""" Compute Engine definitions for the Pipeline API. """ from abc import ( ABCMeta, abstractmethod, ) from uuid import uuid4 from six import ( iteritems, with_metaclass, ) from numpy import array from pandas import DataFrame, MultiIndex from toolz import groupby, juxt from toolz.curried.operator import getitem from zipline.lib.adjusted_array import ensure_adjusted_array, ensure_ndarray from zipline.errors import NoFurtherDataError from zipline.utils.numpy_utils import ( as_column, repeat_first_axis, repeat_last_axis, ) from zipline.utils.pandas_utils import explode from .term import AssetExists, InputDates, LoadableTerm class PipelineEngine(with_metaclass(ABCMeta)): @abstractmethod def run_pipeline(self, pipeline, start_date, end_date): """ Compute values for `pipeline` between `start_date` and `end_date`. Returns a DataFrame with a MultiIndex of (date, asset) pairs. Parameters ---------- pipeline : zipline.pipeline.Pipeline The pipeline to run. start_date : pd.Timestamp Start date of the computed matrix. end_date : pd.Timestamp End date of the computed matrix. Returns ------- result : pd.DataFrame A frame of computed results. The columns `result` correspond to the entries of `pipeline.columns`, which should be a dictionary mapping strings to instances of `zipline.pipeline.term.Term`. For each date between `start_date` and `end_date`, `result` will contain a row for each asset that passed `pipeline.screen`. A screen of None indicates that a row should be returned for each asset that existed each day. """ raise NotImplementedError("run_pipeline") class NoEngineRegistered(Exception): """ Raised if a user tries to call pipeline_output in an algorithm that hasn't set up a pipeline engine. """ class ExplodingPipelineEngine(PipelineEngine): """ A PipelineEngine that doesn't do anything. """ def run_pipeline(self, pipeline, start_date, end_date): raise NoEngineRegistered( "Attempted to run a pipeline but no pipeline " "resources were registered." ) class SimplePipelineEngine(object): """ PipelineEngine class that computes each term independently. Parameters ---------- get_loader : callable A function that is given a loadable term and returns a PipelineLoader to use to retrieve raw data for that term. calendar : DatetimeIndex Array of dates to consider as trading days when computing a range between a fixed start and end. asset_finder : zipline.assets.AssetFinder An AssetFinder instance. We depend on the AssetFinder to determine which assets are in the top-level universe at any point in time. """ __slots__ = ( '_get_loader', '_calendar', '_finder', '_root_mask_term', '_root_mask_dates_term', '__weakref__', ) def __init__(self, get_loader, calendar, asset_finder): self._get_loader = get_loader self._calendar = calendar self._finder = asset_finder self._root_mask_term = AssetExists() self._root_mask_dates_term = InputDates() def run_pipeline(self, pipeline, start_date, end_date): """ Compute a pipeline. Parameters ---------- pipeline : zipline.pipeline.Pipeline The pipeline to run. start_date : pd.Timestamp Start date of the computed matrix. end_date : pd.Timestamp End date of the computed matrix. The algorithm implemented here can be broken down into the following stages: 0. Build a dependency graph of all terms in `pipeline`. Topologically sort the graph to determine an order in which we can compute the terms. 1. Ask our AssetFinder for a "lifetimes matrix", which should contain, for each date between start_date and end_date, a boolean value for each known asset indicating whether the asset existed on that date. 2. Compute each term in the dependency order determined in (0), caching the results in a a dictionary to that they can be fed into future terms. 3. For each date, determine the number of assets passing pipeline.screen. The sum, N, of all these values is the total number of rows in our output frame, so we pre-allocate an output array of length N for each factor in `terms`. 4. Fill in the arrays allocated in (3) by copying computed values from our output cache into the corresponding rows. 5. Stick the values computed in (4) into a DataFrame and return it. Step 0 is performed by ``Pipeline.to_graph``. Step 1 is performed in ``SimplePipelineEngine._compute_root_mask``. Step 2 is performed in ``SimplePipelineEngine.compute_chunk``. Steps 3, 4, and 5 are performed in ``SimplePiplineEngine._to_narrow``. See Also -------- PipelineEngine.run_pipeline """ if end_date < start_date: raise ValueError( "start_date must be before or equal to end_date \n" "start_date=%s, end_date=%s" % (start_date, end_date) ) screen_name = uuid4().hex graph = pipeline.to_execution_plan( screen_name, self._root_mask_term, self._calendar, start_date, end_date, ) extra_rows = graph.extra_rows[self._root_mask_term] root_mask = self._compute_root_mask(start_date, end_date, extra_rows) dates, assets, root_mask_values = explode(root_mask) results = self.compute_chunk( graph, dates, assets, initial_workspace={ self._root_mask_term: root_mask_values, self._root_mask_dates_term: as_column(dates.values) }, ) return self._to_narrow( graph.outputs, results, results.pop(screen_name), dates[extra_rows:], assets, ) def _compute_root_mask(self, start_date, end_date, extra_rows): """ Compute a lifetimes matrix from our AssetFinder, then drop columns that didn't exist at all during the query dates. Parameters ---------- start_date : pd.Timestamp Base start date for the matrix. end_date : pd.Timestamp End date for the matrix. extra_rows : int Number of extra rows to compute before `start_date`. Extra rows are needed by terms like moving averages that require a trailing window of data. Returns ------- lifetimes : pd.DataFrame Frame of dtype `bool` containing dates from `extra_rows` days before `start_date`, continuing through to `end_date`. The returned frame contains as columns all assets in our AssetFinder that existed for at least one day between `start_date` and `end_date`. """ calendar = self._calendar finder = self._finder start_idx, end_idx = self._calendar.slice_locs(start_date, end_date) if start_idx < extra_rows: raise NoFurtherDataError.from_lookback_window( initial_message="Insufficient data to compute Pipeline:", first_date=calendar[0], lookback_start=start_date, lookback_length=extra_rows, ) # Build lifetimes matrix reaching back to `extra_rows` days before # `start_date.` lifetimes = finder.lifetimes( calendar[start_idx - extra_rows:end_idx], include_start_date=False ) assert lifetimes.index[extra_rows] == start_date assert lifetimes.index[-1] == end_date if not lifetimes.columns.unique: columns = lifetimes.columns duplicated = columns[columns.duplicated()].unique() raise AssertionError("Duplicated sids: %d" % duplicated) # Filter out columns that didn't exist between the requested start and # end dates. existed = lifetimes.iloc[extra_rows:].any() ret = lifetimes.loc[:, existed] shape = ret.shape assert shape[0] * shape[1] != 0, 'root mask cannot be empty' return ret def _mask_and_dates_for_term(self, term, workspace, graph, all_dates): """ Load mask and mask row labels for term. """ mask = term.mask mask_offset = graph.extra_rows[mask] - graph.extra_rows[term] # This offset is computed against _root_mask_term because that is what # determines the shape of the top-level dates array. dates_offset = ( graph.extra_rows[self._root_mask_term] - graph.extra_rows[term] ) return workspace[mask][mask_offset:], all_dates[dates_offset:] @staticmethod def _inputs_for_term(term, workspace, graph): """ Compute inputs for the given term. This is mostly complicated by the fact that for each input we store as many rows as will be necessary to serve **any** computation requiring that input. """ offsets = graph.offset out = [] if term.windowed: # If term is windowed, then all input data should be instances of # AdjustedArray. for input_ in term.inputs: adjusted_array = ensure_adjusted_array( workspace[input_], input_.missing_value, ) out.append( adjusted_array.traverse( window_length=term.window_length, offset=offsets[term, input_], ) ) else: # If term is not windowed, input_data may be an AdjustedArray or # np.ndarray. Coerce the former to the latter. for input_ in term.inputs: input_data = ensure_ndarray(workspace[input_]) offset = offsets[term, input_] # OPTIMIZATION: Don't make a copy by doing input_data[0:] if # offset is zero. if offset: input_data = input_data[offset:] out.append(input_data) return out def get_loader(self, term): return self._get_loader(term) def compute_chunk(self, graph, dates, assets, initial_workspace): """ Compute the Pipeline terms in the graph for the requested start and end dates. Parameters ---------- graph : zipline.pipeline.graph.TermGraph dates : pd.DatetimeIndex Row labels for our root mask. assets : pd.Int64Index Column labels for our root mask. initial_workspace : dict Map from term -> output. Must contain at least entry for `self._root_mask_term` whose shape is `(len(dates), len(assets))`, but may contain additional pre-computed terms for testing or optimization purposes. Returns ------- results : dict Dictionary mapping requested results to outputs. """ self._validate_compute_chunk_params(dates, assets, initial_workspace) get_loader = self.get_loader # Copy the supplied initial workspace so we don't mutate it in place. workspace = initial_workspace.copy() # If loadable terms share the same loader and extra_rows, load them all # together. loader_group_key = juxt(get_loader, getitem(graph.extra_rows)) loader_groups = groupby(loader_group_key, graph.loadable_terms) refcounts = graph.initial_refcounts(workspace) for term in graph.ordered(): # `term` may have been supplied in `initial_workspace`, and in the # future we may pre-compute loadable terms coming from the same # dataset. In either case, we will already have an entry for this # term, which we shouldn't re-compute. if term in workspace: continue # Asset labels are always the same, but date labels vary by how # many extra rows are needed. mask, mask_dates = self._mask_and_dates_for_term( term, workspace, graph, dates ) if isinstance(term, LoadableTerm): to_load = sorted( loader_groups[loader_group_key(term)], key=lambda t: t.dataset ) loader = get_loader(term) loaded = loader.load_adjusted_array( to_load, mask_dates, assets, mask, ) workspace.update(loaded) else: workspace[term] = term._compute( self._inputs_for_term(term, workspace, graph), mask_dates, assets, mask, ) if term.ndim == 2: assert workspace[term].shape == mask.shape else: assert workspace[term].shape == (mask.shape[0], 1) # Decref dependencies of ``term``, and clear any terms whose # refcounts hit 0. for garbage_term in graph.decref_dependencies(term, refcounts): del workspace[garbage_term] out = {} graph_extra_rows = graph.extra_rows for name, term in iteritems(graph.outputs): # Truncate off extra rows from outputs. out[name] = workspace[term][graph_extra_rows[term]:] return out def _to_narrow(self, terms, data, mask, dates, assets): """ Convert raw computed pipeline results into a DataFrame for public APIs. Parameters ---------- terms : dict[str -> Term] Dict mapping column names to terms. data : dict[str -> ndarray[ndim=2]] Dict mapping column names to computed results for those names. mask : ndarray[bool, ndim=2] Mask array of values to keep. dates : ndarray[datetime64, ndim=1] Row index for arrays `data` and `mask` assets : ndarray[int64, ndim=2] Column index for arrays `data` and `mask` Returns ------- results : pd.DataFrame The indices of `results` are as follows: index : two-tiered MultiIndex of (date, asset). Contains an entry for each (date, asset) pair corresponding to a `True` value in `mask`. columns : Index of str One column per entry in `data`. If mask[date, asset] is True, then result.loc[(date, asset), colname] will contain the value of data[colname][date, asset]. """ if not mask.any(): # Manually handle the empty DataFrame case. This is a workaround # to pandas failing to tz_localize an empty dataframe with a # MultiIndex. It also saves us the work of applying a known-empty # mask to each array. # # Slicing `dates` here to preserve pandas metadata. empty_dates = dates[:0] empty_assets = array([], dtype=object) return DataFrame( data={ name: array([], dtype=arr.dtype) for name, arr in iteritems(data) }, index=MultiIndex.from_arrays([empty_dates, empty_assets]), ) resolved_assets = array(self._finder.retrieve_all(assets)) dates_kept = repeat_last_axis(dates.values, len(assets))[mask] assets_kept = repeat_first_axis(resolved_assets, len(dates))[mask] final_columns = {} for name in data: # Each term that computed an output has its postprocess method # called on the filtered result. # # As of Mon May 2 15:38:47 2016, we only use this to convert # LabelArrays into categoricals. final_columns[name] = terms[name].postprocess(data[name][mask]) return DataFrame( data=final_columns, index=MultiIndex.from_arrays([dates_kept, assets_kept]), ).tz_localize('UTC', level=0) def _validate_compute_chunk_params(self, dates, assets, initial_workspace): """ Verify that the values passed to compute_chunk are well-formed. """ root = self._root_mask_term clsname = type(self).__name__ # Writing this out explicitly so this errors in testing if we change # the name without updating this line. compute_chunk_name = self.compute_chunk.__name__ if root not in initial_workspace: raise AssertionError( "root_mask values not supplied to {cls}.{method}".format( cls=clsname, method=compute_chunk_name, ) ) shape = initial_workspace[root].shape implied_shape = len(dates), len(assets) if shape != implied_shape: raise AssertionError( "root_mask shape is {shape}, but received dates/assets " "imply that shape should be {implied}".format( shape=shape, implied=implied_shape, ) )