Add a data browser

README.md

@@ -4,7 +4,7 @@ emoji: 💐
 colorFrom: green
 colorTo: blue
 sdk: gradio
-sdk_version: 5.14.0
+sdk_version: 5.49.1
 app_file: app.py
 pinned: false
 license: other

app.py

@@ -11,6 +11,7 @@ import pandas as pd
 from collections import defaultdict
 
 from leaderboard import leaderboard_tab
+from data_samples import data_browse_tab
 
 DLA = """
 Note that the dataset collected by the BOUQuET initiative and your contributions to this dataset will be released under the following open source license.
@@ -243,6 +244,9 @@ with gr.Blocks(
     css="""
     #cla textarea {min-height: 60em;}
     main.app  {max-width: 90em; margin: auto;}
+    .small-font {
+        font-size: 0.8em;
+    }
     """,
     theme=gr.themes.Glass(
         font=[gr.themes.GoogleFont("Roboto"), "Arial", "sans-serif"]
@@ -256,6 +260,7 @@ with gr.Blocks(
     """)
         intro_tab()
         leaderboard_tab()
+        data_browse_tab()
         guidelines_tab()
         dla_tab()
 

data_samples.py

@@ -0,0 +1,74 @@
+
+import datasets
+import pandas as pd
+import gradio as gr
+import csv
+from collections import defaultdict
+import random
+
+INTRO = """
+The table below demonstrates a sample paragraph from the dev split of BOUQuET.
+"""
+ALL = "All"
+
+
+def data_browse_tab():
+    # Load the data
+    ds = datasets.load_dataset("facebook/bouquet", "sentence_level", split="dev")
+    long_df = ds.to_pandas()
+    lang2df = {
+        lang: part.drop(columns=["tgt_text", "tgt_lang"]).reset_index(drop=True) 
+        for lang, part in long_df.groupby('src_lang')
+    }
+    eng_df = lang2df["eng_Latn"]
+    langs = sorted(lang2df.keys())
+    domains = sorted(set(eng_df["domain"]))
+    paragraph_ids = sorted(set(eng_df["par_id"]))
+    domain2par_ids = {domain: sorted(set(group["par_id"])) for domain, group in eng_df.groupby("domain")}
+
+    def select_data(src_lang, tgt_lang, par_id):
+        src_df = lang2df[src_lang]
+        tgt_df = lang2df[tgt_lang]
+        df = src_df.copy()
+        df["tgt_text"] = tgt_df["src_text"]
+        par = df[df['par_id'].eq(par_id)].copy()
+        part = par[['domain', 'uniq_id', 'orig_text', 'src_text', 'tgt_text', 'tags', 'register']]
+        # TODO: add 'par_comment' in a text field below
+        return gr.update(value=part, wrap=True)
+
+    with gr.Tab("Data samples"):
+        gr.Markdown("# BOUQuET data browser")
+        # Define the controls
+        with gr.Row():
+            gr_src_lang = gr.Dropdown(langs, label="Source lang", value=random.choice(langs))
+            gr_tgt_lang = gr.Dropdown(langs, label="Target lang", value=random.choice(langs))
+            gr_domain = gr.Dropdown([ALL] + domains, label="Domain", value=ALL)
+            gr_par_id = gr.Dropdown(paragraph_ids, label="Paragraph ID", value=random.choice(paragraph_ids))
+            inputs = [gr_src_lang, gr_tgt_lang, gr_par_id]
+            gr_sample_btn = gr.Button(value="Sample a paragraph")
+            gr_sample_btn.click(fn=lambda: random.choice(paragraph_ids), inputs=None, outputs=gr_par_id)
+
+        # Define the data
+        df_all = select_data(*[inp.value for inp in inputs])
+        gr_df = gr.Dataframe(
+            df_all, 
+            wrap=True, 
+            show_fullscreen_button=True,
+            column_widths=["10%", "5%", "20%", "20%", "20%", "15%", "6%"],
+            elem_classes=["small-font"],
+        )
+        # Interactivity
+        for inp in inputs:
+            inp.change(fn=select_data, inputs=inputs, outputs=gr_df)
+        
+        def change_domain(domain, par_id):
+            if domain == ALL:
+                par_ids = paragraph_ids
+            else:
+                par_ids = domain2par_ids[domain]
+            if par_id not in par_ids:
+                par_id = random.choice(par_ids)
+            print(f"par_id: {par_id} is one of {par_ids}")
+            return gr.Dropdown(choices=par_ids, value=par_id)
+        
+        gr_domain.change(fn=change_domain, inputs=[gr_domain, gr_par_id], outputs=[gr_par_id])

requirements.txt

@@ -1,3 +1,4 @@
 gradio[oauth]
 pandas
 matplotlib # for background_gradient
+datasets