updated module import

.gitignore

@@ -0,0 +1 @@
+*embeddings.pkl

app.py

@@ -1,14 +1,13 @@
 import pandas as pd
 import numpy as np
 import os
-from copy import deepcopy
-import pickle
+from prefixspan import PrefixSpan
 import gradio as gr
 import altair as alt
 alt.data_transformers.enable("vegafusion")
-from dynabench.task_evaluator import *
+# from dynabench.task_evaluator import *
 
-BASE_DIR = "../db"
+BASE_DIR = "db"
 MODELS = ['qwenvl-chat', 'qwenvl', 'llava15-7b', 'llava15-13b', 'instructblip-vicuna13b', 'instructblip-vicuna7b']
 VIDEO_MODELS = ['video-chat2-7b','video-llama2-7b','video-llama2-13b','chat-univi-7b','chat-univi-13b','video-llava-7b','video-chatgpt-7b']
 domains = ["imageqa-2d-sticker", "imageqa-3d-tabletop", "imageqa-scene-graph", "videoqa-3d-tabletop", "videoqa-scene-graph"]
@@ -19,6 +18,28 @@ domain2folder = {"imageqa-2d-sticker": "2d",
                  "videoqa-scene-graph": "video-sg", 
                  None: '2d'}
 
+def find_frequent_patterns(k, df, scores=None):
+	if len(df) == 0:
+		return []
+
+	df = df.reset_index(drop=True)
+	cols = df.columns.to_list()
+	df = df.fillna('').astype('str')
+	db = [[(c, v) for c, v in zip(cols, d) if v] for d in df.values.tolist()]
+
+	ps = PrefixSpan(db)
+	patterns = ps.topk(k, closed=True)
+	if scores is None:
+		return patterns
+	else:
+		aggregated_scores = []
+		scores = np.asarray(scores)
+		for count, pattern in patterns:
+			q = ' and '.join([f"`{k}` == {repr(v)}" for k, v in pattern])
+			indices = df.query(q).index.to_numpy()
+			aggregated_scores.append(np.mean(scores[indices]))
+		return patterns, aggregated_scores
+
 def update_partition_and_models(domain):
     domain = domain2folder[domain]
     path = f"{BASE_DIR}/{domain}"