def get_dict_features_from_df_parallel(self, df, nworkers=8):
print("extracting features...")
df_split = np.array_split(df, nworkers)
pool = Pool(nworkers)
res_dicts = pool.map(self.get_dict_features_from_df, df_split)
pool.close(
) # informs the processor that no new tasks will be added to the pool
pool.join(
) # stops and waits for all of the results to be finished and collected before proceeding with the rest of
big_dic = defaultdict(lambda: defaultdict(int))
# merge feature dictionaries created for data frame splits into one big dictionary
for dic in res_dicts:
for k, v in dic.items():
big_dic[k] = v
return pd.concat([
pd.get_dummies(df[df.columns.difference(["event", "venue"])],
prefix="@",
columns=["month", "weekday"]),
pd.DataFrame.from_dict(big_dic, orient='index')
],
axis=1,
join_axes=[df.index]).fillna(0.)
def parallelize_dataframe(self, df, func): df_split = np.array_split(df, 1) pool = Pool(1) rr = pool.map(func, df_split) df = pd.concat(rr) pool.close() pool.join() return df
def parallel_launcher(data_dir, data, worker, pool_size, files_num):
files = modified_get_files(data_dir)
batches = [(files[i:i + files_num], data, j)
for j, i in enumerate(range(0, len(files), files_num))]
pool = Pool(pool_size)
output = pool.starmap(worker, batches)
pool.close()
pool.join()
return output
