def worker(tsobj):
global features_to_use
thisfeats = featurize.featurize_single_ts(
tsobj,
features_to_use=features_to_use,
custom_functions=feature_sets.custom_features,
raise_exceptions=False)
return thisfeats
def avg_double_to_single_step(df):
# http://cesium-ml.org/docs/feature_table.html
bands = df.groupby('passband')
ts = time_series.TimeSeries(t=bands['mjd'].apply(lambda x: x.values),
m=bands['flux'].apply(lambda x: x.values),
e=bands['flux_err'].apply(lambda x: x.values))
name = 'avg_double_to_single_step'
features = featurize.featurize_single_ts(
ts, features_to_use=[name])[name].to_dict()
return pd.Series(features)
positions = [1, 12076, 24634, ...]
#store pb names for labelling the features accurately
pbnames = ['u', 'g', 'r', 'i', 'z', 'Y']
#loop through each object
for i in len(positions):
#seek to it's starting position
f.seek(positions[i])
#read up to the start of the subsequent object, split by row
s = f.read(positions[i + 1] - positions[i]).split('/n')
#assign id, create empty lists for t/m/e
idnum = s.split(',')[0]
t = [[], [], [], [], [], []]
m = [[], [], [], [], [], []]
e = [[], [], [], [], [], []]
#Append each t/m/e value to the right passband list within t/m/e/ lists
for row in s.split(','):
pb = row[2]
t[pb].append(row[1])
m[pb].append(row[3])
e[pb].append(row[4])
#create ts obj, then generate features
tsobj = TimeSeries(t=t, m=m, e=e, name=idnum, channel_names=pbnames)
thisfeats = featurize.featurize_single_ts(
tsobj, features_to_use=features_to_use, raise_exceptions=False
) #this is a dict where keys are feature names and values are the respective scalars
with open('featfile.csv', 'a') as ff:
ff.write(thisfeats) #Need to check if this would actually work.
del s, idnum, pb, t, m, e, tsobj, thisfeats
list_of_indices = np.arange(
0, len(t)) #array of indicies for plotting purposes
tsdict['object_ID'].append(
current_object_id) #add id and target data to dictionary
tsdict['target'].append(current_object_target)
#create time series object for the source in question and store for transformation of test data
timeobj = TimeSeries(t=t,
m=m,
e=e,
label=current_object_target,
name=current_object_id)
#featurize the time series object from above
features_of_time_series = featurize.featurize_single_ts(
timeobj, features_to_use=feature_list, raise_exceptions=False)
#print(features_of_time_series.values)
tsdict['features'].append(
list(features_of_time_series.values
)) #add the list of time series features to the dictionary
tsdict['features'][i] += list(
train_meta_data.iloc[i, 1:11]) #add the static data features
#Going to delete the dataframes with all of the data now that it is all organized in a dictionary
del (train_data)
del (train_meta_data)
#*******************************************************************************************************************
#Now the data is organized in the tsdict dictionary object and will be used
#for the construction of a logistic regression model with l1 (LASSO) regularization
def worker(ts_obj, features_to_use):
this_feats = featurize.featurize_single_ts(ts_obj,
features_to_use=features_to_use,
raise_exceptions=False)
return this_feats