Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -3,7 +3,7 @@ from langdetect import detect
|
|
3 |
import gradio as gr
|
4 |
import pandas as pd
|
5 |
|
6 |
-
# Chargement
|
7 |
classifier = pipeline(
|
8 |
"sentiment-analysis",
|
9 |
model="mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis"
|
@@ -19,21 +19,36 @@ translator_to_fr = pipeline(
|
|
19 |
model="Helsinki-NLP/opus-mt-en-fr"
|
20 |
)
|
21 |
|
22 |
-
#
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
|
|
|
|
|
|
|
|
28 |
return pipeline("text2text-generation", model=model_name)
|
29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
# Fonction complète d'analyse + explication
|
31 |
-
def full_analysis(text,
|
32 |
if not text:
|
33 |
return "Entrez une phrase.", "", "", count, history, None
|
34 |
|
35 |
-
# Charger
|
36 |
-
explainer = load_explainer(
|
37 |
|
38 |
# Détection de langue
|
39 |
try:
|
@@ -49,11 +64,11 @@ def full_analysis(text, mode, count, history):
|
|
49 |
result = classifier(text)[0]
|
50 |
sentiment_output = f"Sentiment : {result['label']} (Score: {result['score']:.2f})"
|
51 |
|
52 |
-
#
|
53 |
-
prompt = f"
|
54 |
-
explanation_en = explainer(prompt, max_length=
|
55 |
|
56 |
-
# Traduction
|
57 |
explanation_fr = translator_to_fr(explanation_en, max_length=512)[0]['translation_text']
|
58 |
|
59 |
# Mise à jour historique
|
@@ -64,12 +79,12 @@ def full_analysis(text, mode, count, history):
|
|
64 |
"Score": f"{result['score']:.2f}",
|
65 |
"Explication_EN": explanation_en,
|
66 |
"Explication_FR": explanation_fr,
|
67 |
-
"
|
68 |
})
|
69 |
|
70 |
return sentiment_output, explanation_en, explanation_fr, count, history, None
|
71 |
|
72 |
-
# Fonction pour
|
73 |
def download_history(history):
|
74 |
if not history:
|
75 |
return None
|
@@ -78,23 +93,24 @@ def download_history(history):
|
|
78 |
df.to_csv(file_path, index=False)
|
79 |
return file_path
|
80 |
|
81 |
-
# Interface
|
82 |
with gr.Blocks() as iface:
|
83 |
-
gr.Markdown("# 📈 Analyse de Sentiment Financier Multilingue +
|
84 |
-
gr.Markdown("Entrez une actualité financière.
|
85 |
|
86 |
count = gr.State(0)
|
87 |
history = gr.State([])
|
88 |
|
89 |
with gr.Row():
|
90 |
input_text = gr.Textbox(lines=4, placeholder="Entrez votre actualité ici...")
|
91 |
-
|
92 |
with gr.Row():
|
93 |
-
|
94 |
-
choices=
|
95 |
-
label="Choisissez le
|
96 |
-
value="
|
97 |
)
|
|
|
98 |
|
99 |
analyze_btn = gr.Button("Analyser")
|
100 |
download_btn = gr.Button("Télécharger l'historique CSV")
|
@@ -115,13 +131,24 @@ with gr.Blocks() as iface:
|
|
115 |
|
116 |
download_file = gr.File(label="Téléchargement du CSV")
|
117 |
|
118 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
119 |
analyze_btn.click(
|
120 |
lambda: gr.update(visible=True),
|
121 |
outputs=[loading_text]
|
122 |
).then(
|
123 |
full_analysis,
|
124 |
-
inputs=[input_text,
|
125 |
outputs=[sentiment_output, explanation_output_en, explanation_output_fr, count, history, download_file]
|
126 |
).then(
|
127 |
lambda c: gr.update(value=f"{c} analyses réalisées"),
|
@@ -132,7 +159,6 @@ with gr.Blocks() as iface:
|
|
132 |
outputs=[loading_text]
|
133 |
)
|
134 |
|
135 |
-
# Clic sur "Télécharger historique"
|
136 |
download_btn.click(
|
137 |
download_history,
|
138 |
inputs=[history],
|
|
|
3 |
import gradio as gr
|
4 |
import pandas as pd
|
5 |
|
6 |
+
# Chargement modèle de sentiment
|
7 |
classifier = pipeline(
|
8 |
"sentiment-analysis",
|
9 |
model="mrm8488/distilroberta-finetuned-financial-news-sentiment-analysis"
|
|
|
19 |
model="Helsinki-NLP/opus-mt-en-fr"
|
20 |
)
|
21 |
|
22 |
+
# Modèles disponibles
|
23 |
+
MODEL_OPTIONS = {
|
24 |
+
"Flan-T5 Small (rapide)" : "google/flan-t5-small",
|
25 |
+
"Flan-T5 Base (équilibré)" : "google/flan-t5-base",
|
26 |
+
"Flan-T5 Large (précis)" : "google/flan-t5-large",
|
27 |
+
}
|
28 |
+
|
29 |
+
# Fonction pour charger un modèle d'explication
|
30 |
+
def load_explainer(model_choice):
|
31 |
+
model_name = MODEL_OPTIONS.get(model_choice, "google/flan-t5-small")
|
32 |
return pipeline("text2text-generation", model=model_name)
|
33 |
|
34 |
+
# Fonction pour suggérer le meilleur modèle
|
35 |
+
def suggest_model(text):
|
36 |
+
word_count = len(text.split())
|
37 |
+
if word_count < 50:
|
38 |
+
suggestion = "Flan-T5 Small (rapide)"
|
39 |
+
elif word_count <= 200:
|
40 |
+
suggestion = "Flan-T5 Base (équilibré)"
|
41 |
+
else:
|
42 |
+
suggestion = "Flan-T5 Large (précis)"
|
43 |
+
return suggestion
|
44 |
+
|
45 |
# Fonction complète d'analyse + explication
|
46 |
+
def full_analysis(text, model_choice, count, history):
|
47 |
if not text:
|
48 |
return "Entrez une phrase.", "", "", count, history, None
|
49 |
|
50 |
+
# Charger modèle sélectionné
|
51 |
+
explainer = load_explainer(model_choice)
|
52 |
|
53 |
# Détection de langue
|
54 |
try:
|
|
|
64 |
result = classifier(text)[0]
|
65 |
sentiment_output = f"Sentiment : {result['label']} (Score: {result['score']:.2f})"
|
66 |
|
67 |
+
# Prompt amélioré pour meilleure explication
|
68 |
+
prompt = f"""Given the following financial news: \"{text}\", explain in natural, clear and detailed language why the sentiment is {result['label'].lower()}. Focus on the main financial aspects and keep it concise."""
|
69 |
+
explanation_en = explainer(prompt, max_length=150)[0]['generated_text']
|
70 |
|
71 |
+
# Traduction française
|
72 |
explanation_fr = translator_to_fr(explanation_en, max_length=512)[0]['translation_text']
|
73 |
|
74 |
# Mise à jour historique
|
|
|
79 |
"Score": f"{result['score']:.2f}",
|
80 |
"Explication_EN": explanation_en,
|
81 |
"Explication_FR": explanation_fr,
|
82 |
+
"Modèle utilisé": model_choice
|
83 |
})
|
84 |
|
85 |
return sentiment_output, explanation_en, explanation_fr, count, history, None
|
86 |
|
87 |
+
# Fonction pour générer historique CSV
|
88 |
def download_history(history):
|
89 |
if not history:
|
90 |
return None
|
|
|
93 |
df.to_csv(file_path, index=False)
|
94 |
return file_path
|
95 |
|
96 |
+
# Interface
|
97 |
with gr.Blocks() as iface:
|
98 |
+
gr.Markdown("# 📈 Analyse de Sentiment Financier Multilingue + Raisonnement 🌍")
|
99 |
+
gr.Markdown("Entrez une actualité financière. L'analyse détecte automatiquement le sentiment et génère une explication **bilingue**.\nChoisissez ou laissez l'IA vous suggérer le meilleur modèle selon la longueur de votre texte.")
|
100 |
|
101 |
count = gr.State(0)
|
102 |
history = gr.State([])
|
103 |
|
104 |
with gr.Row():
|
105 |
input_text = gr.Textbox(lines=4, placeholder="Entrez votre actualité ici...")
|
106 |
+
|
107 |
with gr.Row():
|
108 |
+
model_selector = gr.Dropdown(
|
109 |
+
choices=list(MODEL_OPTIONS.keys()),
|
110 |
+
label="Choisissez le modèle d'explication",
|
111 |
+
value="Flan-T5 Base (équilibré)"
|
112 |
)
|
113 |
+
model_suggestion = gr.Markdown("")
|
114 |
|
115 |
analyze_btn = gr.Button("Analyser")
|
116 |
download_btn = gr.Button("Télécharger l'historique CSV")
|
|
|
131 |
|
132 |
download_file = gr.File(label="Téléchargement du CSV")
|
133 |
|
134 |
+
# Suggestions dynamiques du modèle
|
135 |
+
input_text.change(
|
136 |
+
lambda text: gr.update(value=suggest_model(text)),
|
137 |
+
inputs=[input_text],
|
138 |
+
outputs=[model_selector]
|
139 |
+
).then(
|
140 |
+
lambda text: gr.update(value=f"💡 Suggestion IA : {suggest_model(text)}"),
|
141 |
+
inputs=[input_text],
|
142 |
+
outputs=[model_suggestion]
|
143 |
+
)
|
144 |
+
|
145 |
+
# Actions boutons
|
146 |
analyze_btn.click(
|
147 |
lambda: gr.update(visible=True),
|
148 |
outputs=[loading_text]
|
149 |
).then(
|
150 |
full_analysis,
|
151 |
+
inputs=[input_text, model_selector, count, history],
|
152 |
outputs=[sentiment_output, explanation_output_en, explanation_output_fr, count, history, download_file]
|
153 |
).then(
|
154 |
lambda c: gr.update(value=f"{c} analyses réalisées"),
|
|
|
159 |
outputs=[loading_text]
|
160 |
)
|
161 |
|
|
|
162 |
download_btn.click(
|
163 |
download_history,
|
164 |
inputs=[history],
|