Limited use of `pandas` data structures in both `HistoryContainer` and
`RollingPanel`. Where possible, methods were amended to return raw
`ndarrays` with the indexing logic done separately. This allows us to
cut down the number of times pandas objects are created both as returns
and intermediate values. The separation of indexing from data access
allowed us to minimize the times we’d make use of pandas indexes.
This required that that certain methods like `NDFrame.ffill` be replaced
with versions that work with `ndarrays`. Some of this was done via
straight numpy methods and others by access pandas internal
machinery. Outside of allowing us to use faster ndarrays, many of these
function provided speedups over their pandas counterparts as we didn’t
require the extra features like handling multiple dtypes. i.e. np.isnan
is faster than pd.isnull, but only works with certain dtypes.
and added a new test case
was not iterating over lookup date directory names, and
therefore mising all by one list of stocks.
discovered because of differing sort orders between
my local machine, other devs, and travis ci.
getting filled with the wrong datetimes and causing errors.
Updates the logic for addressing missing datetimes and adds unit tests
for the 2 main cases (no missing datetimes, and some missing datetimes).
script.
Adds an option kwarg to TradingAlgorithm named 'algo_filename' that
represents the file where the algoscript came from (if any). The
run_algo.py script will pass this argument with the value passed to the
'-f' flag. The default name is '<string>' to represent that the script
is coming from a string in python and not a file. This matches the
behavior of exec and the python convention for compiling code objects.
makes it an offset from 13:30 UTC.
This is to be more consistent with the market_close, which is an offset
from 20:00 UTC.
This also makes market_open and market_close cache the dt to offset from
for each day.
Previously, all specs had to be pre-allocated by using the 'add_history'
function. This is now no longer required and instead serves as a hint to
the HistoryContainer to pre-allocate the space for the given spec.
History can grow by increasing the length for a frequency, adding a
frequency, or adding a field. It can grow with any combination of
these.
HistoryContainer now is aware of the data_frequency of the algorithm,
and no longer uses the daily_at_midnight flag; instead, this is the
default behavior.
- NotHalfDay only worked at midnight
- week_(start|end) were actually month_(start|end)
- Removes check_args from api.
- Default offset of 30mins for market_(open|close)
zipline.utils.events is imported.
Changes the class level attribute `env` on EventRule to a property so
that the environment is only looked up at when needed.
schedule_function takes a date rule, a time rule, and a function and
will call the function, passing context and data only when the two rules
fire. This allows for code that is conditional to the datetime of the
algo.
This is implemented internally with `Event` objects which are pairings
of `EventRule`s and callbacks.
handle_data becomes a special event with a rule that always fires. This
makes the logic for handling events more complete and compact.
Overhaul the core HistoryContainer logic to be more robust to changing
universes.
Major Changes
-------------
* Remove `return_frame` cache. The original purpose of using
return_frames was to avoid having to create new DataFrames on each
iteration of handle_data, but we ended up having to copy the return
frames anyway because user code could mutate the frames in place.
Removing the return_frames reduces unnecessary copying, and reduces
the logic of `get_history` to just forward-filling and concatenating
two DataFrames.
* Use a `MultiIndex`ed DataFrame to represent
`last_known_prior_values`. This makes lookups faster and greatly
simplifies the logic of adding and dropping sids.
* HistoryContainer no longer attempts to determine its universe based on
the contents of its internal buffers. The TradingAlgorithm
controlling the container is now responsible for explicitly calling
`add_sids` or `drop_sids` when securities enter or leave the
algorithm's universe. These methods, along with the internal
`_realign` method, provide a clean interface for changing the universe
of securities managed by the container.
* Refactor index mutation logic in `RollingPanel` into a
`MutableIndexRollingPanel` subclass. Maintenance of the old behavior
is regrettably necessary to support `BatchTransform`.
* Refactor shared logic from `roll` and `get_history` into a single
`aggregate_ohlcv_panel` method that's responsible for collapsing an
OHLCV buffer into a frame.
Removes support for handling dividends as part of the algorithm
simulation stream, replacing it with an API in `TradingAlgorithm` for
supplying dividends as a DataFrame.
Previously, calling order() in initalize resulted in a weird
stack trace. It now returns a well formulated error that is
readable to the user through the API. Adding a slippage
kwarg to test_algorithm and simfactor was necessary because
slippage can only be called during init. Previously initaliazed
was never set to true and calls to init-only function were sprinkled
around the code in non-init sections. Code changes were to enforce
init-only rules.
There were sevaral places you could supply sim_params
in TradingAlgorithm (__init__, run). This got confusing
as its not clear who updated what and which one was the
correct one to use at each time.
Then there were to ways to define data_frequency, one in
__init__() and one in the sim_params which also added code
complexity.
This refactor makes it explicit that sim_params are to be
passed to __init__() only. Moreover, data_frequency is
only stored in sim_params. For backwards compatibility,
it can still be supplied separately but will link to
the one in sim_params.
For example, you could create new sim params via:
sim_params = create_simulation_parameters(data_frequency='minute')
algo = MyAlgo(sim_params)
algo.run(data)
In addition, perf_tracker only gets initialized in one place:
_create_generator() which should also make the various ways
of running an algorithm more deterministic.
This also fixes a bug with SimulationParameters where
you could not change the period_start. Unfortunately, the
current implementation still requieres an implicit call to
update the internal variables.
There quite some bugs in certain corner cases. Dropping of obsolete
axes was not working correctly, roll over could cause obsolete axes
to not drop. The tests are much more stringent now as well.
The IPython magic still created an output file because
the output argument was only removed after the pipeline
was run. This fix simply removes the argument before
the call to run_pipline() when running the IPython magic.
Overhauls `HistoryContainer` in prep for support of more than one frequency.
Major changes:
- Methods/variables referring to "day" have been renamed/generalized.
- `current_day_panel` became `buffer_panel`, which is now a `RollingPanel`
- `prior_day_panel` became a dictionary mapping `Frequency` objects to
"digest panels", which are instances of `RollingPanel`.
- Hard-coded daily rollover replaced with a notion of a "current window" for
each unique frequency managed by the panel.
- When the end of the current window is reached for a given frequency, we
compute an aggregate bar (code refers to this as a "digest"), which is
appended to a panel associated with that frequency.
- Window rollover dates are managed by a pair of dictionaries,
`cur_window_starts` and `cur_window_closes`. The `Frequency` class is
responsible for computing window bounds based on the open/close of the
previous window.
- Semantic change to the `open_price` field: `open_price` now always
contains the price of the first trade occurring in the given window.
Previously it contained the price of the first minute in the window,
returning NaN it the security happened not to trade in the first minute.
Adds four new methods to the Zipline API that can be used as circuit-breakers
to interrupt the execution of an algorithm. The API methods are:
`set_max_position_size`
`set_max_order_size`
`set_max_order_count`
`set_long_only`
Internally, these methods are implemented by each registering a TradingControl
callback object with the TradingAlgorithm. During
TradingAlgorithm.__validate_order_params (and thus before any side-effects of
the order call occur), each callback's `validate` method is called with
information about the order to be placed and the algorithm's current state,
raising an exception if the callback detects that an error condition has been breached.
When zipline is imported it checks whether
it runs in the IPython notebook. If it does,
it registers a %%zipline magic that takes the
same arguments as the CLI with the addition of
a -o for specifying the output variable to store
the performance frame in.
The algo code in the cell is, as of yet, executed
in its own environment rather than that of the
IPython NB which is probably what we want.
Also adds cli option to save the perf dataframe
to a pickle file.
Also adds an IPython notebook buyapple example.
Add a CLI that reads in an algorithm, loads data,
run the algorithm, and output performance metrics.
The examples are adapted to the new zipline API and
analyses are split into separate files.
Also add config files that run the example
algorithms with preset settings.
recent 2 for 1 stock split, where 1 class C share was distributed
for each share of class A held.
Now a dividend can specify a sid and ratio of stock that will be paid
to owners of the original security. If the ratio is 2.0, then for every
existing share, two shares will be paid.
TradingAlgorithm always uses set_algo_instance in pairs of
set_algo_instance(self) and set_algo_instance(None). Refactoring this to use a
context manager.
This creates a data source for csv and hdf5 files, a generator to create a sample csv, and a pytables generator to go from a list of dated gzipped csv's in a directory to a pytables data source.
This does not add a unittest yet which we should write for the future.
Dale Jung