Skip to content

Commit e99224e

Browse files
you-n-gyou-n-g
authored
Update benchmark based on new backtest (#634)
* free random seed * update model baselines * more robust for parameters
1 parent 8c8d133 commit e99224e

File tree

12 files changed

+239
-209
lines changed

12 files changed

+239
-209
lines changed

‎README.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -306,7 +306,7 @@ All the models listed above are runnable with ``Qlib``. Users can find the confi
306306
- Users can use the tool `qrun` mentioned above to run a model's workflow based from a config file.
307307
- Users can create a `workflow_by_code` python script based on the [one](examples/workflow_by_code.py) listed in the `examples` folder.
308308

309-
- Users can use the script [`run_all_model.py`](examples/run_all_model.py) listed in the `examples` folder to run a model. Here is an example of the specific shell command to be used: `python run_all_model.py --models=lightgbm`, where the `--models` arguments can take any number of models listed above(the available models can be found in [benchmarks](examples/benchmarks/)). For more use cases, please refer to the file's [docstrings](examples/run_all_model.py).
309+
- Users can use the script [`run_all_model.py`](examples/run_all_model.py) listed in the `examples` folder to run a model. Here is an example of the specific shell command to be used: `python run_all_model.py run --models=lightgbm`, where the `--models` arguments can take any number of models listed above(the available models can be found in [benchmarks](examples/benchmarks/)). For more use cases, please refer to the file's [docstrings](examples/run_all_model.py).
310310
- **NOTE**: Each baseline has different environment dependencies, please make sure that your python version aligns with the requirements(e.g. TFT only supports Python 3.6~3.7 due to the limitation of `tensorflow==1.15.0`)
311311
312312
## Run multiple models
@@ -316,7 +316,7 @@ The script will create a unique virtual environment for each model, and delete t
316316

317317
Here is an example of running all the models for 10 iterations:
318318
```python
319-
python run_all_model.py 10
319+
python run_all_model.py run 10
320320
```
321321

322322
It also provides the API to run specific models at once. For more use cases, please refer to the file's [docstrings](examples/run_all_model.py).

‎examples/benchmarks/README.md‎

Lines changed: 40 additions & 36 deletions
Original file line numberDiff line numberDiff line change
@@ -8,44 +8,48 @@ The numbers shown below demonstrate the performance of the entire `workflow` of
88
>
99
> In the new version of qlib, the default dataset is **v2**. Since the data is collected from the YahooFinance API (which is not very stable), the results of *v2* and *v1* may differ
1010
11+
## Alpha158 dataset
12+
13+
| Model Name | Dataset | IC | ICIR | Rank IC | Rank ICIR | Annualized Return | Information Ratio | Max Drawdown |
14+
|------------------------------------------|-------------------------------------|-------------|-------------|-------------|-------------|-------------------|-------------------|--------------|
15+
| TabNet(Sercan O. Arik, et al.) | Alpha158 | 0.0204±0.01 | 0.1554±0.07 | 0.0333±0.00 | 0.2552±0.05 | 0.0227±0.04 | 0.3676±0.54 | -0.1089±0.08 |
16+
| Transformer(Ashish Vaswani, et al.) | Alpha158 | 0.0264±0.00 | 0.2053±0.02 | 0.0407±0.00 | 0.3273±0.02 | 0.0273±0.02 | 0.3970±0.26 | -0.1101±0.02 |
17+
| GRU(Kyunghyun Cho, et al.) | Alpha158(with selected 20 features) | 0.0315±0.00 | 0.2450±0.04 | 0.0428±0.00 | 0.3440±0.03 | 0.0344±0.02 | 0.5160±0.25 | -0.1017±0.02 |
18+
| LSTM(Sepp Hochreiter, et al.) | Alpha158(with selected 20 features) | 0.0318±0.00 | 0.2367±0.04 | 0.0435±0.00 | 0.3389±0.03 | 0.0381±0.03 | 0.5561±0.46 | -0.1207±0.04 |
19+
| Localformer(Juyong Jiang, et al.) | Alpha158 | 0.0356±0.00 | 0.2756±0.03 | 0.0468±0.00 | 0.3784±0.03 | 0.0438±0.02 | 0.6600±0.33 | -0.0952±0.02 |
20+
| SFM(Liheng Zhang, et al.) | Alpha158 | 0.0379±0.00 | 0.2959±0.04 | 0.0464±0.00 | 0.3825±0.04 | 0.0465±0.02 | 0.5672±0.29 | -0.1282±0.03 |
21+
| ALSTM (Yao Qin, et al.) | Alpha158(with selected 20 features) | 0.0362±0.01 | 0.2789±0.06 | 0.0463±0.01 | 0.3661±0.05 | 0.0470±0.03 | 0.6992±0.47 | -0.1072±0.03 |
22+
| GATs (Petar Velickovic, et al.) | Alpha158(with selected 20 features) | 0.0349±0.00 | 0.2511±0.01 | 0.0462±0.00 | 0.3564±0.01 | 0.0497±0.01 | 0.7338±0.19 | -0.0777±0.02 |
23+
| TRA(Hengxu Lin, et al.) | Alpha158(with selected 20 features) | 0.0404±0.00 | 0.3197±0.05 | 0.0490±0.00 | 0.4047±0.04 | 0.0649±0.02 | 1.0091±0.30 | -0.0860±0.02 |
24+
| Linear | Alpha158 | 0.0397±0.00 | 0.3000±0.00 | 0.0472±0.00 | 0.3531±0.00 | 0.0692±0.00 | 0.9209±0.00 | -0.1509±0.00 |
25+
| TRA(Hengxu Lin, et al.) | Alpha158 | 0.0440±0.00 | 0.3535±0.05 | 0.0540±0.00 | 0.4451±0.03 | 0.0718±0.02 | 1.0835±0.35 | -0.0760±0.02 |
26+
| CatBoost(Liudmila Prokhorenkova, et al.) | Alpha158 | 0.0481±0.00 | 0.3366±0.00 | 0.0454±0.00 | 0.3311±0.00 | 0.0765±0.00 | 0.8032±0.01 | -0.1092±0.00 |
27+
| XGBoost(Tianqi Chen, et al.) | Alpha158 | 0.0498±0.00 | 0.3779±0.00 | 0.0505±0.00 | 0.4131±0.00 | 0.0780±0.00 | 0.9070±0.00 | -0.1168±0.00 |
28+
| TFT (Bryan Lim, et al.) | Alpha158(with selected 20 features) | 0.0358±0.00 | 0.2160±0.03 | 0.0116±0.01 | 0.0720±0.03 | 0.0847±0.02 | 0.8131±0.19 | -0.1824±0.03 |
29+
| MLP | Alpha158 | 0.0376±0.00 | 0.2846±0.02 | 0.0429±0.00 | 0.3220±0.01 | 0.0895±0.02 | 1.1408±0.23 | -0.1103±0.02 |
30+
| LightGBM(Guolin Ke, et al.) | Alpha158 | 0.0448±0.00 | 0.3660±0.00 | 0.0469±0.00 | 0.3877±0.00 | 0.0901±0.00 | 1.0164±0.00 | -0.1038±0.00 |
31+
| DoubleEnsemble(Chuheng Zhang, et al.) | Alpha158 | 0.0544±0.00 | 0.4340±0.00 | 0.0523±0.00 | 0.4284±0.01 | 0.1168±0.01 | 1.3384±0.12 | -0.1036±0.01 |
32+
33+
34+
1135
## Alpha360 dataset
12-
| Model Name | Dataset | IC | ICIR | Rank IC | Rank ICIR | Annualized Return | Information Ratio | Max Drawdown |
13-
|---|---|---|---|---|---|---|---|---|
14-
| Linear | Alpha360 | 0.0150±0.00 | 0.1049±0.00| 0.0284±0.00 | 0.1970±0.00 | -0.0659±0.00 | -0.7072±0.00| -0.2955±0.00 |
15-
| CatBoost (Liudmila Prokhorenkova, et al.) | Alpha360 | 0.0397±0.00 | 0.2878±0.00| 0.0470±0.00 | 0.3703±0.00 | 0.0342±0.00 | 0.4092±0.00| -0.1057±0.00 |
16-
| XGBoost (Tianqi Chen, et al.) | Alpha360 | 0.0400±0.00 | 0.3031±0.00| 0.0461±0.00 | 0.3862±0.00 | 0.0528±0.00 | 0.6307±0.00| -0.1113±0.00 |
17-
| LightGBM (Guolin Ke, et al.) | Alpha360 | 0.0399±0.00 | 0.3075±0.00| 0.0492±0.00 | 0.4019±0.00 | 0.0323±0.00 | 0.4370±0.00| -0.0917±0.00 |
18-
| MLP | Alpha360 | 0.0285±0.00 | 0.1981±0.02| 0.0402±0.00 | 0.2993±0.02 | 0.0073±0.02 | 0.0880±0.22| -0.1446±0.03 |
19-
| GRU (Kyunghyun Cho, et al.) | Alpha360 | 0.0490±0.01 | 0.3787±0.05| 0.0581±0.00 | 0.4664±0.04 | 0.0726±0.02 | 0.9817±0.34| -0.0902±0.03 |
20-
| LSTM (Sepp Hochreiter, et al.) | Alpha360 | 0.0443±0.01 | 0.3401±0.05| 0.0536±0.01 | 0.4248±0.05 | 0.0627±0.03 | 0.8441±0.48| -0.0882±0.03 |
21-
| ALSTM (Yao Qin, et al.) | Alpha360 | 0.0493±0.01 | 0.3778±0.06| 0.0585±0.00 | 0.4606±0.04 | 0.0513±0.03 | 0.6727±0.38| -0.1085±0.02 |
22-
| GATs (Petar Velickovic, et al.) | Alpha360 | 0.0475±0.00 | 0.3515±0.02| 0.0592±0.00 | 0.4585±0.01 | 0.0876±0.02 | 1.1513±0.27| -0.0795±0.02 |
23-
| DoubleEnsemble (Chuheng Zhang, et al.) | Alpha360 | 0.0407±0.00| 0.3053±0.00 | 0.0490±0.00 | 0.3840±0.00 | 0.0380±0.02 | 0.5000±0.21 | -0.0984±0.02 |
24-
| TabNet (Sercan O. Arik, et al.)| Alpha360 | 0.0192±0.00 | 0.1401±0.00| 0.0291±0.00 | 0.2163±0.00 | -0.0258±0.00 | -0.2961±0.00| -0.1429±0.00 |
25-
| TCTS (Xueqing Wu, et al.)| Alpha360 | 0.0485±0.00 | 0.3689±0.04| 0.0586±0.00 | 0.4669±0.02 | 0.0816±0.02 | 1.1572±0.30| -0.0689±0.02 |
26-
| Transformer (Ashish Vaswani, et al.)| Alpha360 | 0.0141±0.00 | 0.0917±0.02| 0.0331±0.00 | 0.2357±0.03 | -0.0259±0.03 | -0.3323±0.43| -0.1763±0.07 |
27-
| Localformer (Juyong Jiang, et al.)| Alpha360 | 0.0408±0.00 | 0.2988±0.03| 0.0538±0.00 | 0.4105±0.02 | 0.0275±0.03 | 0.3464±0.37| -0.1182±0.03 |
28-
| TRA (Hengxu Lin, et al.)| Alpha360 | 0.0491±0.01 | 0.3868±0.06 | 0.0589±0.00 | 0.4802±0.04 | 0.0898±0.02 | 1.2490±0.32 | -0.0778±0.02 |
2936

30-
## Alpha158 dataset
31-
| Model Name | Dataset | IC | ICIR | Rank IC | Rank ICIR | Annualized Return | Information Ratio | Max Drawdown |
32-
|---|---|---|---|---|---|---|---|---|
33-
| Linear | Alpha158 | 0.0393±0.00 | 0.2980±0.00| 0.0475±0.00 | 0.3546±0.00 | 0.0795±0.00 | 1.0712±0.00| -0.1449±0.00 |
34-
| CatBoost (Liudmila Prokhorenkova, et al.) | Alpha158 | 0.0503±0.00 | 0.3586±0.00| 0.0483±0.00 | 0.3667±0.00 | 0.1080±0.00 | 1.1561±0.00| -0.0787±0.00 |
35-
| XGBoost (Tianqi Chen, et al.) | Alpha158 | 0.0481±0.00 | 0.3659±0.00| 0.0495±0.00 | 0.4033±0.00 | 0.1111±0.00 | 1.2915±0.00| -0.0893±0.00 |
36-
| LightGBM (Guolin Ke, et al.) | Alpha158 | 0.0475±0.00 | 0.3979±0.00| 0.0485±0.00 | 0.4123±0.00 | 0.1143±0.00 | 1.2744±0.00| -0.0800±0.00 |
37-
| MLP | Alpha158 | 0.0358±0.00 | 0.2738±0.03| 0.0425±0.00 | 0.3221±0.01 | 0.0836±0.02 | 1.0323±0.25| -0.1127±0.02 |
38-
| TFT (Bryan Lim, et al.) | Alpha158 (with selected 20 features) | 0.0343±0.00 | 0.2071±0.02| 0.0107±0.00 | 0.0660±0.02 | 0.0623±0.02 | 0.5818±0.20| -0.1762±0.01 |
39-
| GRU (Kyunghyun Cho, et al.) | Alpha158 (with selected 20 features) | 0.0311±0.00 | 0.2418±0.04| 0.0425±0.00 | 0.3434±0.02 | 0.0330±0.02 | 0.4805±0.30| -0.1021±0.02 |
40-
| LSTM (Sepp Hochreiter, et al.) | Alpha158 (with selected 20 features) | 0.0312±0.00 | 0.2394±0.04| 0.0418±0.00 | 0.3324±0.03 | 0.0298±0.02 | 0.4198±0.33| -0.1348±0.03 |
41-
| ALSTM (Yao Qin, et al.) | Alpha158 (with selected 20 features) | 0.0385±0.01 | 0.3022±0.06| 0.0478±0.00 | 0.3874±0.04 | 0.0486±0.03 | 0.7141±0.45| -0.1088±0.03 |
42-
| GATs (Petar Velickovic, et al.) | Alpha158 (with selected 20 features) | 0.0349±0.00 | 0.2511±0.01| 0.0457±0.00 | 0.3537±0.01 | 0.0578±0.02 | 0.8221±0.25| -0.0824±0.02 |
43-
| DoubleEnsemble (Chuheng Zhang, et al.) | Alpha158 | 0.0544±0.00 | 0.4338±0.01 | 0.0523±0.00 | 0.4257±0.01 | 0.1253±0.01 | 1.4105±0.14 | -0.0902±0.01 |
44-
| TabNet (Sercan O. Arik, et al.)| Alpha158 | 0.0383±0.00 | 0.3414±0.00| 0.0388±0.00 | 0.3460±0.00 | 0.0226±0.00 | 0.2652±0.00| -0.1072±0.00 |
45-
| Transformer (Ashish Vaswani, et al.)| Alpha158 | 0.0274±0.00 | 0.2166±0.04| 0.0409±0.00 | 0.3342±0.04 | 0.0204±0.03 | 0.2888±0.40| -0.1216±0.04 |
46-
| Localformer (Juyong Jiang, et al.)| Alpha158 | 0.0355±0.00 | 0.2747±0.04| 0.0466±0.00 | 0.3762±0.03 | 0.0506±0.02 | 0.7447±0.34| -0.0875±0.02 |
47-
| TRA (Hengxu Lin, et al.)| Alpha158 (with selected 20 features)| 0.0409±0.00 | 0.3253±0.04 | 0.0488±0.00 | 0.4045±0.02 | 0.0673±0.02 | 1.0389±0.39 | -0.0830±0.02 |
48-
| TRA (Hengxu Lin, et al.)| Alpha158 | 0.0442±0.00 | 0.3426±0.03 | 0.0555±0.00 | 0.4395±0.03 | 0.0833±0.03 | 1.2064±0.36 | -0.0849±0.02 |
37+
| Model Name | Dataset | IC | ICIR | Rank IC | Rank ICIR | Annualized Return | Information Ratio | Max Drawdown |
38+
|-------------------------------------------|----------|-------------|-------------|-------------|-------------|-------------------|-------------------|--------------|
39+
| Transformer(Ashish Vaswani, et al.) | Alpha360 | 0.0114±0.00 | 0.0716±0.03 | 0.0327±0.00 | 0.2248±0.02 | -0.0270±0.03 | -0.3378±0.37 | -0.1653±0.05 |
40+
| TabNet(Sercan O. Arik, et al.) | Alpha360 | 0.0099±0.00 | 0.0593±0.00 | 0.0290±0.00 | 0.1887±0.00 | -0.0369±0.00 | -0.3892±0.00 | -0.2145±0.00 |
41+
| MLP | Alpha360 | 0.0273±0.00 | 0.1870±0.02 | 0.0396±0.00 | 0.2910±0.02 | 0.0029±0.02 | 0.0274±0.23 | -0.1385±0.03 |
42+
| Localformer(Juyong Jiang, et al.) | Alpha360 | 0.0404±0.00 | 0.2932±0.04 | 0.0542±0.00 | 0.4110±0.03 | 0.0246±0.02 | 0.3211±0.21 | -0.1095±0.02 |
43+
| CatBoost((Liudmila Prokhorenkova, et al.) | Alpha360 | 0.0378±0.00 | 0.2714±0.00 | 0.0467±0.00 | 0.3659±0.00 | 0.0292±0.00 | 0.3781±0.00 | -0.0862±0.00 |
44+
| XGBoost(Tianqi Chen, et al.) | Alpha360 | 0.0394±0.00 | 0.2909±0.00 | 0.0448±0.00 | 0.3679±0.00 | 0.0344±0.00 | 0.4527±0.02 | -0.1004±0.00 |
45+
| DoubleEnsemble(Chuheng Zhang, et al.) | Alpha360 | 0.0404±0.00 | 0.3023±0.00 | 0.0495±0.00 | 0.3898±0.00 | 0.0468±0.01 | 0.6302±0.20 | -0.0860±0.01 |
46+
| LightGBM(Guolin Ke, et al.) | Alpha360 | 0.0400±0.00 | 0.3037±0.00 | 0.0499±0.00 | 0.4042±0.00 | 0.0558±0.00 | 0.7632±0.00 | -0.0659±0.00 |
47+
| ALSTM (Yao Qin, et al.) | Alpha360 | 0.0497±0.00 | 0.3829±0.04 | 0.0599±0.00 | 0.4736±0.03 | 0.0626±0.02 | 0.8651±0.31 | -0.0994±0.03 |
48+
| LSTM(Sepp Hochreiter, et al.) | Alpha360 | 0.0448±0.00 | 0.3474±0.04 | 0.0549±0.00 | 0.4366±0.03 | 0.0647±0.03 | 0.8963±0.39 | -0.0875±0.02 |
49+
| GRU(Kyunghyun Cho, et al.) | Alpha360 | 0.0493±0.00 | 0.3772±0.04 | 0.0584±0.00 | 0.4638±0.03 | 0.0720±0.02 | 0.9730±0.33 | -0.0821±0.02 |
50+
| TCTS(Xueqing Wu, et al.) | Alpha360 | 0.0454±0.01 | 0.3457±0.06 | 0.0566±0.01 | 0.4492±0.05 | 0.0744±0.03 | 1.0594±0.41 | -0.0761±0.03 |
51+
| GATs (Petar Velickovic, et al.) | Alpha360 | 0.0476±0.00 | 0.3508±0.02 | 0.0598±0.00 | 0.4604±0.01 | 0.0824±0.02 | 1.1079±0.26 | -0.0894±0.03 |
52+
| TRA(Hengxu Lin, et al.) | Alpha360 | 0.0485±0.00 | 0.3787±0.03 | 0.0587±0.00 | 0.4756±0.03 | 0.0920±0.03 | 1.2789±0.42 | -0.0834±0.02 |
4953

5054
- The selected 20 features are based on the feature importance of a lightgbm-based model.
5155
- The base model of DoubleEnsemble is LGBM.

‎examples/benchmarks/TCTS/workflow_config_tcts_Alpha360.yaml‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -69,6 +69,7 @@ task:
6969
steps: 3
7070
target_label: 1
7171
lowest_valid_performance: 0.993
72+
seed: 0
7273
dataset:
7374
class: DatasetH
7475
module_path: qlib.data.dataset

‎examples/benchmarks/TFT/data_formatters/qlib_Alpha158.py‎

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -195,7 +195,8 @@ def format_predictions(self, predictions):
195195

196196
for col in column_names:
197197
if col not in {"forecast_time", "identifier"}:
198-
output[col] = self._target_scaler.inverse_transform(predictions[col])
198+
# Using [col] is for aligning with the format when fitting
199+
output[col] = self._target_scaler.inverse_transform(predictions[[col]])
199200

200201
return output
201202

‎examples/benchmarks/TFT/tft.py‎

Lines changed: 7 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -311,5 +311,11 @@ def to_pickle(self, path: Union[Path, str]):
311311
# self.model.save(path)
312312

313313
# save qlib model wrapper
314-
self.model = None
314+
drop_attrs = ["model", "tf_graph", "sess", "data_formatter"]
315+
orig_attr = {}
316+
for attr in drop_attrs:
317+
orig_attr[attr] = getattr(self, attr)
318+
setattr(self, attr, None)
315319
super(TFTModel, self).to_pickle(path)
320+
for attr in drop_attrs:
321+
setattr(self, attr, orig_attr[attr])

‎examples/benchmarks/TRA/src/model.py‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -38,7 +38,7 @@ def __init__(
3838
model_init_state=None,
3939
lamb=0.0,
4040
rho=0.99,
41-
seed=0,
41+
seed=None,
4242
logdir=None,
4343
eval_train=True,
4444
eval_test=False,

‎examples/benchmarks/TabNet/workflow_config_TabNet_Alpha158.yaml‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -50,6 +50,7 @@ task:
5050
kwargs:
5151
d_feat: 158
5252
pretrain: True
53+
seed: 993
5354
dataset:
5455
class: DatasetH
5556
module_path: qlib.data.dataset

‎examples/benchmarks/TabNet/workflow_config_TabNet_Alpha360.yaml‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -50,6 +50,7 @@ task:
5050
kwargs:
5151
d_feat: 360
5252
pretrain: True
53+
seed: 993
5354
dataset:
5455
class: DatasetH
5556
module_path: qlib.data.dataset

0 commit comments

Comments
 (0)