basically delirious lol

app.py

@@ -224,57 +224,62 @@ def pluck_img(user_id, user_emb):
 def background_next_image():
     global prevs_df
     global is_started
-
-    # only let it get N (maybe 3) ahead of the user
-    #not_rated_rows = prevs_df[[i[1]['user:rating'] == {' ': ' '} for i in prevs_df.iterrows()]]
-    rated_rows = prevs_df[[i[1]['user:rating'] != {' ': ' '} for i in prevs_df.iterrows()]]
-    while len(rated_rows) < 4:
-        time.sleep(.2)
-
-    user_id_list = set(rated_rows['latest_user_to_rate'].to_list())
-    for uid in user_id_list:
-        rated_rows = prevs_df[[i[1]['user:rating'].get(uid, None) is not None for i in prevs_df.iterrows()]]
-        not_rated_rows = prevs_df[[i[1]['user:rating'].get(uid, None) is None for i in prevs_df.iterrows()]]
-        
-        # we need to intersect not_rated_rows from this user's embed > 7. Just add a new column on which user_id spawned the 
-        #   media. 
-        
-        unrated_from_user = not_rated_rows[[i[1]['from_user_id'] == uid for i in not_rated_rows.iterrows()]]
-        rated_from_user = rated_rows[[i[1]['from_user_id'] == uid for i in rated_rows.iterrows()]]
-
-        # we pop previous ratings if there are > 10
-        if len(rated_from_user) >= 10:
-            oldest = rated_from_user.iloc[0]['paths']
-            prevs_df = prevs_df[prevs_df['paths'] != oldest]
-            print('popping 10th rated')
-        # we don't compute more after 10 are in the queue for them
-        if len(unrated_from_user) >= 10:
-            print('skipping because >=10 in queue unrated for user')
-            continue
-        
-        print(uid)
-        embs, ys = pluck_embs_ys(uid)
-        
-        user_emb = get_user_emb(embs, ys)
-        img, embs = generate(user_emb)
-        print(img)
-        if img:
-            tmp_df = pd.DataFrame(columns=['paths', 'embeddings', 'ips', 'user:rating', 'latest_user_to_rate'])
-            tmp_df['paths'] = [img]
-            tmp_df['embeddings'] = [embs]
-            tmp_df['user:rating'] = [{' ': ' '}]
-            tmp_df['from_user_id'] = [uid]
-            prevs_df = pd.concat((prevs_df, tmp_df))
-            # we can free up storage by deleting the image
-            if len(prevs_df) > 50:
-                oldest_path = prevs_df.iloc[6]['paths']
-                if os.path.isfile(oldest_path):
-                    os.remove(oldest_path)
-                else:
-                    # If it fails, inform the user.
-                    print("Error: %s file not found" % oldest_path)
-                # only keep 50 images & embeddings & ips, then remove oldest besides calibrating
-                prevs_df = pd.concat((prevs_df.iloc[:6], prevs_df.iloc[7:]))
+    while True:
+        # only let it get N (maybe 3) ahead of the user
+        #not_rated_rows = prevs_df[[i[1]['user:rating'] == {' ': ' '} for i in prevs_df.iterrows()]]
+        rated_rows = prevs_df[[i[1]['user:rating'] != {' ': ' '} for i in prevs_df.iterrows()]]
+        while len(rated_rows) < 4:
+        #    not_rated_rows = prevs_df[[i[1]['user:rating'] == {' ': ' '} for i in prevs_df.iterrows()]]
+            rated_rows = prevs_df[[i[1]['user:rating'] != {' ': ' '} for i in prevs_df.iterrows()]]
+            time.sleep(.01)
+            print('all users have 4 or less rows rated')
+
+        user_id_list = set(rated_rows['latest_user_to_rate'].to_list())
+        for uid in user_id_list:
+            rated_rows = prevs_df[[i[1]['user:rating'].get(uid, None) is not None for i in prevs_df.iterrows()]]
+            not_rated_rows = prevs_df[[i[1]['user:rating'].get(uid, None) is None for i in prevs_df.iterrows()]]
+            
+            # we need to intersect not_rated_rows from this user's embed > 7. Just add a new column on which user_id spawned the 
+            #   media. 
+            
+            unrated_from_user = not_rated_rows[[i[1]['from_user_id'] == uid for i in not_rated_rows.iterrows()]]
+            rated_from_user = rated_rows[[i[1]['from_user_id'] == uid for i in rated_rows.iterrows()]]
+
+            # we pop previous ratings if there are > 10
+            if len(rated_from_user) >= 10:
+                oldest = rated_from_user.iloc[0]['paths']
+                prevs_df = prevs_df[prevs_df['paths'] != oldest]
+            # we don't compute more after 10 are in the queue for them
+            if len(unrated_from_user) >= 10:
+                continue
+            
+            if len(rated_rows) < 4:
+                print(f'latest user {uid} has < 4 rows') # or > 7 unrated rows')
+                continue
+            
+            print(uid)
+            embs, ys = pluck_embs_ys(uid)
+            
+            user_emb = get_user_emb(embs, ys)
+            img, embs = generate(user_emb)
+            print(img)
+            if img:
+                tmp_df = pd.DataFrame(columns=['paths', 'embeddings', 'ips', 'user:rating', 'latest_user_to_rate'])
+                tmp_df['paths'] = [img]
+                tmp_df['embeddings'] = [embs]
+                tmp_df['user:rating'] = [{' ': ' '}]
+                tmp_df['from_user_id'] = [uid]
+                prevs_df = pd.concat((prevs_df, tmp_df))
+                # we can free up storage by deleting the image
+                if len(prevs_df) > 50:
+                    oldest_path = prevs_df.iloc[6]['paths']
+                    if os.path.isfile(oldest_path):
+                        os.remove(oldest_path)
+                    else:
+                        # If it fails, inform the user.
+                        print("Error: %s file not found" % oldest_path)
+                    # only keep 50 images & embeddings & ips, then remove oldest besides calibrating
+                    prevs_df = pd.concat((prevs_df.iloc[:6], prevs_df.iloc[7:]))
     
 
 def pluck_embs_ys(user_id):
@@ -292,14 +297,14 @@ def pluck_embs_ys(user_id):
     return embs, ys
 
 def next_image(calibrate_prompts, user_id):
-    global is_started
     print(prevs_df)
+    
     with torch.no_grad():
         if len(calibrate_prompts) > 0:
             print('######### Calibrating with sample media #########')
             cal_video = calibrate_prompts.pop(0)
             image = prevs_df[prevs_df['paths'] == cal_video]['paths'].to_list()[0]
-            is_started = True
+            
             return image, calibrate_prompts
         else:
             print('######### Roaming #########')
@@ -312,31 +317,21 @@ def next_image(calibrate_prompts, user_id):
 
 
 
-scheduler = BackgroundScheduler()
-#thread = threading.Thread(target=background_next_image)
 
-is_started = False
 
-def bg_check(_):
-    global is_started
-    global scheduler
-    if not is_started:
-        scheduler.add_job(func=background_next_image, trigger="interval", seconds=.1)
-        scheduler.start()
-        is_started = True
+
 
 def start(_, calibrate_prompts, user_id, request: gr.Request):
     global is_started
-    global scheduler
     user_id = int(str(time.time())[-7:].replace('.', ''))
     image, calibrate_prompts = next_image(calibrate_prompts, user_id)
-    
-
+    if not is_started:
+        background_next_image()
     return [
             gr.Button(value='Like (L)', interactive=True), 
-            gr.Button(value='_', interactive=False, visible=False),
+            gr.Button(value='Neither (Space)', interactive=True, visible=False), 
             gr.Button(value='Dislike (A)', interactive=True),
-            gr.Button(value=bg_check, interactive=False),
+            gr.Button(value='Start', interactive=False),
             image,
             calibrate_prompts,
             user_id
@@ -454,7 +449,7 @@ Explore the latent space without text prompts based on your preferences. Learn m
         img.play(l, js='''document.querySelector('[data-testid="Lightning-player"]').loop = true''')
     with gr.Row(equal_height=True):
         b3 = gr.Button(value='Dislike (A)', interactive=False, elem_id="dislike")
-        b2 = gr.Button(value='_', interactive=False, visible=False)
+        b2 = gr.Button(value='Neither (Space)', interactive=False, elem_id="neither", visible=False)
         b1 = gr.Button(value='Like (L)', interactive=False, elem_id="like")
         b1.click(
         choose, 
@@ -489,7 +484,12 @@ Explore the latent space without text prompts based on your preferences. Learn m
 log = logging.getLogger('log_here')
 log.setLevel(logging.ERROR)
 
+#scheduler = BackgroundScheduler()
+#scheduler.add_job(func=background_next_image, trigger="interval", seconds=.1)
+#scheduler.start()
 
+#thread = threading.Thread(target=background_next_image,)
+#thread.start()
 
 @spaces.GPU()
 def encode_space(x):