Modifier and Type | Class and Description |
---|---|
class |
BaseEvaluationListener
A base listener class that will preform the provided evaluations, and provide the results in epochEnd and validationDone
Instead of overriding requiredVariables, epochStart, epochEnd, validationDone, and/or opExecution,
override otherRequiredVariables, epochStartEvaluations, epochEndEvaluations, validationDoneEvaluations, and/or opExecutionEvaluations
If you want to use Evaluations in your listener, extend this class
|
class |
BaseListener
A base/abstract
Listener with all methods implemented as no-op. |
Modifier and Type | Class and Description |
---|---|
class |
CheckpointListener
CheckpointListener: The goal of this listener is to periodically save a copy of the model during training..
Model saving may be done: 1. |
Modifier and Type | Class and Description |
---|---|
class |
ArraySavingListener |
class |
ExecDebuggingListener
A listener that logs operation execution for debugging purposes.
|
class |
OpBenchmarkListener
A simple listener for benchmarking single operations in SameDiff
Supports 2 modes: - SINGLE_ITER_PRINT: Print the runtime of the first iteration - AGGREGATE: Collect statistics for multiple runs, that can be accessed (by op name) via #getAggregateModeMap() |
Modifier and Type | Class and Description |
---|---|
class |
HistoryListener
HistoryListener is mainly used internally to collect information such as the loss curve and evaluations,
which will be reported later in a
History instance |
class |
ScoreListener
A listener that reports scores and performance metrics for each epoch.
At every N iterations, the following is reported: (a) Epoch and iteration number (b) Loss value (total loss) (c) ETL time (if > 0) - this represents how long training was blocked waiting for data. |
class |
UIListener
User interface listener for SameDiff
Basic usage: |
Modifier and Type | Method and Description |
---|---|
List<Listener> |
SameDiff.getListeners()
Gets the current SameDiff-wide listeners.
|
Modifier and Type | Method and Description |
---|---|
void |
SameDiff.addListeners(Listener... listeners)
Add SameDiff-wide
Listener instances. |
void |
SameDiff.evaluate(DataSetIterator iterator,
Map<String,IEvaluation> variableEvals,
Listener... listeners)
Evaluation for multiple-output networks.
See SameDiff.evaluate(MultiDataSetIterator, Map, Map, Listener[]) . |
void |
SameDiff.evaluate(MultiDataSetIterator iterator,
Map<String,List<IEvaluation>> variableEvals,
Map<String,Integer> predictionLabelMapping,
Listener... listeners)
Perform evaluation using classes such as
Evaluation for classifier outputs
and RegressionEvaluation for regression outputs.Example: classifier evaluation Predictions variable name: "softmaxOutput" Evaluations to perform: Evaluation Data: single input, single output MultiDataSets Code: |
void |
SameDiff.evaluateMultiple(DataSetIterator iterator,
Map<String,List<IEvaluation>> variableEvals,
Listener... listeners)
Evaluation for multiple output networks - one or more.
|
History |
SameDiff.fit(DataSetIterator iter,
int numEpochs,
DataSetIterator validationIter,
int validationFrequency,
Listener... listeners)
Fit the SameDiff instance based on DataSetIterator for the specified number of epochs.
This method can only be used for singe input, single output SameDiff instances as DataSet only supports a single input and a single output. Note that a TrainingConfig must be set via SameDiff.setTrainingConfig(TrainingConfig) before training can
be performed. |
History |
SameDiff.fit(DataSetIterator iter,
int numEpochs,
Listener... listeners)
See
SameDiff.fit(DataSetIterator, int, DataSetIterator, int, Listener...) , does not preform validation. |
History |
SameDiff.fit(DataSet dataSet,
Listener... listeners)
Fit the SameDiff instance based on a single DataSet (i.e., a single minibatch for one iteration).
This method can only be used for singe input, single output SameDiff instances as DataSet only supports a single input and a single output. Note that a TrainingConfig must be set via SameDiff.setTrainingConfig(TrainingConfig) before training can
be performed. |
protected History |
SameDiff.fit(MultiDataSetIterator iter,
int numEpochs,
boolean incrementEpochCount,
MultiDataSetIterator validationData,
int validationFrequency,
Listener... listeners) |
History |
SameDiff.fit(MultiDataSetIterator iter,
int numEpochs,
Listener... listeners)
See
SameDiff.fit(MultiDataSetIterator, int, MultiDataSetIterator, int, Listener...) , does not preform validation. |
History |
SameDiff.fit(MultiDataSetIterator iter,
int numEpochs,
MultiDataSetIterator validationIter,
int validationFrequency,
Listener... listeners)
Fit the SameDiff instance based on MultiDataSetIterator for the specified number of epochs.
This method can both singe input, single output and multi-input, multi-output SameDiff instances Note that a TrainingConfig must be set via SameDiff.setTrainingConfig(TrainingConfig) before training can
be performed. |
History |
SameDiff.fit(MultiDataSet dataSet,
Listener... listeners)
Fit the SameDiff instance based on a single MultiDataSet (i.e., a single minibatch for one iteration).
Note that a TrainingConfig must be set via SameDiff.setTrainingConfig(TrainingConfig) before training can
be performed. |
void |
SameDiff.setListeners(Listener... listeners)
Set the current SameDiff-wide
Listener instances. |
Modifier and Type | Method and Description |
---|---|
void |
SameDiff.addListeners(Collection<? extends Listener> listeners)
|
protected Map<String,INDArray> |
SameDiff.batchOutputHelper(Map<String,INDArray> placeholders,
List<Listener> listeners,
Operation operation,
String... outputs) |
protected Map<String,INDArray> |
SameDiff.directExecHelper(Map<String,INDArray> placeholders,
At at,
MultiDataSet batch,
Collection<String> requiredActivations,
List<Listener> activeListeners,
String... outputs)
Do inference for the given variables for a single batch, with training information
|
void |
SameDiff.evaluate(DataSetIterator iterator,
String outputVariable,
List<Listener> listeners,
IEvaluation... evaluations)
Evaluate the performance of a single variable's prediction.
For example, if the variable to evaluatate was called "softmax" you would use: |
void |
SameDiff.evaluate(MultiDataSetIterator iterator,
String outputVariable,
int labelIndex,
List<Listener> listeners,
IEvaluation... evaluations)
Evaluate the performance of a single variable's prediction.
For example, if the variable to evaluatate was called "softmax" you would use: |
protected History |
SameDiff.fitHelper(MultiDataSetIterator iter,
int numEpochs,
boolean incrementEpochCount,
MultiDataSetIterator validationData,
int validationFrequency,
List<Listener> listeners) |
Map<String,INDArray> |
SameDiff.output(DataSetIterator iterator,
List<Listener> listeners,
String... outputs)
Do inference on a network with a single input.
For example, if the variable to infer was called "softmax" you would use: |
Map<String,INDArray> |
SameDiff.output(Map<String,INDArray> placeholders,
List<Listener> listeners,
String... outputs)
Do inference for the given variables for a single batch.
|
Map<String,INDArray> |
SameDiff.output(MultiDataSetIterator iterator,
List<Listener> listeners,
String... outputs)
Perform inference.
Example: classifier inference Predictions variable name: "softmaxOutput" Evaluations to perform: Evaluation Data: single output MultiDataSets Code: |
List<Map<String,INDArray>> |
SameDiff.outputBatches(DataSetIterator iterator,
List<Listener> listeners,
String... outputs)
See
SameDiff.output(DataSetIterator, List, String...) , but without the concatenation of batches. |
List<Map<String,INDArray>> |
SameDiff.outputBatches(MultiDataSetIterator iterator,
List<Listener> listeners,
String... outputs)
Perform inference.
Example: classifier inference Predictions variable name: "softmaxOutput" Evaluations to perform: Evaluation Data: single output MultiDataSets Code: |
void |
SameDiff.setListeners(Collection<? extends Listener> listeners)
|
Modifier and Type | Method and Description |
---|---|
EvaluationConfig |
EvaluationConfig.listeners(Listener... listeners)
Add listeners for this operation
|
BatchOutputConfig |
BatchOutputConfig.listeners(Listener... listeners)
Add listeners for this operation
|
FitConfig |
FitConfig.listeners(Listener... listeners)
Add listeners for this operation
|
OutputConfig |
OutputConfig.listeners(Listener... listeners)
Add listeners for this operation
|
Modifier and Type | Field and Description |
---|---|
protected List<Listener> |
TrainingSession.listeners |
Modifier and Type | Method and Description |
---|---|
abstract T[] |
AbstractSession.getOutputs(O op,
AbstractSession.FrameIter outputFrameIter,
Set<AbstractSession.VarId> inputs,
Set<AbstractSession.VarId> allIterInputs,
Set<String> constAndPhInputs,
List<Listener> listeners,
At at,
MultiDataSet batch,
Set<String> allReqVariables)
Execute the op - calculate INDArrays, or shape info, etc
|
INDArray[] |
TrainingSession.getOutputs(SameDiffOp op,
AbstractSession.FrameIter outputFrameIter,
Set<AbstractSession.VarId> opInputs,
Set<AbstractSession.VarId> allIterInputs,
Set<String> constAndPhInputs,
List<Listener> listeners,
At at,
MultiDataSet batch,
Set<String> allReqVariables) |
INDArray[] |
InferenceSession.getOutputs(SameDiffOp op,
AbstractSession.FrameIter outputFrameIter,
Set<AbstractSession.VarId> opInputs,
Set<AbstractSession.VarId> allIterInputs,
Set<String> constAndPhInputs,
List<Listener> listeners,
At at,
MultiDataSet batch,
Set<String> allReqVariables) |
Map<String,T> |
AbstractSession.output(List<String> variables,
Map<String,T> placeholderValues,
MultiDataSet batch,
Collection<String> requiredActivations,
List<Listener> listeners,
At at)
Get the output of the session - i.e., perform inference/forward pass and return the autputs for the specified variables
|
Loss |
TrainingSession.trainingIteration(TrainingConfig config,
Map<String,INDArray> placeholders,
Set<String> paramsToTrain,
Map<String,GradientUpdater> updaters,
MultiDataSet batch,
List<String> lossVariables,
List<Listener> listeners,
At at)
Perform one iteration of training - i.e., do forward and backward passes, and update the parameters
|
Modifier and Type | Class and Description |
---|---|
class |
ActivationGradientCheckListener
A listener used for debugging and testing purposes - specifically for gradient checking activations internally in
GradCheckUtil . |
Modifier and Type | Class and Description |
---|---|
class |
NonInplaceValidationListener |
Copyright © 2019. All rights reserved.