def download_yahoo(asset_list):
MS = MongoStorage()
z = MS.Find({
"Name": {
"$in": asset_list
},
"SeriesType": {
"$in": ["yahoo"]
}
})
exist_asset = [j["Name"] for j in z]
to_download_asset_list = [j for j in asset_list if j not in exist_asset]
to_download_asset_list
for asset in to_download_asset_list:
try:
df = query_data(asset)
df[asset] = df["Adj Close"].pct_change()
df = df.dropna()
rdf = df[[asset]]
if rdf.shape[0] > 100:
rs = RSeries()
rs.load_custom_series(rdf,
custom_series_name=asset,
series_type="yahoo")
rs.SaveSeries()
else:
print("Not enough data:", asset)
except Exception as err:
print(err)
print("Error:", asset)
def get_benchmark(benchmark, client=None):
MS = MongoStorage(client=client)
id_list = MS.Find(find_filter={
"SeriesType": "Benchmark",
"Name": benchmark
},
id_only=True)
rs = MS.Load(id_list)[0]
return rs.df
class SingleIntervalAllocator(object):
# Example args:
# s1, s2 = datetime.datetime(2017,1,1), datetime.datetime(2018,1,1)
# a1, a2 = datetime.datetime(2018,1,1), datetime.datetime(2019,1,1)
# f1, f2 = datetime.datetime(2019,1,1), datetime.datetime(2019,6,1)
# find_filter, sample_threshold = 2
# sample_population_n = 10000
# sample_population_seed = 123
# gen_weight = False
# excluder_kwargs = [
# {"name": "exclude_high_correlation", "kwargs": {"corr_threshold": 0.9}},
# {"name": "exclude_seasonality_month", "kwargs": {}},
# {"name": "exclude_data_mined", "kwargs": {}}
# ] # apply each in sequence
# selector_kwargs = {"select_method": "select_all"}
# allocator_kwargs = {"allocate_method": "constrained_risk_parity", "upperbound": 0.3, "adj_lowerbound": 0.008, "strat_min_alloc": {}}
# evaluator_kwargs = {"f1": None, "f2": None}
def __init__(self,
s1,
s2,
a1,
a2,
excluder_kwargs,
selector_kwargs,
allocator_kwargs,
find_filter,
sample_threshold=2,
sample_population_n=10000,
sample_population_seed=123,
f1=None,
f2=None,
**kwargs):
date_kwargs = self.construct_dates(
s1, s2, a1, a2, f1, f2
) # Construct boundary dates for filtering population strategiess
selector_kwargs = {**selector_kwargs, **date_kwargs}
allocator_kwargs = {**allocator_kwargs, **date_kwargs}
excluded = {}
self.MS = MongoStorage() # Initialize MongoStorage
meta_list = self.get_full_meta(
find_filter, **date_kwargs) # Get full population metadata
if len(meta_list) >= sample_threshold:
track_df, excluded_0 = self.sample_from_population(
meta_list, n=sample_population_n, seed=sample_population_seed)
# Exclude, Select, Allocate
track_df, excluded_1 = Excluder(track_df, date_kwargs,
excluder_kwargs).get_output()
track_df, excluded_2 = Selector(track_df,
**selector_kwargs).get_output()
track_df = Allocator(track_df, **allocator_kwargs).get_output()
excluded = {**excluded_0, **excluded_1, **excluded_2}
forward_df = self.construct_forward_series(track_df, f1, f2)
self.track_df, self.excluded, self.forward_df = track_df, excluded, forward_df
def get_output(self, subset=True):
if subset:
return self.subset_track_df(
self.track_df), self.excluded, self.forward_df
else:
return self.track_df, self.excluded, self.forward_df
def construct_dates(self, s1, s2, a1, a2, f1, f2, **kwargs):
startdate, enddate = min(s1, a1), max(s2, a2)
date_kwargs = {
"s1": s1,
"s2": s2,
"a1": a1,
"a2": a2,
"f1": f1,
"f2": f2,
"startdate": startdate,
"enddate": enddate
}
return date_kwargs
def get_full_meta(self, find_filter, startdate, enddate, **kwargs):
# filter_dict = {"User": {"$nin": ["Deleted"], "$regex": user_regex_filter}, "StartDate": {"$lte": b_startdate}, "EndDate": {"$gte": f_enddate}}
find_filter["StartDate"] = {"$lte": startdate}
find_filter["EndDate"] = {"$gte": enddate}
meta_list = [j for j in self.MS.Find(find_filter)]
new_meta_list = pd.DataFrame(meta_list).sort_values("EndDate").groupby(
"Name").last().reset_index().to_dict(orient="records")
if len(meta_list) > len(new_meta_list):
print("DUPLICATES FOUND! Duplicates will be removed...")
# self.meta_list = meta_list #debug
return new_meta_list
def sample_from_population(self, meta_list, n=10000, seed=123, **kwargs):
"""Sample from population strategies, return id_list for loading."""
if len(meta_list) == 0:
# self.log("SampleFromPopulation: No strategies found!", "SampleFromPopulation")
raise Exception(
"No strategies found! (Likely due to date constraints)")
fdf = df = pd.DataFrame(meta_list)
N = df.shape[0]
exclude_sample_meta_df = pd.DataFrame()
if n >= N:
full_meta = meta_list
else:
df["filter"] = df.groupby("Name")["EndDate"].apply(
lambda x: x == np.max(x))
fdf = df[df["filter"]].sort_values("Name")
fdf = fdf.set_index("Name")
fdf = fdf.loc[~fdf.index.duplicated(keep='last')]
fdf = fdf.reset_index()
fdf_full = fdf.copy()
fdf = fdf_full.sample(n, random_state=seed).copy()
exclude_sample_meta_df = fdf_full.loc[
~fdf_full["Name"].isin(fdf["Name"]), :].copy()
# self.log("SampleStrategies: {}".format(str(fdf["Name"].values)), "SampleFromPopulation_2")
id_list = [_id for _id in fdf["_id"].values]
fdf["rs"] = [self.MS.Load([_id])[0] for _id in fdf["_id"].values]
excluded = {}
excluded['exclude_sample'] = exclude_sample_meta_df
return fdf, excluded
def subset_track_df(self, track_df):
needed_columns = ["Name", "StartDate", "OOSDate", "EndDate", "_id"]
p1 = track_df.loc[:, track_df.columns.str.startswith("weight")]
p2 = track_df.loc[:, needed_columns]
sub_track_df = pd.concat([p2, p1], axis=1)
return sub_track_df
def construct_forward_series(self, track_df, f1, f2, **kwargs):
rs_list = track_df["rs"].values
w_list = track_df["weight"].values
dfc = get_df_combined_from_rs_list(rs_list=rs_list, d1=f1, d2=f2)
allocated_df = dfc * w_list
allocated_df = allocated_df.fillna(0)
portfolio_df = pd.DataFrame(allocated_df.sum(axis=1), columns=["pret"])
portfolio_df["pcret"] = cumulative_returns(portfolio_df["pret"])
return portfolio_df.copy()
# if self.meta_df.shape[0] > 0:
# self.Select(**kwargs)
# self.ExcludeZeroVariance(**kwargs)
# self.Allocate(**kwargs)
# if not gen_weight:
# self.BackwardTest(b1 = a1, b2 = a2)
# # self.ForwardTest(**kwargs)
# self.ForwardTest(f1 = f1, f2 = f2)
# self.Evaluate()
# self.format_meta_df()
# Debug
# try:
# if self.meta_df.shape[0] > 0:
# self.Select(**kwargs)
# self.ExcludeZeroVariance(**kwargs)
# self.Allocate(**kwargs)
# if not gen_weight:
# self.ForwardTest(**kwargs)
# self.Evaluate()
# self.format_meta_df()
# except Exception as err:
# print("Debug:", err)
# @timeit
# def ForwardTest(self, f1, f2, **kwargs):
# rs_list = self.meta_df["rs"].values
# w_list = self.meta_df["weight"].values
# dfc = get_df_combined_from_rs_list(rs_list = rs_list, d1 = f1, d2 = f2)
# allocated_df = dfc*w_list
# allocated_df = allocated_df.fillna(0)
# portfolio_df = pd.DataFrame(allocated_df.sum(axis = 1), columns = ["pret"])
# portfolio_df["pcret"] = cumulative_returns(portfolio_df["pret"])
# self.portfolio_df = portfolio_df.copy()
# @timeit
# def BackwardTest(self, b1, b2, **kwargs):
# rs_list = self.meta_df["rs"].values
# w_list = self.meta_df["weight"].values
# dfc = get_df_combined_from_rs_list(rs_list = rs_list, d1 = b1, d2 = b2)
# # todel1
# self.debug1 = dfc.copy()
# self.debug2 = self.meta_df["weight"]
# # todel2
# allocated_df = dfc*w_list
# allocated_df = allocated_df.fillna(0)
# portfolio_df = pd.DataFrame(allocated_df.sum(axis = 1), columns = ["pret"])
# portfolio_df["pcret"] = cumulative_returns(portfolio_df["pret"])
# self.is_portfolio_df = portfolio_df.copy()
# @timeit
# def Evaluate(self):
# rs = RSeries()
# df = self.portfolio_df[["pret"]]
# rs.load_custom_series(df, custom_series_name = "PortfolioSegment", series_type = "PortfolioSegment")
# metrics = rs.get_metrics_by_date(d1 = df.index[0], d2 = df.index[-1])
# self.oos_metrics = metrics.copy()
# rs = RSeries()
# df = self.is_portfolio_df[["pret"]]
# rs.load_custom_series(df, custom_series_name = "PortfolioSegment", series_type = "PortfolioSegment")
# metrics = rs.get_metrics_by_date(d1 = df.index[0], d2 = df.index[-1])
# self.is_metrics = metrics.copy()
# def format_meta_df(self):
# mdf = self.meta_df.copy()
# mdf_rs = self.exclude_sample_meta_df
# mdf_e1 = self.exclude_corr_meta_df
# mdf_e2 = self.exclude_season_meta_df
# mdf_es = self.exclude_select_meta_df
# mdf_is = self.exclude_is_meta_df
# mdf_ezv = self.exclude_zerovariance_meta_df
# mdf["status"] = "selected"
# mdf_rs["status"] = "exclude_sample"
# mdf_rs["weight"] = 0
# mdf_e1["status"] = "exclude_corr"
# mdf_e1["weight"] = 0
# mdf_e2["status"] = "exclude_season"
# mdf_e2["weight"] = 0
# mdf_es["status"] = "exclude_select"
# mdf_es["weight"] = 0
# mdf_is["status"] = "exclude_is"
# mdf_is["weight"] = 0
# mdf_ezv["status"] = "exclude_zerovariance"
# mdf_ezv["weight"] = 0
# mdf = pd.concat([mdf, mdf_rs, mdf_e1, mdf_e2, mdf_es, mdf_is, mdf_ezv])
# try:
# mdf = mdf[["Name", "_id", "User", "StartDate", "EndDate", "DateAdded", "status", "weight"]]
# except:
# mdf = mdf[["Name", "_id", "User", "StartDate", "EndDate", "status", "weight"]] # quick fix on ExcludeIS (disabled = 1)
# mdf["s1"] = self.s1
# mdf["s2"] = self.s2
# mdf["a1"] = self.a1
# mdf["a2"] = self.a2
# mdf["f1"] = self.f1
# mdf["f2"] = self.f2
# self.clean_meta_df = mdf