Factor concepts¶
Normal data¶
NormalData is the data containing pandas dataframe
with multiple index which level 0 named entity_id and level 1 named timestamp:
entity_id |
timestamp |
col1 |
col2 |
col3 |
col4 |
|---|---|---|---|---|---|
stock_sz_000338 |
2020-05-05 |
1.2 |
0.5 |
0.3 |
a |
… |
2020-05-06 |
1.0 |
0.7 |
0.2 |
b |
stock_sz_000778 |
2020-05-05 |
1.2 |
0.5 |
0.3 |
a |
… |
2020-05-06 |
1.0 |
0.7 |
0.2 |
b |
This data structure is used heavily in zvt computing, you should be familiar with it. Pandas multiple index guide is a good start.
Query data returning normal data in this way:
{Schema}.query_data(index=["entity_id, timestamp"])
e.g.
>>> from zvt.domain import *
>>> entity_ids = ["stock_sz_000338", "stock_sz_000001"]
>>> Stock1dHfqKdata.record_data(entity_ids=entity_ids, provider="em")
>>> df = Stock1dHfqKdata.query_data(entity_ids=entity_ids, provider="em", index=["entity_id", "timestamp"])
>>> print(df)
id entity_id timestamp provider code name level open close high low volume turnover change_pct turnover_rate
entity_id timestamp
stock_sz_000001 1991-04-03 stock_sz_000001_1991-04-03 stock_sz_000001 1991-04-03 em 000001 平安银行 1d 49.00 49.00 49.00 49.00 1.0 5.000000e+03 0.2250 0.0000
1991-04-04 stock_sz_000001_1991-04-04 stock_sz_000001 1991-04-04 em 000001 平安银行 1d 48.76 48.76 48.76 48.76 3.0 1.500000e+04 -0.0049 0.0000
1991-04-05 stock_sz_000001_1991-04-05 stock_sz_000001 1991-04-05 em 000001 平安银行 1d 48.52 48.52 48.52 48.52 2.0 1.000000e+04 -0.0049 0.0000
1991-04-06 stock_sz_000001_1991-04-06 stock_sz_000001 1991-04-06 em 000001 平安银行 1d 48.28 48.28 48.28 48.28 7.0 3.400000e+04 -0.0049 0.0000
1991-04-08 stock_sz_000001_1991-04-08 stock_sz_000001 1991-04-08 em 000001 平安银行 1d 48.04 48.04 48.04 48.04 2.0 1.000000e+04 -0.0050 0.0000
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
stock_sz_000338 2022-01-17 stock_sz_000338_2022-01-17 stock_sz_000338 2022-01-17 em 000338 潍柴动力 1d 296.26 297.64 298.71 293.49 504866.0 8.546921e+08 0.0026 0.0100
2022-01-18 stock_sz_000338_2022-01-18 stock_sz_000338 2022-01-18 em 000338 潍柴动力 1d 298.10 300.87 302.71 296.10 622455.0 1.064735e+09 0.0109 0.0124
2022-01-19 stock_sz_000338_2022-01-19 stock_sz_000338 2022-01-19 em 000338 潍柴动力 1d 299.64 299.48 304.24 298.56 610096.0 1.049195e+09 -0.0046 0.0121
2022-01-20 stock_sz_000338_2022-01-20 stock_sz_000338 2022-01-20 em 000338 潍柴动力 1d 298.10 294.87 299.18 290.11 812949.0 1.361764e+09 -0.0154 0.0161
2022-01-21 stock_sz_000338_2022-01-21 stock_sz_000338 2022-01-21 em 000338 潍柴动力 1d 292.72 287.04 293.34 284.58 754156.0 1.234360e+09 -0.0266 0.0150
[10878 rows x 15 columns]
Factor¶
Factor is a computing facility to build factor according your mind ——— algorithm.
It reads data from schema, use Transformer, Accumulator
or your custom logic to compute, and save the result to new schema if need.
It also provides a standard way to evaluate the targets which could be used by TargetSelector
and Trader for backtesting or real trading.
Transformer¶
Computing factor which depends on input data only.
Here is an example: MaTransformer
Accumulator¶
Computing factor which depends on input data and previous result of the factor.
Here is an example:
Let’s have a look by example:
>>> from zvt.factors import GoldCrossFactor
>>> from zvt.domain import Stock1dHfqKdata
>>> entity_ids = ["stock_sz_000338"]
>>> Stock1dHfqKdata.record_data(entity_ids=entity_ids, provider="em")
>>> factor = GoldCrossFactor(entity_ids=entity_ids, provider="em", start_timestamp="2018-01-01")
>>> print(factor.factor_df)
>>> print(factor.result_df)
>>> factor.draw(show=True)
level turnover high id open low entity_id timestamp close turnover_rate volume diff dea macd live bull live_count
entity_id timestamp
stock_sz_000338 2018-01-02 1d 8.325588e+08 145.97 stock_sz_000338_2018-01-02 141.21 141.06 stock_sz_000338 2018-01-02 145.67 0.0225 972471.0 NaN NaN NaN -1 False -1
2018-01-03 1d 7.530370e+08 147.66 stock_sz_000338_2018-01-03 146.13 144.29 stock_sz_000338 2018-01-03 144.44 0.0202 870225.0 NaN NaN NaN -1 False -2
2018-01-04 1d 4.917067e+08 145.51 stock_sz_000338_2018-01-04 144.75 143.67 stock_sz_000338 2018-01-04 145.21 0.0133 574335.0 NaN NaN NaN -1 False -3
2018-01-05 1d 5.282211e+08 146.59 stock_sz_000338_2018-01-05 146.44 143.21 stock_sz_000338 2018-01-05 143.21 0.0143 616244.0 NaN NaN NaN -1 False -4
2018-01-08 1d 1.255871e+09 150.43 stock_sz_000338_2018-01-08 143.82 143.82 stock_sz_000338 2018-01-08 150.12 0.0331 1426567.0 NaN NaN NaN -1 False -5
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
2022-01-17 1d 8.546921e+08 298.71 stock_sz_000338_2022-01-17 296.26 293.49 stock_sz_000338 2022-01-17 297.64 0.0100 504866.0 -1.386687 1.781615 -6.336603 -1 False -12
2022-01-18 1d 1.064735e+09 302.71 stock_sz_000338_2022-01-18 298.10 296.10 stock_sz_000338 2022-01-18 300.87 0.0124 622455.0 -1.694421 1.086407 -5.561657 -1 False -13
2022-01-19 1d 1.049195e+09 304.24 stock_sz_000338_2022-01-19 299.64 298.56 stock_sz_000338 2022-01-19 299.48 0.0121 610096.0 -2.027097 0.463707 -4.981607 -1 False -14
2022-01-20 1d 1.361764e+09 299.18 stock_sz_000338_2022-01-20 298.10 290.11 stock_sz_000338 2022-01-20 294.87 0.0161 812949.0 -2.632389 -0.155513 -4.953753 -1 False -15
2022-01-21 1d 1.234360e+09 293.34 stock_sz_000338_2022-01-21 292.72 284.58 stock_sz_000338 2022-01-21 287.04 0.0150 754156.0 -3.701237 -0.864657 -5.673159 -1 False -16
[987 rows x 17 columns]
filter_result
entity_id timestamp
stock_sz_000338 2018-01-02 False
2018-01-03 False
2018-01-04 False
2018-01-05 False
2018-01-08 False
... ...
2022-01-17 False
2022-01-18 False
2022-01-19 False
2022-01-20 False
2022-01-21 False
[987 rows x 1 columns]
Follow Extending factor to do the funny part.
TargetSelector¶
The class select targets according to Factors. You could calculate factors in the whole market and use selector to choose the targets.
from zvt.contract import IntervalLevel
from zvt.factors.target_selector import TargetSelector
from zvt.factors.ma.ma_factor import CrossMaFactor
entity_ids = ["stock_sz_000338"]
entity_type = "stock"
start_timestamp = "2018-01-01"
end_timestamp = "2019-06-30"
my_selector = TargetSelector(
entity_ids=entity_ids, entity_schema=entity_type, start_timestamp=start_timestamp, end_timestamp=end_timestamp
)
# add the factors
my_selector.add_factor(
CrossMaFactor(
entity_provider="em",
provider="em",
entity_ids=entity_ids,
start_timestamp=start_timestamp,
end_timestamp=end_timestamp,
computing_window=10,
windows=[5, 10],
need_persist=False,
level=IntervalLevel.LEVEL_1DAY,
adjust_type="hfq",
)
)
my_selector.run()
print(my_selector.open_long_df)
print(my_selector.open_short_df)
my_selector.get_open_long_targets("2019-06-27")
If not set entity_ids arguments, the selected targets would be in whole market. And it provides get_open_long_targets function to select targets on the timestamp. For multiple targets backtesting, this pre computed factor would be very fast.