random calibration & they"re kept in the df

app.py

@@ -102,11 +102,17 @@ def sample_embs(prompt_embeds):
 def get_user_emb(embs, ys):
     positives = [e for e, ys in zip(embs, ys) if ys == 1]
     embs = random.sample(positives, min(8, len(positives)))
-    positives = torch.stack(embs, 1)
+    if len(embs) == 0:
+        positives = torch.zeros_like(im_emb)[None]
+    else:
+        positives = torch.stack(embs, 1)
 
     negs = [e for e, ys in zip(embs, ys) if ys == 0]
     negative_embs = random.sample(negs, min(8, len(negs)))
-    negatives = torch.stack(negative_embs, 1)
+    if len(negative_embs) == 0:
+        negatives = torch.zeros_like(im_emb)[None]
+    else:
+        negatives = torch.stack(negative_embs, 1)
 
     image_embeds = torch.stack([sample_embs(negatives), sample_embs(positives)])
 
@@ -202,10 +208,12 @@ def pluck_img(user_id):
 
 def next_image(calibrate_prompts, user_id):
     with torch.no_grad():
-        if len(calibrate_prompts) > 0:
-            cal_video = calibrate_prompts.pop(0)
+        # once we've done so many random calibration prompts out of the full media
+        if len(m_calibrate) - len(calibrate_prompts) < 5:
+            cal_video = calibrate_prompts.pop(random.randint(0, len(calibrate_prompts)-1))
             image = prevs_df[prevs_df['paths'] == cal_video]['paths'].to_list()[0]
             return image, calibrate_prompts
+        # we switch to just getting media by similarity.
         else:
             image = pluck_img(user_id)
             return image, calibrate_prompts
@@ -330,7 +338,7 @@ Explore the latent space without text prompts based on your preferences. [rynmur
     ''', elem_id="description")
     user_id = gr.State()
     # calibration videos -- this is a misnomer now :D
-    calibrate_prompts = gr.State( [l for l in random.sample(glob.glob('image_init/*'), k=8)] )
+    calibrate_prompts = gr.State( glob.glob('image_init/*') )
     def l():
         return None
 
@@ -428,8 +436,8 @@ for im in m_calibrate:
     tmp_df['embeddings'] = [im_emb.detach().to('cpu')]
     tmp_df['user:rating'] = [{' ': ' '}]
     tmp_df['text'] = ['']
-    # tmp_df['from_user_id'] = [0]
-    # tmp_df['latest_user_to_rate'] = [0]
+    tmp_df['from_user_id'] = [0]
+    tmp_df['latest_user_to_rate'] = [0]
     prevs_df = pd.concat((prevs_df, tmp_df))
 
 glob_idx = 0