def pca_estimator_with_debug_hook():
s3_output_location = 's3://sagemaker/models'
hook_config = DebuggerHookConfig(
s3_output_path='s3://sagemaker/output/debug',
hook_parameters={
"save_interval": "1"
},
collection_configs=[
CollectionConfig("hyperparameters"),
CollectionConfig("metrics")
]
)
rules = [Rule.sagemaker(rule_configs.confusion(),
rule_parameters={
"category_no": "15",
"min_diag": "0.7",
"max_off_diag": "0.3",
"start_step": "17",
"end_step": "19"}
)]
pca = sagemaker.estimator.Estimator(
PCA_IMAGE,
role=EXECUTION_ROLE,
train_instance_count=1,
train_instance_type='ml.c4.xlarge',
output_path=s3_output_location,
debugger_hook_config = hook_config,
rules=rules
)
pca.set_hyperparameters(
feature_dim=50000,
num_components=10,
subtract_mean=True,
algorithm_mode='randomized',
mini_batch_size=200
)
pca.sagemaker_session = MagicMock()
pca.sagemaker_session.boto_region_name = 'us-east-1'
pca.sagemaker_session._default_bucket = 'sagemaker'
return pca
# TODO: Upload source files here given we are not calling fit
debug_hook_config = DebuggerHookConfig(
s3_output_path=debug_output_path,
hook_parameters={"save_interval": "1"},
collection_configs=[
CollectionConfig("hyperparameters"),
CollectionConfig("metrics"),
CollectionConfig("predictions"),
CollectionConfig("labels"),
CollectionConfig("feature_importance")
])
debug_rules = [
Rule.sagemaker(rule_configs.confusion(),
rule_parameters={
"category_no": "15",
"min_diag": "0.7",
"max_off_diag": "0.3",
"start_step": "17",
"end_step": "19"
})
]
hyperparameters = {
"max_depth": "10",
"eta": "0.2",
"gamma": "1",
"min_child_weight": "6",
"silent": "0",