Update app.py

app.py

@@ -1,5 +1,7 @@
-# app.py — Thai Sentiment (WangchanBERTa Variants) GUI with Negative-Aspect Pie
-import os, json, importlib.util, traceback, sys, re, math, tempfile
+# app.py — Thai Sentiment (WangchanBERTa Variants) GUI
+# No Single tab. Tabs: Batch (textarea) + CSV Upload
+# Extended B: Negative-aspect pie + Top-N bar + Time series (POS/NEG & aspects)
+import os, json, importlib.util, traceback, re, math, tempfile
 import gradio as gr
 import torch, pandas as pd
 import torch.nn.functional as F
@@ -9,7 +11,7 @@ from safetensors.torch import load_file
 from transformers import AutoTokenizer
 
 # ================= Settings =================
-REPO_ID       = os.getenv("REPO_ID", "Dusit-P/thai-sentiment")  # รีโปโมเดลใหม่ของคุณ
+REPO_ID       = os.getenv("REPO_ID", "Dusit-P/thai-sentiment")
 DEFAULT_MODEL = os.getenv("DEFAULT_MODEL", "WCB")
 HF_TOKEN      = os.getenv("HF_TOKEN", None)
 
@@ -17,15 +19,13 @@ AVAILABLE_CHOICES = ["WCB", "WCB_BiLSTM", "WCB_CNN_BiLSTM", "WCB_4Layer_BiLSTM"]
 if DEFAULT_MODEL not in AVAILABLE_CHOICES:
     DEFAULT_MODEL = "WCB"
 
-# สีและธีม
-NEG_COLOR = "#F87171"
-POS_COLOR = "#34D399"
+NEG_COLOR = "#F87171"   # red
+POS_COLOR = "#34D399"   # green
 TEMPLATE  = "plotly_white"
 
 CACHE = {}
 
-# ================= Aspect Keywords =================
-# ปรับแก้/เพิ่มคีย์เวิร์ดได้ตามโดเมนงานของคุณ
+# ================ Aspect Keywords  ================
 ASPECT_KEYWORDS = {
     "บริการ": [
         "บริการ", "เซอร์วิส", "service", "ดูแล", "เอาใจใส่", "ใส่ใจ",
@@ -162,7 +162,6 @@ _RE_HAS_LETTER = re.compile(r"[ก-๙A-Za-z]")
 def _norm_text(v) -> str:
     if v is None: return ""
     if isinstance(v, float) and math.isnan(v): return ""
-    # ล้างเครื่องหมายที่พบบ่อยจากการพิมพ์/CSV
     return str(v).strip().strip('"').strip("'").strip(",")
 
 def _is_substantive_text(s: str, min_chars: int = 2) -> bool:
@@ -183,7 +182,6 @@ def _compile_aspect_regex():
         return _aspect_regex_cache
     cache = {}
     for aspect, kws in ASPECT_KEYWORDS.items():
-        # สร้าง regex OR จากคีย์เวิร์ด (หนีอักขระพิเศษ)
         parts = [re.escape(k) for k in kws if k.strip()]
         if not parts:
             continue
@@ -193,10 +191,6 @@ def _compile_aspect_regex():
     return cache
 
 def detect_aspect_for_negative(text: str) -> str:
-    """
-    คืนชื่อหมวดแรกที่พบจากคีย์เวิร์ด ถ้าไม่พบเลย → "อื่นๆ"
-    (เราจงใจให้เป็น single-label ต่อ 1 ข้อความเพื่ออ่าน pie ง่าย)
-    """
     text = text or ""
     regs = _compile_aspect_regex()
     for aspect, rx in regs.items():
@@ -204,39 +198,40 @@ def detect_aspect_for_negative(text: str) -> str:
             return aspect
     return ASPECT_OTHER_LABEL
 
-def make_charts_with_negative_aspects(df: pd.DataFrame):
-    """
-    df: columns = review, negative(%), positive(%), label
-    return: (bar_counts, pie_negative_aspects, info_md)
-    """
+# ================= Charts (no time) =================
+def make_basic_charts(df: pd.DataFrame):
     total = len(df)
     neg_df = df[df["label"] == "negative"].copy()
     pos_df = df[df["label"] == "positive"].copy()
 
-    # ---- bar: label counts
+    # bar counts
     fig_bar = go.Figure()
     fig_bar.add_bar(name="negative", x=["negative"], y=[len(neg_df)], marker_color=NEG_COLOR)
     fig_bar.add_bar(name="positive", x=["positive"], y=[len(pos_df)], marker_color=POS_COLOR)
     fig_bar.update_layout(barmode="group", title="Label counts", template=TEMPLATE)
 
-    # ---- pie: negative aspects only
+    # pie negative aspects
     if len(neg_df) > 0:
         neg_df["aspect"] = neg_df["review"].apply(detect_aspect_for_negative)
         counts = neg_df["aspect"].value_counts()
         labels = list(counts.index)
         values = list(counts.values)
     else:
-        labels, values = [ASPECT_OTHER_LABEL], [1]  # placeholder เพื่อแสดงกราฟว่าง
-
-    fig_pie = go.Figure(go.Pie(
-        labels=labels,
-        values=values,
-        hole=0.35,
-        sort=False
-    ))
+        labels, values = [ASPECT_OTHER_LABEL], [1]  # placeholder
+
+    fig_pie = go.Figure(go.Pie(labels=labels, values=values, hole=0.35, sort=False))
     fig_pie.update_layout(title="Negative aspect distribution", template=TEMPLATE)
 
-    # ---- info text
+    # top-N aspects bar (negative only, overall)
+    fig_topn = go.Figure()
+    if len(neg_df) > 0:
+        top_counts = neg_df["aspect"].value_counts().sort_values(ascending=True)
+        fig_topn.add_bar(y=top_counts.index.tolist(), x=top_counts.values.tolist(),
+                         orientation="h", marker_color=NEG_COLOR)
+        fig_topn.update_layout(title="Top negative aspects (overall)", template=TEMPLATE)
+    else:
+        fig_topn.update_layout(title="Top negative aspects (no negative rows)", template=TEMPLATE)
+
     neg_avg = pd.to_numeric(df["negative(%)"].str.rstrip("%"), errors="coerce").mean()
     pos_avg = pd.to_numeric(df["positive(%)"].str.rstrip("%"), errors="coerce").mean()
     info = (
@@ -245,10 +240,62 @@ def make_charts_with_negative_aspects(df: pd.DataFrame):
         f"- Negative: {len(neg_df)}  \n"
         f"- Positive: {len(pos_df)}  \n"
         f"- Avg negative: {neg_avg:.2f}%  \n"
-        f"- Avg positive: {pos_avg:.2f}%  \n"
-        f"- *Pie แสดงเฉพาะข้อความเชิงลบ จำแนกเป็นหมวดด้วยคีย์เวิร์ด*"
+        f"- Avg positive: {pos_avg:.2f}%\n"
+        f"- *Pie & Top-N = เฉพาะข้อความเชิงลบ*"
+    )
+    return fig_bar, fig_pie, fig_topn, info
+
+# ================= Charts (with time) =================
+def _to_datetime_safe(s):
+    return pd.to_datetime(s, errors="coerce", infer_datetime_format=True, utc=False)
+
+def _resample_counts(df, date_col, freq):
+    g = df.groupby([pd.Grouper(key=date_col, freq=freq), "label"]).size().unstack(fill_value=0)
+    for col in ["negative","positive"]:
+        if col not in g.columns:
+            g[col] = 0
+    return g[["negative","positive"]].sort_index()
+
+def _rolling_window(freq):
+    return 7 if freq == "D" else (4 if freq == "W" else 3)
+
+def make_time_charts(df: pd.DataFrame, date_col: str, freq: str, use_ma: bool, topn_aspects: int):
+    # 1) line pos/neg over time
+    ts = _resample_counts(df, date_col, freq)
+    if use_ma:
+        win = _rolling_window(freq)
+        ts = ts.rolling(win, min_periods=1).mean()
+
+    fig_line = go.Figure()
+    fig_line.add_scatter(x=ts.index, y=ts["negative"], mode="lines",
+                         name="negative", line=dict(color=NEG_COLOR))
+    fig_line.add_scatter(x=ts.index, y=ts["positive"], mode="lines",
+                         name="positive", line=dict(color=POS_COLOR))
+    fig_line.update_layout(title="Reviews over time", template=TEMPLATE,
+                           xaxis_title="Date", yaxis_title="Count")
+
+    # 2) stacked area of top-N negative aspects over time
+    neg_df = df[df["label"] == "negative"].copy()
+    if len(neg_df) == 0:
+        return fig_line  # ไม่มีลบก็ไม่มี area
+
+    neg_df["aspect"] = neg_df["review"].apply(detect_aspect_for_negative)
+    top_aspects = neg_df["aspect"].value_counts().head(max(1, int(topn_aspects))).index.tolist()
+
+    area_df = (
+        neg_df.assign(aspect=lambda d: d["aspect"].where(d["aspect"].isin(top_aspects), ASPECT_OTHER_LABEL))
+             .groupby([pd.Grouper(key=date_col, freq=freq), "aspect"]).size()
+             .unstack(fill_value=0).sort_index()
     )
-    return fig_bar, fig_pie, info
+    if use_ma:
+        win = _rolling_window(freq)
+        area_df = area_df.rolling(win, min_periods=1).mean()
+
+    # ผสาน area traces เข้า fig_line (รวมเป็นกราฟเดียว)
+    for col in area_df.columns:
+        fig_line.add_scatter(x=area_df.index, y=area_df[col], stackgroup="one",
+                             mode="lines", name=f"neg:{col}", opacity=0.4)
+    return fig_line
 
 # ================= Core Predict =================
 def _predict_batch(texts, model_name, batch_size=32):
@@ -272,22 +319,7 @@ def _predict_batch(texts, model_name, batch_size=32):
             })
     return results
 
-# ----- single -----
-def predict_one(text: str, model_choice: str):
-    try:
-        s = _norm_text(text)
-        if not _is_substantive_text(s):
-            return {"negative": 0.0, "positive": 0.0}, "invalid"
-        out = _predict_batch([s], model_choice)[0]
-        probs = {
-            "negative": float(out["negative(%)"].rstrip("%"))/100.0,
-            "positive": float(out["positive(%)"].rstrip("%"))/100.0,
-        }
-        return probs, out["label"]
-    except Exception as e:
-        return {"error": str(e)}, "error"
-
-# ----- textarea batch -----
+# ----- Batch (textarea) -----
 def predict_many(text_block: str, model_choice: str):
     try:
         raw_lines = (text_block or "").splitlines()
@@ -296,88 +328,158 @@ def predict_many(text_block: str, model_choice: str):
         skipped = len(all_norm) - len(cleaned)
         if len(cleaned) == 0:
             empty = pd.DataFrame(columns=["review","negative(%)","positive(%)","label"])
-            return empty, go.Figure(), go.Figure(), "No valid text"
+            return empty, go.Figure(), go.Figure(), go.Figure(), "No valid text"
         results = _predict_batch(cleaned, model_choice)
         df = pd.DataFrame(results)
-        fig_bar, fig_pie, info_md = make_charts_with_negative_aspects(df)
+        fig_bar, fig_pie, fig_topn, info_md = make_basic_charts(df)
         info_md = f"{info_md}  \n- Skipped: {skipped}"
-        return df, fig_bar, fig_pie, info_md
+        return df, fig_bar, fig_pie, fig_topn, info_md
     except Exception:
         tb = traceback.format_exc()
         empty = pd.DataFrame(columns=["review","negative(%)","positive(%)","label"])
-        return empty, go.Figure(), go.Figure(), f"**Error**\n```\n{tb}\n```"
+        return empty, go.Figure(), go.Figure(), go.Figure(), f"**Error**\n```\n{tb}\n```"
 
 # ----- CSV upload -----
 LIKELY_TEXT_COLS = ["text","review","message","comment","content","sentence","body","ข้อความ","รีวิว"]
+LIKELY_DATE_COLS = ["date","created_at","time","timestamp","datetime","วันที่","วันเวลา","เวลา"]
+
+def predict_csv(file_obj, model_choice: str, text_col_name: str,
+                date_col_name: str, date_from: str, date_to: str,
+                freq_choice: str, use_ma: bool, topn_aspects: int):
 
-def predict_csv(file_obj, model_choice: str, text_col_name: str):
     try:
         if file_obj is None:
-            return pd.DataFrame(), go.Figure(), go.Figure(), "Please upload a CSV.", None
+            return pd.DataFrame(), go.Figure(), go.Figure(), go.Figure(), go.Figure(), "Please upload a CSV.", None
 
-        df = pd.read_csv(file_obj.name)
-        cols = [c for c in df.columns]
-        # autodetect column if not provided
-        col = (text_col_name or "").strip()
-        if not col or col not in df.columns:
-            found = None
+        df_raw = pd.read_csv(file_obj.name)
+
+        # ----- pick text column -----
+        cols = list(df_raw.columns)
+        col_text = (text_col_name or "").strip()
+        if not col_text or col_text not in cols:
             low = {c.lower(): c for c in cols}
+            found = None
             for k in LIKELY_TEXT_COLS:
                 if k in low:
                     found = low[k]; break
             if found is None:
-                cand = [c for c in cols if df[c].dtype == object]
+                cand = [c for c in cols if df_raw[c].dtype == object]
                 found = cand[0] if cand else cols[0]
-            col = found
+            col_text = found
 
-        texts = [_norm_text(v) for v in df[col].tolist()]
+        texts = [_norm_text(v) for v in df_raw[col_text].tolist()]
         texts = [t for t in texts if _is_substantive_text(t)]
         if len(texts) == 0:
-            return pd.DataFrame(), go.Figure(), go.Figure(), "No valid texts in selected column.", None
+            return pd.DataFrame(), go.Figure(), go.Figure(), go.Figure(), go.Figure(), \
+                   "No valid texts in selected column.", None
 
+        # ----- predict -----
         results = _predict_batch(texts, model_choice)
         out_df = pd.DataFrame(results)
-        fig_bar, fig_pie, info_md = make_charts_with_negative_aspects(out_df)
 
-        # write downloadable csv
+        # ----- attach date if available -----
+        col_date = (date_col_name or "").strip()
+        if col_date and col_date in cols:
+            date_series = df_raw[col_date]
+        else:
+            found_d = None
+            low = {c.lower(): c for c in cols}
+            for k in LIKELY_DATE_COLS:
+                if k in low:
+                    found_d = low[k]; break
+            date_series = df_raw[found_d] if found_d else None
+
+        have_time = False
+        if date_series is not None:
+            dts = _to_datetime_safe(date_series)
+            if dts.notna().any():
+                out_df["__dt__"] = dts  # บางแถว parse ได้
+                have_time = True
+
+        # ----- basic charts (always) -----
+        neg_mask = out_df["label"] == "negative"
+        if neg_mask.any():
+            out_df.loc[neg_mask, "aspect"] = out_df.loc[neg_mask, "review"].apply(detect_aspect_for_negative)
+        else:
+            out_df["aspect"] = None
+
+        fig_bar, fig_pie, fig_topn, info_basic = make_basic_charts(out_df)
+
+        # ----- time charts (optional) -----
+        fig_time = go.Figure()
+        info_time = ""
+        if have_time:
+            df_time = out_df.dropna(subset=["__dt__"]).copy()
+            if date_from:
+                df_time = df_time[df_time["__dt__"] >= pd.to_datetime(date_from, errors="coerce")]
+            if date_to:
+                df_time = df_time[df_time["__dt__"] <= pd.to_datetime(date_to, errors="coerce")]
+            if len(df_time) == 0:
+                info_time = "**Note:** Selected date range has no data."
+            else:
+                fig_time = make_time_charts(df_time, "__dt__", freq_choice, use_ma, topn_aspects)
+                info_time = f"Time charts based on column: **{col_date or 'auto-detected'}**, Freq: **{freq_choice}**, MA: **{use_ma}**, TopN: **{topn_aspects}**"
+        else:
+            info_time = "_No date/timestamp column detected — time charts hidden._"
+
+        # ----- downloadable CSV -----
         fd, out_path = tempfile.mkstemp(prefix="pred_", suffix=".csv")
         os.close(fd)
-        out_df.to_csv(out_path, index=False, encoding="utf-8-sig")
+        out_df.drop(columns=["__dt__"], errors="ignore").to_csv(out_path, index=False, encoding="utf-8-sig")
+
+        info_md = info_basic + "\n\n" + info_time
+        return (
+            out_df.drop(columns=["__dt__"], errors="ignore"),
+            fig_bar, fig_pie, fig_topn, fig_time, info_md, out_path
+        )
 
-        info_md = f"{info_md}  \n- Column used: **{col}**"
-        return out_df, fig_bar, fig_pie, info_md, out_path
     except Exception:
         tb = traceback.format_exc()
-        return pd.DataFrame(), go.Figure(), go.Figure(), f"**Error**\n```\n{tb}\n```", None
+        return pd.DataFrame(), go.Figure(), go.Figure(), go.Figure(), go.Figure(), f"**Error**\n```\n{tb}\n```", None
 
 # ================= Gradio UI =================
 with gr.Blocks(title="Thai Sentiment (WangchanBERTa Variants)") as demo:
     gr.Markdown("### Thai Sentiment (WangchanBERTa Variants)")
     model_radio = gr.Radio(choices=AVAILABLE_CHOICES, value=DEFAULT_MODEL, label="เลือกโมเดล")
 
-    with gr.Tab("Single"):
-        t1 = gr.Textbox(lines=3, label="ข้อความรีวิว (1 ข้อความ)")
-        probs = gr.Label(label="Probabilities")
-        pred  = gr.Textbox(label="Prediction", interactive=False)
-        gr.Button("Predict").click(predict_one, [t1, model_radio], [probs, pred])
-
+    # ---- Batch (Textarea) ----
     with gr.Tab("Batch (หลายข้อความ)"):
         t2 = gr.Textbox(lines=8, label="พิมพ์หลายรีวิว (บรรทัดละ 1 รีวิว)")
         df2  = gr.Dataframe(label="ผลลัพธ์", interactive=False)
         bar2 = gr.Plot(label="Label counts (bar)")
-        pie2 = gr.Plot(label="Negative aspects (pie)")  # << เปลี่ยนเป็นพายหมวดลบ
+        pie2 = gr.Plot(label="Negative aspects (pie)")
+        top2 = gr.Plot(label="Top-N negative aspects (overall)")
         sum2 = gr.Markdown()
-        gr.Button("Run Batch").click(predict_many, [t2, model_radio], [df2, bar2, pie2, sum2])
+        gr.Button("Run Batch").click(
+            predict_many, [t2, model_radio], [df2, bar2, pie2, top2, sum2]
+        )
 
+    # ---- CSV Upload ----
     with gr.Tab("CSV Upload"):
-        file_in = gr.File(label="อัปโหลดไฟล์ .csv", file_types=[".csv"])
-        col_in  = gr.Textbox(label="ชื่อคอลัมน์ข้อความ (เว้นว่างให้เลือกอัตโนมัติได้)", value="")
+        with gr.Row():
+            file_in = gr.File(label="อัปโหลดไฟล์ .csv", file_types=[".csv"])
+            col_in  = gr.Textbox(label="ชื่อคอลัมน์ข้อความ (เว้นว่างให้เลือกอัตโนมัติ)", value="")
+            date_in = gr.Textbox(label="ชื่อคอลัมน์วันเวลา (เว้นว่างให้ auto-detect)", value="")
+        with gr.Row():
+            date_from = gr.Textbox(label="เริ่มวันที่ (YYYY-MM-DD)", value="")
+            date_to   = gr.Textbox(label="ถึงวันที่ (YYYY-MM-DD)", value="")
+            freq      = gr.Radio(choices=["D","W","M"], value="D", label="ความถี่ (Day/Week/Month)")
+            use_ma    = gr.Checkbox(value=True, label="Moving average (7/4/3)")
+            topn      = gr.Slider(3, 10, value=5, step=1, label="Top-N aspects (ลบ)")
+
         df3  = gr.Dataframe(label="ผลลัพธ์", interactive=False)
         bar3 = gr.Plot(label="Label counts (bar)")
-        pie3 = gr.Plot(label="Negative aspects (pie)")  # << พายหมวดลบ
+        pie3 = gr.Plot(label="Negative aspects (pie)")
+        top3 = gr.Plot(label="Top-N negative aspects (overall)")
+        line = gr.Plot(label="Reviews over time (POS/NEG + stacked negative aspects)")
         sum3 = gr.Markdown()
         dl3  = gr.File(label="ดาวน์โหลดผลเป็น CSV", interactive=False)
-        gr.Button("Predict CSV").click(predict_csv, [file_in, model_radio, col_in], [df3, bar3, pie3, sum3, dl3])
+
+        gr.Button("Predict CSV").click(
+            predict_csv,
+            [file_in, model_radio, col_in, date_in, date_from, date_to, freq, use_ma, topn],
+            [df3, bar3, pie3, top3, line, sum3, dl3]
+        )
 
 if __name__ == "__main__":
     demo.launch()