update frontend to prepare for office plot

app.py

@@ -9,19 +9,32 @@ from transformers import (
     TextClassificationPipeline,
     pipeline,
 )
+from sklearn import preprocessing
 from langdetect import detect
 from matplotlib import pyplot as plt
 import imageio
 
 # move constants into extra file
-ML_MODEL_SURE = 0.6
+DESCRIPTION = """Diese Anwendung klassifiziert Vorstöße in Departements und schlägt auch ein 
+    mögliches Office vor. Bitte bewerten Sie für sich, ob Sie dem Office-Vorschlag 
+    nachkommen wollen, oder Ihren Vorstoß in einem anderen Office sehen, und leiten Sie 
+    nach eigenem Ermessen weiter. \n\n
+    Cette application classe les requêtes dans les départements et propose également un 
+    office possible. Veuillez évaluer pour vous-même si vous souhaitez suivre la 
+    proposition d'office ou si vous souhaitez voir votre démarche dans un autre office
+    et transmettez à votre discrétion."""
+TITLE_DE = "Automatisierte Einteilung von Vorstößen in Departements & Offices"
+TITLE_FR = "Où aller ? Classification des départements & bureaux"
 UNKNOWN_LANG_TEXT = (
     "The language is not recognized, it must be either in German or in French."
 )
-PLACEHOLDER_TEXT = "Geben Sie bitte den Titel und den Sumbmitted Text des Vorstoss ein.\nVeuillez entrer le titre et le Submitted Text de la requête."
+PLACEHOLDER_TEXT = "Geben Sie bitte den Titel und den 'Submitted Text' des Vorstoss ein.\nVeuillez entrer le titre et le 'Submitted Text' de la requête."
 
 UNSURE_DE_TEXT = "Das ML-Modell ist nicht sicher. Das Departement könnte sein : \n\n"
 UNSURE_FR_TEXT = "Le modèle ML n'est pas sûr. Le département pourrait être : \n\n"
+
+ML_MODEL_SURE = 0.6
+
 BARS_DEP_FR = (
     "DDPS",
     "DFI",
@@ -60,10 +73,10 @@ def load_model(modelFolder):
     return pipe
 
 
-def translate_to_de(inputText):
+def translate_to_de(SubmittedText):
     """Translates french user input to German for the model to reach better classification."""
     translator = pipeline("translation", model="Helsinki-NLP/opus-mt-fr-de")
-    translatedText = translator(inputText[0:1000])
+    translatedText = translator(SubmittedText[0:1000])
     text = translatedText[0]["translation_text"]
     return text
 
@@ -115,35 +128,76 @@ def show_chosen_category(barnames, rates, language):
 
 
 pipeDep = load_model("saved_model_dep")
-# pipeOffice = load_model("saved_model_office")
+pipeOffice = load_model("saved_model_dep")
+
+labelencoderOffice = preprocessing.LabelEncoder()
+labelencoderOffice.classes_ = np.load("classes_office.npy")
 
 
-# Function called by the UI
-def attribution(inputText):
+def textclassification(SubmittedText):
     plt.clf()
-    language = detect(inputText)
+    language = detect(SubmittedText)
 
     # Translate the input to german if necessary
     if language == "fr":
-        inputText = translate_to_de(inputText)
+        SubmittedText = translate_to_de(SubmittedText)
     elif language != "de":
         return UNKNOWN_LANG_TEXT, None
 
     # Make the prediction with the 1000 first characters
-    prediction = pipeDep(inputText[0:1000], return_all_scores=True)
-    rates = [row["score"] for row in prediction[0]]
-
-    # Create barplot & output text
-    im, barnames = create_bar_plot(rates, language)
-    chosenCategoryText = show_chosen_category(barnames, rates, language)
-
-    return chosenCategoryText, im
-
-
-# display the UI
-interface = gr.Interface(
-    fn=attribution,
-    inputs=[gr.components.Textbox(lines=20, placeholder=PLACEHOLDER_TEXT)],
-    outputs=["text", "image"],
-)
-interface.launch()
+    images = []
+    chosenCategoryTexts = []
+    for pipe in (pipeDep, pipeOffice):
+        prediction = pipe(SubmittedText[0:1000], return_all_scores=True)
+        rates = [row["score"] for row in prediction[0]]
+
+        # Create barplot & output text
+        im, barnames = create_bar_plot(rates, language)
+        images.append(im)
+
+        chosenCategoryText = show_chosen_category(barnames, rates, language)
+        chosenCategoryTexts.append(chosenCategoryText)
+
+    # return chosenCategoryText & image for both predictions
+    return chosenCategoryTexts[0], images[0], chosenCategoryTexts[1], images[1]
+
+
+# TODO set example picture upon loading
+# TODO vielleicht ein paar Sachen zum Einstellen im Frontend?
+
+# Launch UI
+with gr.Blocks(
+    # Set theme matching BK CH
+    gr.themes.Monochrome(
+        primary_hue="red",
+        secondary_hue="red",
+        font=[gr.themes.GoogleFont("Inter"), "Arial", "sans-serif"],
+    )
+) as demo:
+    gr.Markdown(f"# {TITLE_DE}\n # {TITLE_FR}\n\n {DESCRIPTION}")
+
+    # Organize layout in three columns for input, prediction I and prediction II
+    with gr.Row():
+        with gr.Column(scale=2):
+            name = gr.Textbox(label="", lines=28, placeholder=PLACEHOLDER_TEXT)
+            predict_btn = gr.Button("Submit | Soumettre")
+        with gr.Column(scale=2):
+            output_text_dep = gr.Textbox(label="Departement prediction:")
+            output_image_dep = gr.Image(label="Departement")
+        with gr.Column(scale=2):
+            output_text_office = gr.Textbox(label="Office prediction:")
+            output_image_office = gr.Image(label="Office")
+
+    predict_btn.click(
+        fn=textclassification,
+        inputs=name,
+        outputs=[
+            output_text_dep,
+            output_image_dep,
+            output_text_office,
+            output_image_office,
+        ],
+        api_name="predict",
+    )
+
+demo.launch()

classes_office.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:91aa3c28bb43aeb228af856650169f97f6326064b2dedb4cb438d5541918a94f
+size 1480

requirements.txt

@@ -5,6 +5,7 @@ matplotlib
 imageio
 torch
 sentencepiece
+sklearn
 
 gradio
 langdetect