Qlib Sandwich Model Workflow for CSI 300 Stock Ranking
Summary
This Qlib configuration defines a supervised stock-ranking experiment using the CSI 300 universe and the Sandwich neural model. Its data handler applies robust feature normalization with outlier clipping, fills missing feature values, drops missing labels, and cross-sectionally rank-normalizes labels. The target is a forward close-to-close return over the next two reference points. The dataset is divided into training, validation, and test periods, with the model trained on the earliest segment and evaluated later.
For portfolio analysis, predicted scores feed a top-k dropout strategy that holds 50 names and replaces up to 5 positions at a time. The configuration specifies a Shanghai Composite benchmark, a 2017–2020 backtest window, close-price execution, transaction costs, and a price-limit threshold. It also configures signal and portfolio analysis records. This is an experiment specification, not a report of results: it gives no returns, risk statistics, or comparison with alternatives. The setup depends on the configured Chinese market data and Qlib components, and the stated backtest assumptions do not establish live tradability or robustness.
Key ideas
- The workflow uses Qlib’s Alpha360 handler and the CSI 300 universe to construct a stock prediction task.
- Features receive robust normalization and missing-value filling, while labels are rank-normalized across stocks.
- The target measures a forward close-to-close return, and the data is split into chronological train, validation, and test segments.
- Predictions are evaluated with a top-k dropout portfolio strategy using a benchmark, close-price execution, and specified costs.
- The configuration describes an experiment but reports no performance results or evidence of live robustness.
Tags
Full text
# workflow_config_sandwich_Alpha360.yaml
```yaml
qlib_init:
provider_uri: "~/.qlib/qlib_data/cn_data"
region: cn
market: &market csi300
benchmark: &benchmark SH000300
data_handler_config: &data_handler_config
start_time: 2008-01-01
end_time: 2020-08-01
fit_start_time: 2008-01-01
fit_end_time: 2014-12-31
instruments: *market
infer_processors:
- class: RobustZScoreNorm
kwargs:
fields_group: feature
clip_outlier: true
- class: Fillna
kwargs:
fields_group: feature
learn_processors:
- class: DropnaLabel
- class: CSRankNorm
kwargs:
fields_group: label
label: ["Ref($close, -2) / Ref($close, -1) - 1"]
port_analysis_config: &port_analysis_config
strategy:
class: TopkDropoutStrategy
module_path: qlib.contrib.strategy
kwargs:
signal: <PRED>
topk: 50
n_drop: 5
backtest:
start_time: 2017-01-01
end_time: 2020-08-01
account: 100000000
benchmark: *benchmark
exchange_kwargs:
limit_threshold: 0.095
deal_price: close
open_cost: 0.0005
close_cost: 0.0015
min_cost: 5
task:
model:
class: Sandwich
module_path: qlib.contrib.model.pytorch_sandwich
kwargs:
fea_dim: 6
cnn_dim_1: 16
cnn_dim_2: 16
cnn_kernel_size: 3
rnn_dim_1: 8
rnn_dim_2: 8
rnn_dups: 2
rnn_layers: 2
n_epochs: 200
lr: 0.001
early_stop: 20
batch_size: 2000
metric: loss
GPU: 0
dataset:
class: DatasetH
module_path: qlib.data.dataset
kwargs:
handler:
class: Alpha360
module_path: qlib.contrib.data.handler
kwargs: *data_handler_config
segments:
train: [2008-01-01, 2014-12-31]
valid: [2015-01-01, 2016-12-31]
test: [2017-01-01, 2020-08-01]
record:
- class: SignalRecord
module_path: qlib.workflow.record_temp
kwargs:
model: <MODEL>
dataset: <DATASET>
- class: SigAnaRecord
module_path: qlib.workflow.record_temp
kwargs:
ana_long_short: False
ann_scaler: 252
- class: PortAnaRecord
module_path: qlib.workflow.record_temp
kwargs:
config: *port_analysis_config
```Shown in full with attribution under the source's licence. Licence: MIT
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.