multiprocessing

engine.py

@@ -1,6 +1,6 @@
 from simpful import *
 
-FS = FuzzySystem()
+FS = FuzzySystem(show_banner=False)
 
 # Define fuzzy sets for the variable
 # RELEASE_YEAR

main.py

@@ -5,9 +5,13 @@
 !pip install "uvicorn[standard]"
 !uvicorn main:app --reload
 """
+import multiprocessing
+import time
+from multiprocessing import Pool
 
 import numpy as np
 import pandas as pd
+import pandas.core.series
 from fastapi import FastAPI
 from scipy.spatial.distance import cosine
 from sklearn.preprocessing import MultiLabelBinarizer
@@ -18,9 +22,11 @@ app = FastAPI()
 data = pd.DataFrame()
 
 
-def inference(first_id: str, second_id: str):
-    first = data.loc[first_id]
-    second = data.loc[second_id]
+def inference(first: pandas.core.series.Series, second_id: str, df=None):
+    if df is not None:
+        second = df.loc[second_id]
+    else:
+        second = data.loc[second_id]
 
     year_diff = int(first['release_year'] - second['release_year'])
     FS.set_variable('RELEASE_YEAR', year_diff)
@@ -40,9 +46,15 @@ def inference(first_id: str, second_id: str):
     emotion_diff = 1 - cosine(first['emotions'], second['emotions'])
     FS.set_variable('EMOTIONS', emotion_diff)
 
-    return FS.inference(['RECOMMENDATION'])
+    return second_id, FS.inference(['RECOMMENDATION'])['RECOMMENDATION']
 
 
+def process_dataframe(df, production):
+    scores = []
+    for index, row in df.iterrows():
+        scores.append(inference(production, str(index), df))
+    return scores
+
 
 @app.on_event('startup')
 async def startup_event():
@@ -66,7 +78,7 @@ def rec_score(first_id: str, second_id: str):
     except KeyError:
         return {'error': f'{second_id} is not a valid id'}
 
-    return inference(first_id, second_id)
+    return inference(first, second_id)
 
 
 @app.get('/recs/{production_id}')
@@ -77,12 +89,17 @@ async def recs(production_id: str, count: int | None):
         return {'error': f'{production_id} is not a valid id'}
 
     scores = []
+    time_start = time.time()
+    cpus = multiprocessing.cpu_count()
+    df_list = np.array_split(data, cpus)
+    pool = Pool(cpus)
+    results = [pool.apply_async(process_dataframe, [df, first]) for df in df_list]
 
-    for index, row in data.iterrows():
-        if str(index) == production_id:
-            continue
-        scores.append((index, inference(production_id, str(index))['RECOMMENDATION']))
-
-    scores = [idx[0] for idx in sorted(scores, key=lambda x: x[1], reverse=True)[:count]]
-
-    return list(scores)
+    for r in results:
+        r.wait()
+    for r in results:
+        scores += r.get()
+    print(f'time elapsed = {time.time() - time_start}')
+    scores = [idx[0] for idx in sorted(scores, key=lambda x: x[1], reverse=True)[:count+1]]
+    scores.remove(production_id)
+    return scores