dazzleun-7's picture
Update app.py
1b19fcd verified
# gradio final ver ----------------------------
import torch
from torch import nn
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
import gluonnlp as nlp
import numpy as np
from tqdm import tqdm, tqdm_notebook
import pandas as pd
import ast
import os
# Hugging Face๋ฅผ ํ†ตํ•œ ๋ชจ๋ธ ๋ฐ ํ† ํฌ๋‚˜์ด์ € Import
from kobert_tokenizer import KoBERTTokenizer
from transformers import BertModel
from transformers import AdamW
from transformers.optimization import get_cosine_schedule_with_warmup
n_devices = torch.cuda.device_count()
print(n_devices)
for i in range(n_devices):
print(torch.cuda.get_device_name(i))
if torch.cuda.is_available():
device = torch.device("cuda")
print('There are %d GPU(s) available.' % torch.cuda.device_count())
print('We will use the GPU:', torch.cuda.get_device_name(0))
else:
device = torch.device("cpu")
print('No GPU available, using the CPU instead.')
max_len = 64
batch_size = 32
warmup_ratio = 0.1
num_epochs = 5
max_grad_norm = 1
log_interval = 200
learning_rate = 1e-5
class BERTSentenceTransform:
r"""BERT style data transformation.
Parameters
----------
tokenizer : BERTTokenizer.
Tokenizer for the sentences.
max_seq_length : int.
Maximum sequence length of the sentences.
pad : bool, default True
Whether to pad the sentences to maximum length.
pair : bool, default True
Whether to transform sentences or sentence pairs.
"""
# ์ž…๋ ฅ์œผ๋กœ ๋ฐ›์€ tokenizerm ์ตœ๋Œ€ ์‹œํ€€์Šค ๊ธธ์ด, vocab, pad ๋ฐ pair ์„ค์ •
def __init__(self, tokenizer, max_seq_length,vocab, pad=True, pair=True):
self._tokenizer = tokenizer
self._max_seq_length = max_seq_length
self._pad = pad
self._pair = pair
self._vocab = vocab
# ์ž…๋ ฅ๋œ ๋ฌธ์žฅ ๋˜๋Š” ๋ฌธ์žฅ ์Œ์„ BERT ๋ชจ๋ธ์ด ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ํ˜•์‹์œผ๋กœ ๋ณ€ํ™˜
def __call__(self, line):
"""Perform transformation for sequence pairs or single sequences.
The transformation is processed in the following steps:
- tokenize the input sequences
- insert [CLS], [SEP] as necessary
- generate type ids to indicate whether a token belongs to the first
sequence or the second sequence.
- generate valid length
For sequence pairs, the input is a tuple of 2 strings:
text_a, text_b.
Inputs:
text_a: 'is this jacksonville ?'
text_b: 'no it is not'
Tokenization:
text_a: 'is this jack ##son ##ville ?'
text_b: 'no it is not .'
Processed:
tokens: '[CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]'
type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1
valid_length: 14
For single sequences, the input is a tuple of single string:
text_a.
Inputs:
text_a: 'the dog is hairy .'
Tokenization:
text_a: 'the dog is hairy .'
Processed:
text_a: '[CLS] the dog is hairy . [SEP]'
type_ids: 0 0 0 0 0 0 0
valid_length: 7
Parameters
----------
line: tuple of str
Input strings. For sequence pairs, the input is a tuple of 2 strings:
(text_a, text_b). For single sequences, the input is a tuple of single
string: (text_a,).
Returns
-------
np.array: input token ids in 'int32', shape (batch_size, seq_length)
np.array: valid length in 'int32', shape (batch_size,)
np.array: input token type ids in 'int32', shape (batch_size, seq_length)
"""
# convert to unicode
text_a = line[0]
if self._pair:
assert len(line) == 2
text_b = line[1]
tokens_a = self._tokenizer.tokenize(text_a)
tokens_b = None
if self._pair:
tokens_b = self._tokenizer(text_b)
if tokens_b:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
self._truncate_seq_pair(tokens_a, tokens_b,
self._max_seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > self._max_seq_length - 2:
tokens_a = tokens_a[0:(self._max_seq_length - 2)]
# The embedding vectors for `type=0` and `type=1` were learned during
# pre-training and are added to the wordpiece embedding vector
# (and position vector). This is not *strictly* necessary since
# the [SEP] token unambiguously separates the sequences, but it makes
# it easier for the model to learn the concept of sequences.
# For classification tasks, the first vector (corresponding to [CLS]) is
# used as as the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
#vocab = self._tokenizer.vocab
vocab = self._vocab
tokens = []
tokens.append(vocab.cls_token)
tokens.extend(tokens_a)
tokens.append(vocab.sep_token)
segment_ids = [0] * len(tokens)
if tokens_b:
tokens.extend(tokens_b)
tokens.append(vocab.sep_token)
segment_ids.extend([1] * (len(tokens) - len(segment_ids)))
input_ids = self._tokenizer.convert_tokens_to_ids(tokens)
# The valid length of sentences. Only real tokens are attended to.
valid_length = len(input_ids)
if self._pad:
# Zero-pad up to the sequence length.
padding_length = self._max_seq_length - valid_length
# use padding tokens for the rest
input_ids.extend([vocab[vocab.padding_token]] * padding_length)
segment_ids.extend([0] * padding_length)
return np.array(input_ids, dtype='int32'), np.array(valid_length, dtype='int32'),\
np.array(segment_ids, dtype='int32')
class BERTDataset(Dataset):
def __init__(self, dataset, sent_idx, label_idx, bert_tokenizer, vocab, max_len,
pad, pair):
transform = BERTSentenceTransform(bert_tokenizer, max_seq_length=max_len,vocab=vocab, pad=pad, pair=pair)
#transform = nlp.data.BERTSentenceTransform(
# tokenizer, max_seq_length=max_len, pad=pad, pair=pair)
self.sentences = [transform([i[sent_idx]]) for i in dataset]
self.labels = [np.int32(i[label_idx]) for i in dataset]
def __getitem__(self, i):
return (self.sentences[i] + (self.labels[i], ))
def __len__(self):
return (len(self.labels))
tokenizer = KoBERTTokenizer.from_pretrained('skt/kobert-base-v1')
bertmodel = BertModel.from_pretrained('skt/kobert-base-v1', return_dict=False)
vocab = nlp.vocab.BERTVocab.from_sentencepiece(tokenizer.vocab_file, padding_token='[PAD]')
# Kobert_softmax
class BERTClassifier(nn.Module):
def __init__(self,
bert,
hidden_size=768,
num_classes=6,
dr_rate=None,
params=None):
super(BERTClassifier, self).__init__()
self.bert = bert
self.dr_rate = dr_rate
self.softmax = nn.Softmax(dim=1) # Softmax๋กœ ๋ณ€๊ฒฝ
self.classifier = nn.Sequential(
nn.Dropout(p=0.5),
nn.Linear(in_features=hidden_size, out_features=512),
nn.Linear(in_features=512, out_features=num_classes),
)
# ์ •๊ทœํ™” ๋ ˆ์ด์–ด ์ถ”๊ฐ€ (Layer Normalization)
self.layer_norm = nn.LayerNorm(768)
# ๋“œ๋กญ์•„์›ƒ
self.dropout = nn.Dropout(p=dr_rate)
def gen_attention_mask(self, token_ids, valid_length):
attention_mask = torch.zeros_like(token_ids)
for i, v in enumerate(valid_length):
attention_mask[i][:v] = 1
return attention_mask.float()
def forward(self, token_ids, valid_length, segment_ids):
attention_mask = self.gen_attention_mask(token_ids, valid_length)
_, pooler = self.bert(input_ids=token_ids, token_type_ids=segment_ids.long(), attention_mask=attention_mask.float().to(token_ids.device))
pooled_output = self.dropout(pooler)
normalized_output = self.layer_norm(pooled_output)
out = self.classifier(normalized_output)
# LayerNorm ์ ์šฉ
pooler = self.layer_norm(pooler)
if self.dr_rate:
pooler = self.dropout(pooler)
logits = self.classifier(pooler) # ๋ถ„๋ฅ˜๋ฅผ ์œ„ํ•œ ๋กœ์ง“ ๊ฐ’ ๊ณ„์‚ฐ
probabilities = self.softmax(logits) # Softmax๋กœ ๊ฐ ํด๋ž˜์Šค์˜ ํ™•๋ฅ  ๊ณ„์‚ฐ
return probabilities # ๊ฐ ํด๋ž˜์Šค์— ๋Œ€ํ•œ ํ™•๋ฅ  ๋ฐ˜ํ™˜
model = torch.load('./model_weights_softmax(model).pth', map_location=torch.device('cpu'))
model.eval()
# ๋ฉœ๋ก  ๋ฐ์ดํ„ฐ ๋ถˆ๋Ÿฌ์˜ค๊ธฐ
melon_data = pd.read_csv('./melon_data.csv')
melon_emotions = pd.read_csv('./melon_emotions_final.csv')
melon_emotions = pd.merge(melon_emotions, melon_data, left_on='Title', right_on='title', how='inner')
melon_emotions = melon_emotions[['singer', 'Title', 'genre','Emotions']]
melon_emotions = melon_emotions.drop_duplicates(subset='Title', keep='first')
melon_emotions['Emotions'] = melon_emotions['Emotions'].apply(lambda x: ast.literal_eval(x))
emotions = melon_emotions['Emotions'].to_list()
#gradio
import numpy as np
import pandas as pd
import requests
from PIL import Image
import torch
from transformers import AutoProcessor, AutoModelForZeroShotImageClassification, pipeline
import gradio as gr
import openai
from sklearn.metrics.pairwise import cosine_similarity
import ast
###### ๊ธฐ๋ณธ ์„ค์ • ######
# OpenAI API ํ‚ค ์„ค์ •
api_key = os.getenv("OPEN_AI_KEY")
openai.api_key = api_key
if openai.api_key:
print("Private Key:", openai.api_key)
else:
print("Private Key not set.")
# ๋ชจ๋ธ ๋ฐ ํ”„๋กœ์„ธ์„œ ๋กœ๋“œ
processor = AutoProcessor.from_pretrained("openai/clip-vit-large-patch14")
model_clip = AutoModelForZeroShotImageClassification.from_pretrained("openai/clip-vit-large-patch14")
tokenizer = KoBERTTokenizer.from_pretrained('skt/kobert-base-v1')
# ์˜ˆ์ธก ๋ ˆ์ด๋ธ”
labels = ['a photo of a happy face', 'a photo of a joyful face', 'a photo of a loving face',
'a photo of an angry face', 'a photo of a melancholic face', 'a photo of a lonely face']
###### ์–ผ๊ตด ๊ฐ์ • ๋ฒกํ„ฐ ์˜ˆ์ธก ํ•จ์ˆ˜ ######
def predict_face_emotion(image):
# ์ด๋ฏธ์ง€๊ฐ€ None์ด๊ฑฐ๋‚˜ ์ž˜๋ชป๋œ ๊ฒฝ์šฐ
if image is None:
return np.zeros(len(labels)) # ๋นˆ ๋ฒกํ„ฐ ๋ฐ˜ํ™˜
# PIL ์ด๋ฏธ์ง€๋ฅผ RGB๋กœ ๋ณ€ํ™˜
image = image.convert("RGB")
# CLIP ๋ชจ๋ธ์˜ processor๋ฅผ ์ด์šฉํ•œ ์ „์ฒ˜๋ฆฌ
inputs = processor(text=labels, images=image, return_tensors="pt", padding=True)
# pixel_values๊ฐ€ 4์ฐจ์›์ธ์ง€ ํ™•์ธ ํ›„ ๊ฐ•์ œ ๋ณ€ํ™˜
pixel_values = inputs["pixel_values"] # (batch_size, channels, height, width)
# CLIP ๋ชจ๋ธ ์˜ˆ์ธก: forward์— ์˜ฌ๋ฐ”๋ฅธ ์ž…๋ ฅ ์ „๋‹ฌ
with torch.no_grad():
outputs = model_clip(pixel_values=pixel_values, input_ids=inputs["input_ids"])
# ํ™•๋ฅ ๊ฐ’ ๊ณ„์‚ฐ
probs = outputs.logits_per_image.softmax(dim=1)[0]
return probs.numpy()
###### ํ…์ŠคํŠธ ๊ฐ์ • ๋ฒกํ„ฐ ์˜ˆ์ธก ํ•จ์ˆ˜ ######
sentence_emotions = []
def predict_text_emotion(predict_sentence):
if not isinstance(predict_sentence, str):
predict_sentence = str(predict_sentence)
data = [predict_sentence, '0']
dataset_another = [data]
another_test = BERTDataset(dataset_another, 0, 1, tokenizer, vocab, max_len, True, False)
test_dataloader = torch.utils.data.DataLoader(another_test, batch_size=1, num_workers=5)
model.eval()
for batch_id, (token_ids, valid_length, segment_ids, label) in enumerate(test_dataloader):
token_ids = token_ids.long().to(device)
segment_ids = segment_ids.long().to(device)
out = model(token_ids, valid_length, segment_ids)
for i in out:
logits = i.detach().cpu().numpy()
emotions = [value.item() for value in i]
sentence_emotions.append(emotions)
return sentence_emotions[0] # ์ตœ์ข… ๋ฆฌ์ŠคํŠธ ๋ฐ˜ํ™˜
###### ์ตœ์ข… ๊ฐ์ • ๋ฒกํ„ฐ ๊ณ„์‚ฐ ######
def generate_final_emotion_vector(diary_input, image_input):
# ํ…์ŠคํŠธ ๊ฐ์ • ๋ฒกํ„ฐ ์˜ˆ์ธก
text_vector = predict_text_emotion(diary_input)
# ์–ผ๊ตด ๊ฐ์ • ๋ฒกํ„ฐ ์˜ˆ์ธก
image_vector = predict_face_emotion(image_input)
text_vector = np.array(text_vector, dtype=float)
image_vector = np.array(image_vector, dtype=float)
print(text_vector)
print(image_vector)
# ์ตœ์ข… ๊ฐ์ • ๋ฒกํ„ฐ ๊ฐ€์ค‘์น˜ ์ ์šฉ
return (text_vector * 0.7) + (image_vector * 0.3)
####### ์ฝ”์‚ฌ์ธ ์œ ์‚ฌ๋„ ํ•จ์ˆ˜ ######
def cosine_similarity_fn(vec1, vec2):
dot_product = np.dot(vec1, vec2)
norm_vec1 = np.linalg.norm(vec1)
norm_vec2 = np.linalg.norm(vec2)
if norm_vec1 == 0 or norm_vec2 == 0:
return np.nan # ์ œ๋กœ ๋ฒกํ„ฐ์ธ ๊ฒฝ์šฐ NaN ๋ฐ˜ํ™˜
return dot_product / (norm_vec1 * norm_vec2)
####### ์ด๋ฏธ์ง€ ๋‹ค์šด๋กœ๋“œ ํ•จ์ˆ˜ (PIL ๊ฐ์ฒด ๋ฐ˜ํ™˜) ######
def download_image(image_url):
try:
response = requests.get(image_url)
response.raise_for_status()
return Image.open(requests.get(image_url, stream=True).raw)
except Exception as e:
print(f"์ด๋ฏธ์ง€ ๋‹ค์šด๋กœ๋“œ ์˜ค๋ฅ˜: {e}")
return None
# ์Šคํƒ€์ผ ์˜ต์…˜
options = {
1: "๐ŸŒผ ์นœ๊ทผํ•œ",
2: "๐Ÿ”ฅ ํŠธ๋ Œ๋””ํ•œ MZ์„ธ๋Œ€",
3: "๐Ÿ˜„ ์œ ๋จธ๋Ÿฌ์Šคํ•œ ์žฅ๋‚œ๊พธ๋Ÿฌ๊ธฐ",
4: "๐Ÿง˜ ์ฐจ๋ถ„ํ•œ ๋ช…์ƒ๊ฐ€",
5: "๐ŸŽจ ์ฐฝ์˜์ ์ธ ์˜ˆ์ˆ ๊ฐ€",
}
# ์ผ๊ธฐ ๋ถ„์„ ํ•จ์ˆ˜
def chatbot_diary_with_image(style_option, diary_input, image_input, playlist_input):
style = options.get(int(style_option.split('.')[0]), "๐ŸŒผ ์นœ๊ทผํ•œ")
# GPT ์‘๋‹ต (์ผ๊ธฐ ์ฝ”๋ฉ˜ํŠธ)
try:
response_comment = openai.ChatCompletion.create(
model="gpt-4-turbo",
messages=[{"role": "system", "content": f"๋„ˆ๋Š” {style} ์ฑ—๋ด‡์ด์•ผ."}, {"role": "user", "content": diary_input}],
)
comment = response_comment.choices[0].message.content
except Exception as e:
comment = f"๐Ÿ’ฌ ์˜ค๋ฅ˜: {e}"
# GPT ๊ธฐ๋ฐ˜ ์ผ๊ธฐ ์ฃผ์ œ ์ถ”์ฒœ
try:
topics = get_initial_response(style_option, diary_input)
except Exception as e:
topics = f"๐Ÿ“ ์ฃผ์ œ ์ถ”์ฒœ ์˜ค๋ฅ˜: {e}"
# DALLยทE 3 ์ด๋ฏธ์ง€ ์ƒ์„ฑ ์š”์ฒญ (3D ์Šคํƒ€์ผ ์บ๋ฆญํ„ฐ)
try:
response = openai.Image.create(
model="dall-e-3",
prompt=(
f"{diary_input}๋ฅผ ๋ฐ˜์˜ํ•ด์„œ ๊ฐ์ •์„ ํ‘œํ˜„ํ•˜๋Š” 3D ์Šคํƒ€์ผ์˜ ์ผ๋Ÿฌ์ŠคํŠธ ์บ๋ฆญํ„ฐ๋ฅผ ๊ทธ๋ ค์ค˜. "
"์บ๋ฆญํ„ฐ๋Š” ๋ถ€๋“œ๋Ÿฝ๊ณ  ๋‘ฅ๊ทผ ๋””์ž์ธ์— ํ‘œ์ •์ด ๊ฐ์ •์„ ์ž˜ ๋“œ๋Ÿฌ๋‚ด์•ผ ํ•ด. "
"๊ฐ์ •์„ ์‹œ๊ฐ์ ์œผ๋กœ ํ‘œํ˜„ํ•  ์ˆ˜ ์žˆ๋Š” ์†Œํ’ˆ์ด๋‚˜ ์ž‘์€ ์ƒ์ง•์„ ํฌํ•จํ•ด์ค˜. "
"๊ฐ์ •์˜ ๋ถ„์œ„๊ธฐ๋ฅผ ๋ฐ˜์˜ํ•˜๋Š” ์„ ๋ช…ํ•˜๊ณ  ๊นจ๋—ํ•œ ์ƒ‰์ƒ์„ ์‚ฌ์šฉํ•˜๊ณ , ์บ๋ฆญํ„ฐ๊ฐ€ ์—ญ๋™์ ์ด๊ณ  ์žฌ๋ฏธ์žˆ๋Š” ์ž์„ธ๋ฅผ ์ทจํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ด์ค˜. "
"์ด๋ฏธ์ง€์—๋Š” ํ•˜๋‚˜์˜ ์บ๋ฆญํ„ฐ๋งŒ ๋‚˜์˜ค๊ฒŒ ํ•ด์ค˜."
"๋ฐฐ๊ฒฝ์€ ๋‹จ์ˆœํ•˜๊ณ  ๋ฐ์€ ์ƒ‰์ƒ์œผ๋กœ ์„ค์ •ํ•ด์„œ ์บ๋ฆญํ„ฐ๊ฐ€ ๊ฐ•์กฐ๋  ์ˆ˜ ์žˆ๋„๋ก ํ•ด์ค˜."
),
size="1024x1024",
n=1
)
# URL ๊ฐ€์ ธ์˜ค๊ธฐ ๋ฐ ๋‹ค์šด๋กœ๋“œ
image_url = response['data'][0]['url']
print(f"Generated Image URL: {image_url}") # URL ํ™•์ธ
image = download_image(image_url)
except Exception as e:
print(f"์ด๋ฏธ์ง€ ์ƒ์„ฑ ์˜ค๋ฅ˜: {e}") # ์˜ค๋ฅ˜ ์ƒ์„ธ ์ถœ๋ ฅ
image = None
# ์‚ฌ์šฉ์ž ์ตœ์ข… ๊ฐ์ • ๋ฒกํ„ฐ
final_user_emotions = generate_final_emotion_vector(diary_input,image_input)
# ๊ฐ ๋…ธ๋ž˜์— ๋Œ€ํ•œ ์ฝ”์‚ฌ์ธ ์œ ์‚ฌ๋„ ๊ณ„์‚ฐ
similarities = [cosine_similarity_fn(final_user_emotions, song_vec) for song_vec in emotions]
#์œ ํšจํ•œ ์œ ์‚ฌ๋„ ํ•„ํ„ฐ๋ง
valid_indices = [i for i, sim in enumerate(similarities) if not np.isnan(sim)]
filtered_similarities = [similarities[i] for i in valid_indices]
recommendations = np.argsort(filtered_similarities)[::-1] # ๋†’์€ ์œ ์‚ฌ๋„ ์ˆœ์œผ๋กœ ์ •๋ ฌ
results_df = pd.DataFrame({
'Singer' : melon_emotions['singer'].iloc[recommendations].values,
'title' : melon_emotions['Title'].iloc[recommendations].values,
'genre' : melon_emotions['genre'].iloc[recommendations].values,
'Cosine Similarity': [similarities[idx] for idx in recommendations]
})
# ๊ฐ€์ค‘์น˜ ๊ฐ’ ์„ค์ •
gamma = 0.3
similar_playlists = results_df.head(5)
similar_playlists = pd.merge(similar_playlists, melon_emotions, left_on="title", right_on="Title", how="inner")
similar_playlists = similar_playlists[["title", "Emotions", "singer"]]
dissimilar_playlists = results_df.tail(5)
dissimilar_playlists = pd.merge(dissimilar_playlists, melon_emotions, left_on="title", right_on="Title", how="inner")
dissimilar_playlists = dissimilar_playlists[["title", "Emotions", "singer"]]
#๊ฐ์ •๊ณผ ์œ ์‚ฌํ•œ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ
if playlist_input == '๋น„์Šทํ•œ':
results = []
seen_songs = set(similar_playlists["title"].values) # ์ดˆ๊ธฐ seen_songs์— similar_playlists์˜ ๊ณก๋“ค์„ ์ถ”๊ฐ€
# ์‚ฌ์šฉ์ž ๊ฐ์ • ๋ฒกํ„ฐ
user_emotion_vector = generate_final_emotion_vector(diary_input, image_input).reshape(1, -1)
for index, row in similar_playlists.iterrows():
song_title = row["title"]
song_singer = row["singer"]
song_vector = np.array(row["Emotions"]).reshape(1, -1)
song_results = []
for i, emotion_vec in enumerate(emotions):
emotion_title = melon_emotions.iloc[i]["Title"]
emotion_singer = melon_emotions.iloc[i]["singer"]
emotion_vec = np.array(emotion_vec).reshape(1, -1)
# similar_playlists์— ์žˆ๋Š” ๊ณก๊ณผ seen_songs์— ์žˆ๋Š” ๊ณก์€ ์ œ์™ธ
if (
emotion_title != song_title and
emotion_title not in seen_songs
):
try:
# ๊ณก ๊ฐ„ ์œ ์‚ฌ๋„(Song-Song Similarity)
song_song_similarity = cosine_similarity(song_vector, emotion_vec)[0][0]
# ์‚ฌ์šฉ์ž ๊ฐ์ • ๋ฒกํ„ฐ์™€์˜ ์œ ์‚ฌ๋„(User-Song Similarity)
user_song_similarity = cosine_similarity(user_emotion_vector, emotion_vec)[0][0]
# Final Score ๊ณ„์‚ฐ
final_score = gamma * song_song_similarity + (1 - gamma) * user_song_similarity
song_results.append({
"Title": emotion_title,
"Singer": emotion_singer,
"Song-Song Similarity": song_song_similarity,
"User-Song Similarity": user_song_similarity,
"Final Score": final_score
})
except ValueError as e:
print(f"Error with {song_title} vs {emotion_title}: {e}")
continue
# Final Score๋ฅผ ๊ธฐ์ค€์œผ๋กœ ์ƒ์œ„ 3๊ณก ์„ ํƒ
song_results = sorted(song_results, key=lambda x: x["Final Score"], reverse=True)[:3]
seen_songs.update([entry["Title"] for entry in song_results])
results.append({"Song Title": song_title, "Singer": song_singer, "Top 3 Similarities": song_results})
# ๊ฒฐ๊ณผ ์ถœ๋ ฅ
for result in results:
print(f"{result['Singer']} - {result['Song Title']}")
for entry in result["Top 3 Similarities"]:
print(f"{entry['Singer']} - {entry['Title']} : Final Score {entry['Final Score']:.4f}")
print(f" (Song-Song Similarity: {entry['Song-Song Similarity']:.4f}, User-Song Similarity: {entry['User-Song Similarity']:.4f})")
print("-" * 30)
#๋ฐ˜๋Œ€ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ
if playlist_input == '์ƒ๋ฐ˜๋œ':
results = []
seen_songs = set()
# ์‚ฌ์šฉ์ž ๊ฐ์ • ๋ฒกํ„ฐ
user_emotion_vector = generate_final_emotion_vector(diary_input, image_input).reshape(1, -1)
for index, row in dissimilar_playlists.iterrows():
song_title = row["title"]
song_singer = row["singer"]
song_vector = np.array(row["Emotions"]).reshape(1, -1)
song_results = []
for i, emotion_vec in enumerate(emotions):
emotion_title = melon_emotions.iloc[i]["Title"]
emotion_singer = melon_emotions.iloc[i]["singer"]
emotion_vec = np.array(emotion_vec).reshape(1, -1)
if (
emotion_title != song_title and
emotion_title not in dissimilar_playlists["title"].values and
emotion_title not in seen_songs
):
try:
# ๊ณก ๊ฐ„ ์œ ์‚ฌ๋„(Song-Song Similarity)
song_song_similarity = cosine_similarity(song_vector, emotion_vec)[0][0]
# ์‚ฌ์šฉ์ž ๊ฐ์ • ๋ฒกํ„ฐ์™€์˜ ๋ฐ˜๋Œ€ ์œ ์‚ฌ๋„(User-Song Dissimilarity)
opposite_user_song_similarity = 1 - cosine_similarity(user_emotion_vector, emotion_vec)[0][0]
# Final Score ๊ณ„์‚ฐ
final_score = gamma * song_song_similarity + (1 - gamma) * opposite_user_song_similarity
song_results.append({
"Title": emotion_title,
"Singer": emotion_singer,
"Song-Song Similarity": song_song_similarity,
"User-Song Dissimilarity": opposite_user_song_similarity,
"Final Score": final_score
})
except ValueError as e:
print(f"Error with {song_title} vs {emotion_title}: {e}")
continue
# Final Score๋ฅผ ๊ธฐ์ค€์œผ๋กœ ์ƒ์œ„ 3๊ณก ์„ ํƒ (๊ฐ’์ด ํฐ ๊ณก์ด ๋ฐ˜๋Œ€๋˜๋Š” ๊ณก)
song_results = sorted(song_results, key=lambda x: x["Final Score"], reverse=True)[:3]
seen_songs.update(entry["Title"] for entry in song_results)
results.append({"Song Title": song_title, "Singer": song_singer, "Top 3 Similarities": song_results})
# ๊ฒฐ๊ณผ ์ถœ๋ ฅ
for result in results:
print(f"{result['Singer']} - {result['Song Title']}")
for entry in result["Top 3 Similarities"]:
print(f"{entry['Singer']} - {entry['Title']} : Final Score {entry['Final Score']:.4f}")
print(f' (Song-Song Similarity: {entry["Song-Song Similarity"]:.4f}, User-Song Dissimilarity: {entry["User-Song Dissimilarity"]:.4f})')
print("-" * 30)
# ๋ฐ์ดํ„ฐํ”„๋ ˆ์ž„ ๋ณ€ํ™˜์„ ์œ„ํ•œ ๋ฆฌ์ŠคํŠธ ์ƒ์„ฑ
df_rows = []
for result in results:
song_title = result['Song Title']
song_singer = result['Singer']
main_song_info = f"{song_singer} - {song_title}"
for entry in result["Top 3 Similarities"]:
combined_info = f"{entry['Singer']} - {entry['Title']}"
df_rows.append({"1st ์ถ”์ฒœ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ": main_song_info, "2nd ์ถ”์ฒœ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ": combined_info})
# ๋ฐ์ดํ„ฐํ”„๋ ˆ์ž„ ์ƒ์„ฑ
final_music_playlist_recommendation = pd.DataFrame(df_rows)
# ๊ณก ์ œ๋ชฉ ๊ทธ๋ฃนํ™”ํ•˜์—ฌ ์ฒซ ๋ฒˆ์งธ ํ–‰์—๋งŒ ๊ณก ์ œ๋ชฉ ํ‘œ์‹œ
final_music_playlist_recommendation["1st ์ถ”์ฒœ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ"] = final_music_playlist_recommendation.groupby("1st ์ถ”์ฒœ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ")["1st ์ถ”์ฒœ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ"].transform(
lambda x: [x.iloc[0]] + [""] * (len(x) - 1)
)
return final_music_playlist_recommendation, comment, topics, image
# ์ผ๊ธฐ ์ฃผ์ œ ์ถ”์ฒœ ํ•จ์ˆ˜
def get_initial_response(style, sentence):
style = options.get(int(style.split('.')[0]), "๐ŸŒผ ์นœ๊ทผํ•œ")
system_prompt_momentum = (
f"๋„ˆ๋Š” {style}์˜ ์ฑ—๋ด‡์ด์•ผ. ์‚ฌ์šฉ์ž๊ฐ€ ์ž‘์„ฑํ•œ ์ผ๊ธฐ๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ์ƒ๊ฐ์„ ์ •๋ฆฌํ•˜๊ณ  ๋‚ด๋ฉด์„ ๋Œ์•„๋ณผ ์ˆ˜ ์žˆ๋„๋ก "
"๋„์™€์ฃผ๋Š” ๊ตฌ์ฒด์ ์ธ ์ผ๊ธฐ ์ฝ˜ํ…์ธ ๋‚˜ ์งˆ๋ฌธ 4-5๊ฐœ๋ฅผ ์ถ”์ฒœํ•ด์ค˜."
)
try:
response = openai.ChatCompletion.create(
model="gpt-4-turbo",
messages=[
{"role": "system", "content": system_prompt_momentum},
{"role": "user", "content": sentence}
],
temperature=1
)
return response.choices[0].message.content
except Exception as e:
return f"๐Ÿ“ ์ฃผ์ œ ์ถ”์ฒœ ์˜ค๋ฅ˜: {e}"
# Gradio ์ธํ„ฐํŽ˜์ด์Šค
with gr.Blocks() as app:
gr.Markdown("""
# ๐Ÿ“š EmoDiary : ์Šค๋งˆํŠธ ๊ฐ์ • ์ผ๊ธฐ ๋„์šฐ๋ฏธ ๐Ÿ“š
**์˜ค๋Š˜์˜ ํ•˜๋ฃจ๋ฅผ ๊ธฐ๋กํ•˜๋ฉด, ๊ทธ์— ๋งž๋Š” ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ์™€ ์ผ๊ธฐ ํšŒ๊ณ  ์ฝ˜ํ…์ธ ๋ฅผ ์ž๋™์œผ๋กœ ์ƒ์„ฑํ•ด๋“œ๋ฆฝ๋‹ˆ๋‹ค!**
### ์‚ฌ์šฉ ๋ฐฉ๋ฒ•:
1. ์˜ค๋Š˜์˜ ํ•˜๋ฃจ ๊ธฐ๋กํ•˜๊ธฐ: ํ•˜๋ฃจ ๋™์•ˆ์˜ ๊ฐ์ •์ด๋‚˜ ์ผ์–ด๋‚œ ์ผ์„ ํ…์ŠคํŠธ ๋ฐ•์Šค์— ๊ธฐ๋กํ•˜๊ณ , ์ž์‹ ์˜ ๊ฐ์ •์„ ๋ฐ˜์˜ํ•  ์ˆ˜ ์žˆ๋Š” ์–ผ๊ตด ํ‘œ์ • ์ด๋ฏธ์ง€๋ฅผ ์ดฌ์˜ํ•ด์ฃผ์„ธ์š”.
2. AI ์Šคํƒ€์ผ ์„ ํƒํ•˜๊ธฐ: ์ž์‹ ์˜ ์„ ํ˜ธ์— ๋งž๋Š” ์Šคํƒ€์ผ์„ ์„ ํƒํ•˜๋ฉด, ํ•ด๋‹น ์Šคํƒ€์ผ์— ๋งž์ถ˜ ๋‹ต์žฅ๊ณผ ํšŒ๊ณ  ์ฃผ์ œ๋ฅผ ์ถ”์ฒœํ•ด๋“œ๋ฆด๊ฒŒ์š”.
3. ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ ์„ ํƒํ•˜๊ธฐ: ์˜ค๋Š˜์˜ ๊ฐ์ •์— ๋งž์ถฐ "๋น„์Šทํ•œ" ๋˜๋Š” "์ƒ๋ฐ˜๋œ" ๊ฐ์ •์„ ์„ ํƒํ•˜๋ฉด, ๊ทธ์— ๋งž๋Š” ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ๋ฅผ ์ถ”์ฒœํ•ด๋“œ๋ฆด๊ฒŒ์š”.
4. ๋ถ„์„ ์‹œ์ž‘: "๐Ÿš€ ๋ถ„์„ ์‹œ์ž‘" ๋ฒ„ํŠผ์„ ํด๋ฆญํ•˜๋ฉด, ์ž…๋ ฅํ•œ ์ •๋ณด์™€ ์ด๋ฏธ์ง€๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ์ถ”์ฒœ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ, ๊ฐ์ • ์บ๋ฆญํ„ฐ ์ด๋ฏธ์ง€, ๊ทธ๋ฆฌ๊ณ  ํšŒ๊ณ  ์ฃผ์ œ๊ฐ€ ์ƒ์„ฑ๋ฉ๋‹ˆ๋‹ค.""")
with gr.Row():
with gr.Column():
diary_input = gr.Textbox(label="๐Ÿ“œ ์˜ค๋Š˜ ํ•˜๋ฃจ ๊ธฐ๋กํ•˜๊ธฐ", placeholder="ex)์˜ค๋Š˜ ์†Œํ’๊ฐ€์„œ ๋ง›์žˆ๋Š” ๊ฑธ ๋งŽ์ด ๋จน์–ด์„œ ์—„์ฒญ ์‹ ๋‚ฌ์–ด")
chatbot_style = gr.Radio(
choices=[f"{k}. {v}" for k, v in options.items()],
label="๐Ÿค– ์–ด๋–ค ์Šคํƒ€์ผ์˜ AI์—๊ฒŒ ๋‹ต์žฅ๊ณผ ํšŒ๊ณ  ์ฃผ์ œ๋ฅผ ์ถ”์ฒœ๋ฐ›๊ณ  ์‹ถ๋‚˜์š”?"
)
playlist_input = gr.Radio(["๋น„์Šทํ•œ", "์ƒ๋ฐ˜๋œ"], label="๐ŸŽง ์˜ค๋Š˜์˜ ๊ฐ์ •๊ณผ ใ…‡ใ…‡๋˜๋Š” ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ๋ฅผ ์ถ”์ฒœ๋ฐ›๊ณ  ์‹ถ๋‚˜์š”?")
image_input = gr.Image(type="pil", label="๐Ÿ“ท ์–ผ๊ตด ํ‘œ์ • ์‚ฌ์ง„ ์—…๋กœ๋“œ", width=256, height=256)
submit_btn = gr.Button("๐Ÿš€ ๋ถ„์„ ์‹œ์ž‘")
with gr.Column():
output_playlist = gr.Dataframe(label="๐ŸŽง ์ถ”์ฒœ ํ”Œ๋ ˆ์ด๋ฆฌ์ŠคํŠธ ")
output_comment = gr.Textbox(label="๐Ÿ’ฌ AI ์ฝ”๋ฉ˜ํŠธ")
output_topics = gr.Textbox(label="๐Ÿ“ ์˜ค๋Š˜์˜ ๋‚ด๋ฉด์„ ๋Œ์•„๋ณด๋Š” ํšŒ๊ณ  ์ถ”์ฒœ ์ฃผ์ œ")
output_image = gr.Image(label="๐Ÿ–ผ๏ธ ์ƒ์„ฑ๋œ ์˜ค๋Š˜์˜ ๊ฐ์ • ์บ๋ฆญํ„ฐ", type="pil", width=256, height=256)
# ๋ฒ„ํŠผ ํด๋ฆญ ์ด๋ฒคํŠธ ์—ฐ๊ฒฐ
submit_btn.click(
fn=chatbot_diary_with_image,
inputs=[chatbot_style, diary_input, image_input, playlist_input],
outputs=[output_playlist, output_comment, output_topics, output_image]
)
# ์•ฑ ์‹คํ–‰
app.launch(debug=True)