Indexing¶

Most VBT objects, such as data instances and portfolios, can be indexed like regular Pandas objects using the [], iloc, loc, and xs selectors. The operation is passed down to all arrays inside the instance, and a new instance with the new arrays is created.

new_data = data.loc["2020-01-01":"2020-12-31"]

In addition, there's a special selector xloc that accepts a smart indexing instruction. Such an instruction can contain one or more positions, labels, dates, times, ranges, frequencies, or even date offsets. It's parsed automatically and translated into an array with integer positions that are internally passed to the iloc selector.

new_data = data.xloc[::2]  # (1)!
new_data = data.xloc[np.array([10, 20, 30])]  # (2)!
new_data = data.xloc["2020-01-01 17:30"]  # (3)!
new_data = data.xloc["2020-01-01"]  # (4)!
new_data = data.xloc["2020-01"]  # (5)!
new_data = data.xloc["2020"]  # (6)!
new_data = data.xloc["2020-01-01":"2021-01-01"]  # (7)!
new_data = data.xloc["january":"april"]  # (8)!
new_data = data.xloc["monday":"saturday"]  # (9)!
new_data = data.xloc["09:00":"16:00"]  # (10)!
new_data = data.xloc["16:00":"09:00"]  # (11)!
new_data = data.xloc["monday 09:00":"friday 16:00"]  # (12)!
new_data = data.xloc[
    vbt.autoidx(slice("monday", "friday"), closed_end=True) &  # (13)!
    vbt.autoidx(slice("09:00", "16:00"), closed_end=False)
]
new_data = data.xloc["Y"]  # (14)!
new_data = data.xloc[pd.Timedelta(days=7)]  # (15)!
new_data = data.xloc[df.index.weekday == 0]  # (16)!
new_data = data.xloc[pd.tseries.offsets.BDay()]  # (17)!

1. Every second row
2. Rows 10, 20, and 30
3. All rows that match the datetime 2020-01-01 17:30
4. All rows that match the day 2020-01-01
5. All rows that match the month 2020-01
6. All rows that match the year 2020
7. All rows from 2020-01-01 and 2020-12-31 (exclusive)
8. From January to April (exclusive)
9. From Monday to Saturday (exclusive)
10. Between 09:00 and 16:00 (exclusive)
11. Between 16:00 and 09:00 (exclusive)
12. Datetime components can also be combined. Here between Monday 09:00 and Friday 16:00 (exclusive).
13. Indexers can also be combined like masks. Here from Monday to Friday (inclusive) between 09:00 and 16:00 (exclusive).
14. Every year start (frequency string)
15. Every 7 days (timedelta)
16. Every Monday (mask)
17. Every business day (date offset)

Not only rows can be selected but also columns by combining rowidx and colidx instructions.

new_df = df.vbt.xloc[vbt.colidx(0)].get()  # (1)!
new_df = df.vbt.xloc[vbt.colidx("BTC-USD")].get()  # (2)!
new_df = df.vbt.xloc[vbt.colidx((10, "simple", "BTC-USD"))].get()  # (3)!
new_df = df.vbt.xloc[vbt.colidx("BTC-USD", level="symbol")].get()  # (4)!
new_df = df.vbt.xloc["2020", "BTC-USD"].get()  # (5)!

1. First column
2. Column label "BTC-USD"
3. Column label (10, "simple", "BTC-USD")
4. Any column with symbol "BTC-USD"
5. Year 2020 and column label "BTC-USD". Both can be also wrapped with rowidx and colidx respectively, for example, to pass indexer-related keyword arguments.

Pandas accessors can also be used to modify the values under some rows and columns. This isn't possible for more complex VBT objects.

entries.vbt.xloc[vbt.autoidx(slice("mon", "sat")) & vbt.autoidx("09:00")] = True  # (1)!
exits.vbt.xloc[vbt.autoidx(slice("mon", "sat")) & (vbt.autoidx("16:00") << 1)] = True  # (2)!

entries.vbt.xloc[vbt.pointidx(every="B", at_time="09:00")] = True  # (3)!
exits.vbt.xloc[vbt.pointidx(every="B", at_time="16:00", indexer_method="before")] = True

1. Saturday is exclusive
2. Operation << 1 means shifting the mask backwards to get the last time before 16:00
3. Another way by using PointIdxr with a business day frequency