File size: 8,415 Bytes
acb3380 a9c1d92 acb3380 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 |
import concurrent.futures
from collections import defaultdict
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
from ctm.configs.ctm_config_base import (
BaseConsciousnessTuringMachineConfig,
)
from ctm.processors.processor_base import BaseProcessor
from ctm.supervisors.supervisor_base import BaseSupervisor
class BaseConsciousnessTuringMachine(object):
def __call__(self, *args, **kwargs):
return self.forward(*args, **kwargs)
def __init__(self, ctm_name=None, *args, **kwargs):
super().__init__(*args, **kwargs)
if ctm_name:
self.config = BaseConsciousnessTuringMachineConfig.from_ctm(
ctm_name
)
else:
self.config = BaseConsciousnessTuringMachineConfig()
self.processor_list = []
self.processor_group_map = defaultdict(list)
self.load_ctm()
def add_processor(self, processor_name, group_name=None):
processor_instance = BaseProcessor(processor_name)
self.processor_list.append(
{
"processor_name": processor_name,
"processor_instance": processor_instance,
}
)
if group_name:
self.processor_group_map[processor_name] = group_name
def add_supervisor(self, supervisor_name):
supervisor_instance = BaseSupervisor(supervisor_name)
self.supervisor = {
"supervisor_name": supervisor_name,
"supervisor_instance": supervisor_instance,
}
@staticmethod
def ask_processor(
processor, question, context, image_path, audio_path, video_path
):
processor_instance = processor["processor_instance"]
processor_name = processor["processor_name"]
gist, score = processor_instance.ask(
question, context, image_path, audio_path, video_path
)
return {"name": processor_name, "gist": gist, "score": score}
def ask_processors(
self, question, context, image_path, audio_path, video_path
):
with concurrent.futures.ThreadPoolExecutor() as executor:
futures = [
executor.submit(
self.ask_processor,
processor,
question,
context,
image_path,
audio_path,
video_path,
)
for processor in self.processor_list
]
results = [
future.result()
for future in concurrent.futures.as_completed(futures)
]
output = {}
for result in results:
output[result["name"]] = {
"gist": result["gist"],
"score": result["score"],
}
assert len(output) == len(self.processor_list)
return output
def uptree_competition(self, processor_output):
# Unpack processor outputs into lists for easier processing
gists, scores, names = [], [], []
for name, info in processor_output.items():
gists.append(info["gist"])
scores.append(info["score"])
names.append(name)
# Determine the unique group for each processor
unique_groups = set(self.processor_group_map.values())
# Prepare to track the best processor by group
best_processor_by_group = {
group: (None, -1) for group in unique_groups
} # (processor_name, score)
# Iterate through processors to find the best in each group
for name, score in zip(names, scores):
group = self.processor_group_map[name]
if score > best_processor_by_group[group][1]:
best_processor_by_group[group] = (name, score)
# Select the overall best across groups
best_overall = max(
best_processor_by_group.values(), key=lambda x: x[1]
)
best_name = best_overall[0]
index = names.index(best_name)
winning_info = {
"name": best_name,
"gist": gists[index],
"score": scores[index],
}
return winning_info
def ask_supervisor(self, question, processor_info):
final_answer, score = self.supervisor["supervisor_instance"].ask(
question, processor_info["gist"]
)
return final_answer, score
def downtree_broadcast(self, winning_output):
winning_processor_name = winning_output["name"]
winning_processor_gist = winning_output["gist"]
for processor in self.processor_list:
if processor["processor_name"] != winning_processor_name:
processor["processor_instance"].update_info(
winning_processor_gist
)
return
def calc_processor_sim(self, processor_output):
processor_gists = [info["gist"] for info in processor_output.values()]
tfidf_vectorizer = TfidfVectorizer()
tfidf_matrix = tfidf_vectorizer.fit_transform(processor_gists)
cosine_sim = cosine_similarity(tfidf_matrix, tfidf_matrix)
return cosine_sim
def link_form(self, processor_output):
sim = self.calc_processor_sim(processor_output)
print(sim)
# iterate on each sim pair
# if sim > threshold, then link the two processors by combining them into the same group
link_threshold = 0.5
for i in range(len(sim)):
for j in range(i + 1, len(sim)):
if sim[i][j] > 0.5:
processor1_name = list(processor_output.keys())[i]
processor2_name = list(processor_output.keys())[j]
# choose the group that includes more processors
# processor_group_map is a dict with processor_name as key and group_name as value
group1 = self.processor_group_map[processor1_name]
group2 = self.processor_group_map[processor2_name]
# calculate the number of processors in each group
group1_count = sum(
[
1
for group in self.processor_group_map.values()
if group == group1
]
)
group2_count = sum(
[
1
for group in self.processor_group_map.values()
if group == group2
]
)
# choose the group with more processors
group_name = (
group1 if group1_count > group2_count else group2
)
self.processor_group_map[processor1_name] = group_name
self.processor_group_map[processor2_name] = group_name
return
def processor_fuse(self, infos, scores):
return infos, scores
def forward(
self,
question=None,
context=None,
image_path=None,
audio_path=None,
video_path=None,
):
answer_threshold = 0.5
max_iter = 3
for i in range(max_iter):
print("start the {}-th iteration".format(i + 1))
processor_output = self.ask_processors(
question=question,
context=context,
image_path=image_path,
audio_path=audio_path,
video_path=video_path,
)
import pdb
pdb.set_trace()
winning_output = self.uptree_competition(processor_output)
answer, score = self.ask_supervisor(question, winning_output)
if score > answer_threshold:
break
else:
self.downtree_broadcast(winning_output)
self.link_form(processor_output)
return answer, score
def load_ctm(self):
for (
group_name,
processor_list,
) in self.config.groups_of_processors.items():
for processor_name in processor_list:
self.add_processor(processor_name, group_name=group_name)
self.add_supervisor(self.config.supervisor)
|