train_dl_model_anomaly_dataset🔗
Short description🔗
train_dl_model_anomaly_dataset — Train a deep learning model for anomaly detection.
Signature🔗
train_dl_model_anomaly_dataset( dl_model DLModelHandle, dict DLSamples, dict DLTrainParam, out dict DLTrainResult )
Description🔗
The operator train_dl_model_anomaly_dataset performs the training
of a deep learning model with 'type'='anomaly_detection'
contained in DLModelHandle (for deep learning models
with 'type'='gc_anomaly_detection' see
train_dl_model_batch).
This operator processes the full training dataset at once.
This is in contrast to the operator train_dl_model_batch.
The iterations over the dataset are performed internally by the operator.
Consequently, you only need to call this operator once with the full training
dataset to train your anomaly detection model.
The training dataset is handed over in the tuple of dictionaries
DLSamples.
See the chapter Deep Learning / Model for further information to the
used dictionaries and their keys.
The operator expects within the training dataset only images without anomaly
to train the anomaly detection model.
The dictionary DLTrainParam can be used to change the
hyperparameters.
The following values are supported:
-
max_num_epochs: This parameter specifies the maximum number of epochs performed during training. In case the criterion specified byerror_thresholdis reached in an earlier epoch, the training will terminate regardless.Restriction:
max_num_epochs>=1.Default:
max_num_epochs= 30. -
error_threshold: This parameter is a termination criterion for the training. If the training error is less than the specifiederror_threshold, the training terminates successfully.Restriction: 0.0 <=
error_threshold<= 1.0.Default:
error_threshold= 0.001. -
domain_ratio: This parameter determines the percentage of information of each image used for training. Since images tend to contain an abundance of information, it is advisable to reduce its amount. Additionally, reducingdomain_ratiocan decrease the time needed for training. Please note, however, sufficient information needs to remain and therefore this value should not be set too small either. Otherwise the training result might not be satisfactory or the training itself might even fail.Restriction: 0.0 <
domain_ratio<= 1.0.Default:
domain_ratio= 0.1. -
regularization_noise: This parameter can be set to regularize the training in order to improve robustness.Restriction:
regularization_noise>=0.0.Default:
regularization_noise= 0.0.
The output dictionary DLTrainResult contains the following values:
-
final_error: The best error received during training. -
final_epochThe epoch in which the errorfinal_errorwas achieved.
Attention🔗
The operator train_dl_model_anomaly_dataset internally calls functions
that might not be deterministic.
Therefore, results from multiple calls of
train_dl_model_anomaly_dataset can slightly differ, although the same
input values have been used.
System requirements:
To run this operator on GPU by setting 'runtime' to 'gpu'
(see get_dl_model_param), cuDNN and cuBLAS are required.
For further details, please refer to the “Installation Guide”,
paragraph “Requirements for Deep Learning and Deep-Learning-Based Methods”.
Alternatively, this operator
can also be run on CPU by setting 'runtime' to 'cpu'.
Execution information🔗
Execution information
-
Multithreading type: reentrant (runs in parallel with non-exclusive operators).
-
Multithreading scope: global (may be called from any thread).
-
Automatically parallelized on internal data level.
Parameters🔗
DLModelHandle (input_control) dl_model → (handle)
Deep learning model handle.
DLSamples (input_control) dict-array → (handle)
Tuple of Dictionaries with input images and corresponding information.
DLTrainParam (input_control) dict → (handle)
Parameter for training the anomaly detection model.
Default: []
DLTrainResult (output_control) dict → (handle)
Dictionary with the train result data.
Result🔗
If the parameters are valid, the operator
train_dl_model_anomaly_dataset returns the value 2 (H_MSG_TRUE). If necessary,
an exception is raised.
Combinations with other operators🔗
Combinations
Possible predecessors
read_dl_model, set_dl_model_param, get_dl_model_param
Possible successors
See also
Module🔗
This operator uses dynamic licensing (see the ‘Installation Guide’). Which of the following modules is required depends on the specific usage of the operator:
Deep Learning Professional