsPerMonth <- subset(TagsPerMonth, !TagName %in% amazonTags)

guyernest

import boto3
import logging

logger = logging.getLogger()
logger.setLevel(logging.INFO)

client = boto3.client('sagemaker')

def lambda_handler(event, context):
    

guyernest

%%writefile lambda.tf
provider "aws" {
  region = "eu-west-1"
}

resource "aws_lambda_function" "start_stop_sm" {
  function_name = "StopStartSageMakerNotebookInstances"

  # The bucket name as created earlier with "aws s3api create-bucket"
  s3_bucket = "terraform-serverless-repository"

guyernest

%%writefile -a lambda.tf
# IAM role which dictates what other AWS services the Lambda function
# may access.
## Testing a working example
resource "aws_iam_role" "lambda_exec" {
    name = "assume-role"
    assume_role_policy = <<EOF
{
    "Version": "2012-10-17",
    "Statement": [

guyernest

%%writefile cloud_watch.tf
## Based on the location of the instances. This is for Israel where 
## people are working Sunday to Thursday and 5AM GTM is 8AM
resource "aws_cloudwatch_event_rule" "on_duty" {
    name = "on_duty"
    description = "Fires at the beginning of the working day"
    schedule_expression = "cron(0 5 ? * SUN-THU *)"
}

## people are working Sunday to Thursday and 4PM GTM is 7PM

guyernest

guyernest

/*train_1217_category training data aggregate by location and category*/
CREATE TABLE forecast.train_1217_category
WITH (
  format='TEXTFILE',
  external_location='s3://forecast-xxxxxx/sagemaker/train_1217_category/',
  field_delimiter = ','
) AS
SELECT category as item_id, date_format(calendar_date, '%Y-%m-%d') as timestamp, sum(sales) as demand 
FROM forecast.compressed_data
where calendar_date < CAST('2018-01-01' AS DATE)

guyernest

/*Original Data Compressed*/
CREATE TABLE forecast.compressed_data
WITH (
  format='PARQUET',
  external_location='s3://forecast-xxxxxx/sagemaker/data/',
  partitioned_by = ARRAY['year']
) AS
SELECT *, year(calendar_date) as year FROM forecast.data;

guyernest

import boto3
session = boto3.Session(region_name='us-east-1') 
forecast = session.client(service_name='forecast')
forecastquery = session.client(service_name='forecastquery')

# The name of the dataset that we created with Athena
train_dataset_table = 'train_1217_category'

DATASET_FREQUENCY = "D" 
TIMESTAMP_FORMAT = "yyyy-MM-dd"

guyernest

predictorName= project+'_ARIMA'

# We will create forecast prediction for two months (60 days)
forecastHorizon = 60

# Starting the predictor creation job
createPredictorResponse=forecast.create_predictor(
    RecipeName=recipe,
    DatasetGroupName= datasetGroupName ,
    PredictorName=predictorName,