> ## Documentation Index > Fetch the complete documentation index at: https://docs.edgeimpulse.com/llms.txt > Use this file to discover all available pages before exploring further. # Train model (Keras) > Take the output from a DSP block and train a neural network using Keras. Updates are streamed over the websocket API. ## OpenAPI ````yaml /.assets/openapi.yaml post /api/{projectId}/jobs/train/keras/{learnId} openapi: 3.0.0 info: title: Edge Impulse API version: 1.0.0 servers: - url: https://studio.edgeimpulse.com/v1 security: - ApiKeyAuthentication: [] - JWTAuthentication: [] - JWTHttpHeaderAuthentication: [] - OAuth2: [] paths: /api/{projectId}/jobs/train/keras/{learnId}: post: tags: - Jobs summary: Train model (Keras) description: >- Take the output from a DSP block and train a neural network using Keras. Updates are streamed over the websocket API. operationId: trainKerasJob parameters: - $ref: '#/components/parameters/ProjectIdParameter' - $ref: '#/components/parameters/LearnIdParameter' requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/SetKerasParameterRequest' responses: '200': description: OK content: application/json: schema: $ref: '#/components/schemas/StartJobResponse' components: parameters: ProjectIdParameter: name: projectId in: path required: true description: Project ID schema: type: integer LearnIdParameter: name: learnId in: path required: true description: Learn Block ID, use the impulse functions to retrieve the ID schema: type: integer schemas: SetKerasParameterRequest: type: object description: Only fields defined in this object are set properties: mode: type: string description: Whether to use visual or expert mode. enum: - expert - visual minimumConfidenceRating: type: number description: >- DEPRECATED, use "setImpulseThresholds" instead. Minimum confidence score, if the neural network scores a sample below this threshold it will be flagged as uncertain. selectedModelType: $ref: '#/components/schemas/KerasModelTypeEnum' description: The model type to select, as described in the model metadata call. script: type: string description: Raw Keras script (only used in expert mode) visualLayers: type: array description: The visual layers for the neural network (only in visual mode). items: $ref: '#/components/schemas/KerasVisualLayer' trainingCycles: type: integer description: Number of training cycles (only in visual mode). learningRate: type: number description: Learning rate (between 0 and 1) (only in visual mode). batchSize: type: number description: Batch size used during training (only in visual mode). trainTestSplit: type: number description: Train/test split (between 0 and 1) autoClassWeights: type: boolean description: >- Whether to automatically balance class weights, use this for skewed datasets. useLearnedOptimizer: type: boolean description: Use learned optimizer and ignore learning rate. augmentationPolicyImage: $ref: '#/components/schemas/AugmentationPolicyImageEnum' augmentationPolicySpectrogram: $ref: '#/components/schemas/AugmentationPolicySpectrogram' profileInt8: type: boolean description: Whether to profile the i8 model (might take a very long time) skipEmbeddingsAndMemory: type: boolean description: >- If set, skips creating embeddings and measuring memory (used in tests) akidaEdgeLearningConfig: $ref: '#/components/schemas/AkidaEdgeLearningConfig' customValidationMetadataKey: type: string description: >- If the 'custom validation split' experiment is enabled, this metadata key is used to prevent group data leakage between train and validation datasets. showAdvancedTrainingSettings: type: boolean description: >- Whether the 'Advanced training settings' UI element should be expanded. showAugmentationTrainingSettings: type: boolean description: >- Whether the 'Augmentation training settings' UI element should be expanded. customParameters: type: object description: >- Training parameters, this list depends on the list of parameters that the model exposes. additionalProperties: type: string nullable: true anomalyCapacity: $ref: '#/components/schemas/AnomalyCapacity' description: >- Capacity level for visual anomaly detection. Determines which set of default configurations to use. The higher capacity, the higher number of (Gaussian) components, and the more adapted the model becomes to the original distribution lastShownModelVariant: $ref: '#/components/schemas/KerasModelVariantEnum' description: >- Last shown variant on the Keras screen. Used to keep the same view after refreshing. blockParameters: $ref: '#/components/schemas/BlockParameters' description: >- Training parameters specific to the type of the learn block. Parameters may be adjusted depending on the model defined in the visual layers. Used for our built-in blocks. trainingProcessor: $ref: '#/components/schemas/ExperimentalImpulseSpecificGpuTrainingProcessor' description: EXPERIMENTAL. Per-learn-block training processor selection. StartJobResponse: allOf: - $ref: '#/components/schemas/GenericApiResponse' - type: object required: - id properties: id: type: integer description: Job identifier. Status updates will include this identifier. example: 12873488112 KerasModelTypeEnum: type: string enum: - int8 - float32 - akida - requiresRetrain KerasVisualLayer: type: object required: - type properties: type: $ref: '#/components/schemas/KerasVisualLayerType' neurons: type: integer description: >- Number of neurons or filters in this layer (only for dense, conv1d, conv2d) or in the final conv2d layer (only for transfer layers) kernelSize: type: integer description: Kernel size for the convolutional layers (only for conv1d, conv2d) dropoutRate: type: number description: >- Fraction of input units to drop (only for dropout) or in the final layer dropout (only for transfer layers) columns: type: integer description: Number of columns for the reshape operation (only for reshape) stack: type: integer description: >- Number of convolutional layers before the pooling layer (only for conv1d, conv2d) enabled: type: boolean organizationModelId: type: integer description: >- Custom transfer learning model ID (when type is set to transfer_organization) AugmentationPolicyImageEnum: type: string description: The data augmentation policy to use with image input enum: - none - all AugmentationPolicySpectrogram: type: object required: - enabled properties: enabled: type: boolean description: >- True if spectrogram augmentation is enabled. Other properties will be ignored if this is false. warping: type: boolean description: True if warping along the time axis is enabled. freqMasking: type: string enum: - none - low - high description: The amount of frequency masking to apply. timeMasking: type: string enum: - none - low - high description: The amount of time masking to apply. gaussianNoise: type: string enum: - none - low - high description: The amount of Gaussian noise to add. AkidaEdgeLearningConfig: type: object required: - enabled properties: enabled: type: boolean description: >- True if Akida Edge Learning model creation is enabled. Other properties will be ignored if this is false. additionalClasses: type: number description: >- Number of additional classes that will be added to the Edge Learning model. neuronsPerClass: type: number description: >- Number of neurons in each class on the last layer in the Edge Learning model. AnomalyCapacity: type: string description: >- Capacity level for visual anomaly detection. Determines which set of default configurations to use. The higher capacity, the higher number of (Gaussian) components, and the more adapted the model becomes to the original distribution enum: - low - medium - high KerasModelVariantEnum: type: string enum: - int8 - float32 - akida BlockParameters: description: >- Training parameters specific to the type of the learn block. Parameters may be adjusted depending on the model defined in the visual layers. Used for our built-in blocks. oneOf: - $ref: '#/components/schemas/BlockParamsVisualAnomalyPatchcore' - $ref: '#/components/schemas/BlockParamsVisualAnomalyGmm' ExperimentalImpulseSpecificGpuTrainingProcessor: type: string description: > EXPERIMENTAL: Training processor for the impulse-specific-gpu rollout. This field is currently gated by a project experiment and may change in future API versions. enum: - cpu - gpu GenericApiResponse: type: object required: - success properties: success: type: boolean description: Whether the operation succeeded error: type: string description: Optional error description (set if 'success' was false) KerasVisualLayerType: type: string enum: - dense - conv1d - conv2d - reshape - flatten - dropout - batchNormalization - transfer_mobilenetv2_a35 - transfer_mobilenetv2_a1 - transfer_mobilenetv2_a05 - transfer_mobilenetv2_160_a1 - transfer_mobilenetv2_160_a75 - transfer_mobilenetv2_160_a5 - transfer_mobilenetv2_160_a35 - transfer_mobilenetv1_a25_d100 - transfer_mobilenetv1_a2_d100 - transfer_mobilenetv1_a1_d100 - transfer_kws_mobilenetv1_a1_d100 - transfer_kws_mobilenetv2_a35_d100 - transfer_kws_syntiant_ndp10x - transfer_kws_conv2d_tiny - object_ssd_mobilenet_v2_fpnlite_320x320 - fomo_mobilenet_v2_a01 - fomo_mobilenet_v2_a35 - transfer_organization - transfer_akidanet_imagenet_160_a100 - transfer_akidanet_imagenet_160_a50 - transfer_akidanet_imagenet_160_a25 - transfer_akidanet_imagenet_224_a100 - transfer_akidanet_imagenet_224_a50 - transfer_akidanet_imagenet_224_a25 - fomo_akidanet_a50 - fomo_ad_gmm - fomo_ad_patchcore BlockParamsVisualAnomalyPatchcore: type: object properties: backbone: type: string description: The backbone to use for feature extraction numLayers: type: integer description: The number of layers in the feature extractor (1-3) poolSize: type: integer description: The pool size for the feature extractor samplingRatio: type: number description: The sampling ratio for the coreset, used for anomaly scoring numNearestNeighbors: type: integer description: >- The number of nearest neighbors to consider, used for anomaly scoring BlockParamsVisualAnomalyGmm: type: object properties: backbone: type: string description: The backbone to use for feature extraction securitySchemes: ApiKeyAuthentication: type: apiKey in: header name: x-api-key JWTAuthentication: type: apiKey in: cookie name: jwt JWTHttpHeaderAuthentication: type: apiKey in: header name: x-jwt-token OAuth2: type: oauth2 flows: authorizationCode: authorizationUrl: /v1/oauth/authorize tokenUrl: /v1/oauth/token scopes: openid: Access to basic profile information email: Access to email address profile: Access to full profile information implicit: authorizationUrl: /v1/oauth/authorize scopes: openid: Access to basic profile information email: Access to email address profile: Access to full profile information password: tokenUrl: /v1/oauth/token scopes: openid: Access to basic profile information email: Access to email address profile: Access to full profile information clientCredentials: tokenUrl: /v1/oauth/token scopes: openid: Access to basic profile information email: Access to email address profile: Access to full profile information ````