alpha_lab/stock_1d/d033/alpha158_beta/config.yaml

# First, let me create a script to train a VAE model on the 0_7_beta data
# This would need to be done separately as it's a prerequisite for the prediction script above

"""
Workflow configuration to train a VAE model on alpha158 0_7_beta data.
This creates a VAE-encoded version of the 0_7_beta factors that can be used
for prediction comparison with the original 0_7 model.
"""

experiment_name: vae_alpha158_0_7_beta

qlib_init:
    provider_uri: "~/.qlib/data_ops/target"
    region: cn

load_start: &load_start 2013-01-01
load_end: &load_end 2023-09-30

train_start: &train_start 2013-01-01
train_end: &train_end 2018-12-31

benchmark_name: &benchmark_name SH000985
market: &market csiallx

dataset_cache_path: &dataset_cache_path tasks/artifacts/csiallx_dataset_alpha158_0_7_beta_vae.pkl

# DolphinDB configuration
ddb_config: &ddb_config
    host: 192.168.1.146
    port: 8848
    username: "admin"
    password: "123456"

data_handler_config: &data_handler_config
    start_time: *load_start
    end_time: *load_end
    fit_start_time: *train_start
    fit_end_time: *train_end
    instruments: *market
    ddb_config: *ddb_config
    handler_list:
        # Alpha158 0_7_beta features
        - class: DDBZWindDataHandler
          module_path: qlib.contrib.data.ddb_handlers.ddb_wind_handler
          kwargs:
            col_set: "feature"
            query_config:
              - db_path: "dfs://daily_stock_run"
                dtype: "float32"
                field_list: "alpha158"  # All alpha158 factors
                table_name: "stg_1day_wind_alpha158_0_7_beta"  # Use the beta version
        # Additional handlers as needed
        - class: DDBZWindDataHandler
          module_path: qlib.contrib.data.ddb_handlers.ddb_wind_handler
          kwargs:
            col_set: "risk_factor"
            query_config:
              - db_path: "dfs://daily_stock_run"
                dtype: "float32"
                field_list: ["MarketValue as total_size"]
                table_name: "stg_1day_wind_kline_adjusted"
        - class: DDBZWindDataHandler
          module_path: qlib.contrib.data.ddb_handlers.ddb_indus_flag_handler
          kwargs:
            col_set: "indus_flag"
            query_config:
              - db_path: "dfs://daily_stock_run"
                dtype: "bool"
                field_list: "industry_code_cc.csv"
                table_name: "stg_1day_gds_indus_flag_cc1"
        - class: DDBZWindDataHandler
          module_path: qlib.contrib.data.ddb_handlers.ddb_st_flag_handler
          kwargs:
            col_set: "st_flag"
            query_config:
              - db_path: "dfs://daily_stock_run"
                dtype: "bool"
                field_list: ["ST_Y", "ST_S", "ST_T", "ST_L", "ST_Z", "ST_X"]
                table_name: "stg_1day_wind_st_flag"
    infer_processors:
        - class: FlagToOnehot
          module_path: qlib.contrib.data.processor_flag
          kwargs:
              fields_group: indus_flag
              onehot_group: indus_idx
        - class: FactorNtrlInjector
          module_path: qlib.contrib.data.processor_ntrl
          kwargs:
              fields_group: "feature"
              factor_col: "risk_factor"
              dummy_col: "indus_idx"
              ntrl_type: "size_indus"
        - class: RobustZScoreNorm
          kwargs:
              fields_group: ["feature"]
              clip_outlier: true
        - class: Fillna
          kwargs:
              fields_group: ["feature"]

task:
    model:
        class: VAEModel
        module_path: qlib.contrib.model.task.task_vae_flat
        kwargs:
            model_config:
                hidden_size: 32  # Same as the original model for consistency
                nn_module:
                    class: VAE
                    module_path: qlib.contrib.model.module.module_vae
                    kwargs:
                        variational: true
            optim_config:
                seed: 1234567
                bootstrap_config: 1.2
                distort_config: 1e-3
                beta: 1e-3  # KL divergence weight
                n_epochs: 300
                early_stop: 10
                lr: 1e-3
                optimizer: adamw
                batch_size: 10000
                n_jobs: 4
                checkpoint:
                    save_path: tasks/artifacts/checkpoints/csiallx_alpha158_0_7_beta_vae32
    dataset:
        class: DatasetH
        module_path: qlib.data.dataset
        kwargs:
            config_module: qlib.contrib.data.config
            from_cache: *dataset_cache_path
            require_setup: true
            handler:
                class: AggHandler
                module_path: qlib.contrib.data.agg_handler
                kwargs: *data_handler_config
            segments:
                train: [*train_start, *train_end]
                test: [*load_start, *load_end]
    record:
        - class: SignalRecord
          module_path: qlib.contrib.workflow.record_temp
          kwargs:
            model: <MODEL>
            dataset: <DATASET>
            col_set: "feature"