Overview run_dl_job is Stage 3b of the pipeline (alternative to run_ml_job). It triggers the dl-job pipeline job, which trains a PyTorch neural network on the feature-engineered dataset and generates per-ticker price predictions.The tool blocks until the job completes and returns the output blob URL.
Parameters Blob storage URL pointing to a feature_engine_*.json file. This is the output_url returned by run_feature_worker.
Blob storage URL pointing to a data_extractor_*.json file. This is the output_url returned by run_data_extraction. Required alongside feature_url
for proper train/test date alignment.
Neural network configuration. Top-level container for all neural network settings. Training and architecture configuration. Show Dot prediction fields
Enable or disable this prediction block. Default: true.
Model architecture to use. Supported values: Value Architecture "lstm"Multi-layer LSTM (per-ticker) "cnn"1D Convolutional Network (per-ticker) "transformer"Transformer Encoder (per-ticker) "lstm_cnn"Hybrid LSTM + CNN (per-ticker) "mlp"Multi-layer Perceptron (per-ticker) "portfolio_lstm"LSTM trained across all tickers jointly "portfolio_transformer"Transformer trained across all tickers jointly
Runtime settings. Show global_settings fields
Compute device. "auto" selects CUDA if available, else CPU. Default: "auto".
Random seed for reproducibility. Default: 42.
Data preparation settings. Show data_processing fields
Feature scaler. Options: "standard", "minmax", "robust", "maxabs". Default: "robust".
Sequence window length (number of past days fed to the model). Default: 10.
Mini-batch size for training. Default: 64.
Optimizer and training loop settings. Show training_params fields
Maximum number of training epochs. Default: 20.
Learning rate. Default: 0.001.
Early stopping patience (epochs without validation improvement). Default: 5.
Optimizer. Options: "adam", "sgd", "rmsprop". Default: "adam".
L2 regularization coefficient. Default: 0.0.
Loss function. Options: "mse", "mae", "huber", "directional", "sharpe", "sortino", "mahalanobis".
Default: "mse" for per-ticker models; "mahalanobis" for portfolio models.
Architecture-specific hyperparameters. Only the block matching current option is used. Show Per-architecture fields
lstm / portfolio_lstm
hidden_dims (int array) — hidden layer sizes, e.g. [64, 32]dropout (float) — dropout rate, e.g. 0.2 cnn
filters (int array) — number of filters per conv layer, e.g. [32, 64]kernel_size (int) — convolution kernel size, default 3dropout (float) — dropout ratepool_size (int) — max-pool size, default 2 transformer / portfolio_transformer
d_model (int) — model embedding dimension, e.g. 64nhead (int) — number of attention heads, e.g. 4num_layers (int) — number of encoder layers, e.g. 2dim_feedforward (int) — feedforward dimension, e.g. 128dropout (float) — dropout rate lstm_cnn
conv_filters (int) — number of conv filters, e.g. 64lstm_units (int) — LSTM hidden size, e.g. 64dropout (float) — dropout rate mlp
hidden_dims (int array) — hidden layer sizes, e.g. [128, 64]dropout (float) — dropout rate Optional date range hints for train/test split alignment. Show Backtest params fields
End of the training window. Format: YYYY-MM-DD. Pass null to infer from data.
End of the testing window. Format: YYYY-MM-DD. Pass null to infer from data.
Returns { "status" : "Succeeded" , "output_url" : "https://stmcpfabricdev.blob.core.windows.net/data/nn_engine_20260307_125841.json" , "output_name" : "nn_engine_20260307_125841.json" , "execution_name" : "dl-job-abc123xyz" }
Field Description statusJob terminal status (Succeeded) output_urlFull HTTPS URL to the output blob — pass to run_po_job output_nameBlob filename execution_nameJob execution ID for audit/debugging
Example — LSTM (per-ticker) { "feature_url" : "https://stmcpfabricdev.blob.core.windows.net/data/feature_engine_20260307_125103.json" , "data_extractor_url" : "https://stmcpfabricdev.blob.core.windows.net/data/data_extractor_20260307_124937.json" , "config" : { "NN params" : { "Dot prediction" : { "status" : true , "current option" : "lstm" , "global_settings" : { "device" : "auto" , "seed" : 42 }, "data_processing" : { "scaler" : "robust" , "lookback" : 10 , "batch_size" : 64 }, "training_params" : { "epochs" : 20 , "learning_rate" : 0.001 , "patience" : 5 , "optimizer" : "adam" , "weight_decay" : 0.0 , "loss_type" : "mse" }, "models_params" : { "lstm" : { "hidden_dims" : [ 64 , 32 ], "dropout" : 0.2 } } } }, "Backtest params" : { "Learning_end" : null , "Testing_end" : null } } }
{ "feature_url" : "..." , "data_extractor_url" : "..." , "config" : { "NN params" : { "Dot prediction" : { "status" : true , "current option" : "portfolio_transformer" , "global_settings" : { "device" : "auto" , "seed" : 42 }, "data_processing" : { "scaler" : "minmax" , "lookback" : 12 , "batch_size" : 64 }, "training_params" : { "epochs" : 5 , "learning_rate" : 0.0005 , "patience" : 10 , "optimizer" : "sgd" , "weight_decay" : 0.0001 , "loss_type" : "sharpe" }, "models_params" : { "portfolio_transformer" : { "d_model" : 64 , "nhead" : 4 , "num_layers" : 2 , "dropout" : 0.4 } } } } } }
Resources Resource Value Container Apps Job dl-jobContainer name dl-jobEnv vars injected FEATURE_URL, DATA_EXTRACTOR_URL, CONFIGOutput blob prefix nn_engine_Timeout 600 seconds
Next Step Pass output_url to run_po_jobinput_url.
If you prefer classical machine learning, use run_ml_jobml_engine_*.json blob equally compatible with run_po_job. 
