configuration-aml.variables.yml

# Copyright (c) Microsoft Corporation. All rights reserved.
# Licensed under the MIT License.

# Definition of AML-related environment variables
# There are two versions: One is for using multiple classifiers.  The other is for using one best classifier.

# If the data scientist has already picked a best classifier, you can change CLASSIFIER_NUMBER to 1 
#  to use "train_1_classifier.py".  Otherwise, you can change CLASSIFIER_NUMBER to n to 
#   use "train_n_classifier.py" which will use and compare multiple classifers and will automatically pick up the best one for you.

# You can change the variable names to fit your naming convention.

variables:

  PYTHON_VERSION: 3.7
  SDK_VERSION: 1.27

  # Input data
  AML_DATASET: two_class
  KEEP_COLUMNS: Helpfulness_Score|Score|Text|Target
  TARGET_COLUMN: Target
  # The value for the true or positive case should go first!!!
  TARGET_VALUES: toys games|not a toy/game
  TEXT_COLUMNS: Text
  # The number of folds to be used in the cross validation: 1-10
  FOLD_NUMBER: 4
  # The number of classifiers to be used in the model training: 1 or n
  CLASSIFIER_NUMBER: 1

  # Training
  TRAINING_SCRIPT_NAME: train_$(CLASSIFIER_NUMBER)_classifier.py
  AML_MODEL_NAME: two_class_$(CLASSIFIER_NUMBER).pkl
  AML_TRAINING_EXPERIMENT: two-class-$(CLASSIFIER_NUMBER)-experiment
  AML_TRAINING_PIPELINE: two-class-$(CLASSIFIER_NUMBER)-pipeline
  AML_TRAINING_COMPUTE: cpucompute
  AML_TRAINING_ENV_PATH: configuration/environments/environment_training
  TRAINING_MODEL_METRIC_NAME: Recall
  TRAINING_MAXIMIZE: True

  # Batch inference
  AML_BATCHINFERENCE_EXPERIMENT: two-class-$(CLASSIFIER_NUMBER)-bi-experiment
  AML_BATCHINFERENCE_PIPELINE: two-class-$(CLASSIFIER_NUMBER)-bi-pipeline
  AML_BATCHINFERENCE_COMPUTE: $(AML_TRAINING_COMPUTE)
  AML_BATCHINFERENCE_ENV_PATH: configuration/environments/environment_inference
  BATCHINFERENCE_OUTPUT_DIR: forecasting_results
  BATCHINFERENCE_OUTPUT_CONTAINER: modelprediction

  # Real-time inference
  AML_REALTIMEINFERENCE_ENV_PATH: $(AML_BATCHINFERENCE_ENV_PATH)
  AKS_COMPUTE: aks-compute
  AML_WEBSERVICE: two-class-$(CLASSIFIER_NUMBER)-webservice