Update

app.py

@@ -1,3 +1,4 @@
+
 import spaces                        # for ZeroGPU support
 import gradio as gr
 import pandas as pd
@@ -11,7 +12,7 @@ from transformers import (
     TextIteratorStreamer
 )
 
-# MODEL SETUP
+# ─── MODEL SETUP ────────────────────────────────────────────────────────────────
 MODEL_NAME = "bytedance-research/ChatTS-14B"
 
 tokenizer = AutoTokenizer.from_pretrained(
@@ -28,37 +29,70 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 model.eval()
 
-# ─── HELPER FUNCTIONS ──────────────────────────────────────────────────────────
-def create_default_timeseries():
-    """Create default time series with a sudden change.
+# ─── VISUAL THEME (Dark) ───────────────────────────────────────────────────────
+CSS = """
+:root{
+  --bg:#0b1020; --bg-2:#10172e; --panel:#141b34; --panel-2:#0f172a;
+  --border:#263154; --fg:#e7ecf5; --muted:#b9c2d3; --placeholder:#7c8195;
+}
+/* App background */
+.gradio-container{background: linear-gradient(180deg,var(--bg),var(--bg) 40%, var(--bg-2)); color: var(--fg);}
+/* Generic block panels */
+.gradio-container .block, .gradio-container .form, .gradio-container .padded{
+  background: linear-gradient(180deg,var(--panel),var(--panel-2));
+  border: 1px solid var(--border);
+  border-radius: 14px;
+}
+/* Inputs */
+input[type="text"], textarea, select, .gr-textbox, .gr-dropdown, .gr-textbox textarea{
+  background: #0f1630 !important; color: var(--fg) !important;
+  border: 1px solid var(--border) !important; border-radius: 12px !important;
+}
+.gr-textbox textarea::placeholder{ color: var(--placeholder) !important; }
+label, .svelte-1gfkn6j, .svelte-i3tvor, .label{ color: var(--muted) !important; }
 
-    Returns a pandas DataFrame containing two simple synthetic time series.
-    """
-    x1 = np.arange(256)
-    x2 = np.arange(256)
-    ts1 = np.sin(x1 / 10) * 5.0
+/* Upload panel */
+button.svelte-1x5qevo{ background:#0f1630 !important; color:var(--muted) !important; border-color:var(--border) !important; }
+.svelte-12ioyct{ color: var(--muted) !important; }
+
+/* Line plot container */
+.vega-embed, canvas.marks{ background:#0b1020 !important; border-radius: 10px; }
+
+/* Buttons base */
+.gr-button{ border-radius:14px !important; border:1px solid var(--border) !important; color:#0b1020 !important; font-weight:700; }
+.gr-button:hover{ transform: translateY(-1px); box-shadow:0 6px 18px rgba(0,0,0,.35); }
+.gr-button-primary{ background: linear-gradient(180deg,#8ab4ff,#7aa2ff) !important; }
+.gr-button-secondary{ background: linear-gradient(180deg,#a78bfa,#7aa2ff) !important; color:#0b1020 !important; }
+.btn-chip{ padding:12px 16px !important; }
+
+/* Example cards row */
+.examples-row{ display:grid; grid-template-columns:repeat(auto-fit,minmax(180px,1fr)); gap:12px; margin:12px 0 8px; }
+.cardbtn{ border:1px solid var(--border) !important; color:#0b1020 !important; }
+.grad-blue{ background: linear-gradient(180deg,#93c5fd,#60a5fa) !important; }
+.grad-purple{ background: linear-gradient(180deg,#c4b5fd,#a78bfa) !important; }
+.grad-green{ background: linear-gradient(180deg,#bbf7d0,#34d399) !important; }
+.grad-amber{ background: linear-gradient(180deg,#fde68a,#f59e0b) !important; }
+.grad-rose{ background: linear-gradient(180deg,#fecaca,#fb7185) !important; }
+.cardbtn:hover{ filter:brightness(1.02); transform:translateY(-1px); }
+
+/* Subtle spacing fixes */
+.gap, .padded{ padding:12px !important; }
+"""
+
+# ─── HELPERS ───────────────────────────────────────────────────────────────────
+def create_default_timeseries():
+    x = np.arange(256)
+    ts1 = np.sin(x / 10) * 5.0
     ts1[103:] -= 10.0
-    ts2 = x2 * 0.01
+    ts2 = x * 0.01
     ts2[100] += 10.0
-    df = pd.DataFrame({
-        "TS1": ts1,
-        "TS2": ts2
-    })
-    return df
+    return pd.DataFrame({"TS1": ts1, "TS2": ts2})
 
 def process_csv_file(csv_file):
-    """Process CSV file and return DataFrame with validation.
-
-    Validates the uploaded CSV file and ensures that each column represents a
-    time series. Returns (df, message) where df is a pandas DataFrame or None
-    and message is a user friendly status string.
-    """
     if csv_file is None:
         return None, "No file uploaded"
     try:
-        # pandas will handle file objects, but for gradio we want to read by name
         df = pd.read_csv(csv_file.name)
-        # strip and drop empty column names
         df.columns = [str(c).strip() for c in df.columns]
         df = df.loc[:, [c for c in df.columns if c]]
         df = df.dropna(axis=1, how="all")
@@ -66,117 +100,66 @@ def process_csv_file(csv_file):
             return None, "No valid time-series columns found."
         if df.shape[1] > 15:
             return None, f"Too many series ({df.shape[1]}). Max allowed = 15."
-        # Validate each column
-        ts_names, ts_list = [], []
+        ts_names = []
+        ts_list = []
         for name in df.columns:
             series = df[name]
-            # ensure numeric
             if not pd.api.types.is_float_dtype(series):
-                try:
-                    series = pd.to_numeric(series, errors='coerce')
-                except Exception:
-                    return None, f"Series '{name}' cannot be converted to float type."
+                series = pd.to_numeric(series, errors='coerce')
             last_valid = series.last_valid_index()
             if last_valid is None:
                 continue
             trimmed = series.loc[:last_valid].to_numpy(dtype=np.float32)
-            length = len(trimmed)
-            if length < 64 or length > 1024:
-                return None, f"Series '{name}' length {length} invalid. Must be 64 to 1024."
-            ts_names.append(name)
-            ts_list.append(trimmed)
+            L = trimmed.shape[0]
+            if L < 64 or L > 1024:
+                return None, f"Series '{name}' length {L} invalid. Must be 64 to 1024."
+            ts_names.append(name); ts_list.append(trimmed)
         if not ts_list:
             return None, "All time series are empty after trimming NaNs."
-        return df, f"Successfully loaded {len(ts_names)} time series: {', '.join(ts_names)}"
+        return df, f"Loaded {len(ts_names)} series: {', '.join(ts_names)}"
     except Exception as e:
         return None, f"Error processing file: {str(e)}"
 
 def preview_csv(csv_file, use_default):
-    """Preview uploaded CSV file or default time series immediately.
-
-    Returns a gr.LinePlot, status message, dropdown component, and a boolean
-    indicating whether the default dataset is being used.
-    """
     if csv_file is None:
-        # no file, return empty plot and message
         return gr.LinePlot(value=pd.DataFrame()), "Please upload a CSV file first", gr.Dropdown(), False
     df, message = process_csv_file(csv_file)
     if df is None:
         return gr.LinePlot(value=pd.DataFrame()), message, gr.Dropdown(), False
-    # Build dropdown choices and default plot
-    column_choices = list(df.columns)
-    first_column = column_choices[0]
-    df_with_index = df.copy()
-    df_with_index["_internal_idx"] = np.arange(len(df[first_column].values))
-    plot = gr.LinePlot(
-        df_with_index,
-        x="_internal_idx",
-        y=first_column,
-        title=f"Time Series: {first_column}"
-    )
-    dropdown = gr.Dropdown(
-        choices=column_choices,
-        value=first_column,
-        label="Select a Column to Visualize"
-    )
+    choices = list(df.columns)
+    first = choices[0]
+    df_idx = df.copy(); df_idx["_i"] = np.arange(len(df[first].values))
+    plot = gr.LinePlot(df_idx, x="_i", y=first, title=f"Time Series: {first}")
+    dropdown = gr.Dropdown(choices=choices, value=first, label="Select a Column to Visualize")
     return plot, message, dropdown, False
 
 def clear_csv():
-    """Clear uploaded CSV file immediately."""
-    return gr.LinePlot(value=pd.DataFrame()), "No file uploaded", gr.Dropdown()
+    return gr.LinePlot(value=pd.DataFrame()), "Cleared.", gr.Dropdown()
 
 def update_plot(csv_file, selected_column, use_default_state):
-    """Update plot based on selected column or default state."""
     if (csv_file is None and not use_default_state) or selected_column is None:
         return gr.LinePlot(value=pd.DataFrame())
     if csv_file is None and use_default_state:
         df = create_default_timeseries()
     else:
         df, _ = process_csv_file(csv_file)
-        if df is None:
-            return gr.LinePlot(value=pd.DataFrame())
-    df_with_index = df.copy()
-    df_with_index["_internal_idx"] = np.arange(len(df[selected_column].values))
-    plot = gr.LinePlot(
-        df_with_index,
-        x="_internal_idx",
-        y=selected_column,
-        title=f"Time Series: {selected_column}"
-    )
-    return plot
+    if df is None:
+        return gr.LinePlot(value=pd.DataFrame())
+    df_idx = df.copy(); df_idx["_i"] = np.arange(len(df[selected_column].values))
+    return gr.LinePlot(df_idx, x="_i", y=selected_column, title=f"Time Series: {selected_column}")
 
 def initialize_interface():
-    """Initialize interface with default time series.
-
-    Returns a plot, a status message, dropdown and boolean state.
-    """
     df = create_default_timeseries()
-    column_choices = list(df.columns)
-    first_column = column_choices[0]
-    df_with_index = df.copy()
-    df_with_index["_internal_idx"] = np.arange(len(df[first_column].values))
-    plot = gr.LinePlot(
-        df_with_index,
-        x="_internal_idx",
-        y=first_column,
-        title=f"Time Series: {first_column}"
-    )
-    dropdown = gr.Dropdown(
-        choices=column_choices,
-        value=first_column,
-        label="Select a Column to Visualize"
-    )
-    message = "Using default time series (TS1 and TS2). Select a series from the dropdown for visualization."
-    return plot, message, dropdown, True
+    choices = list(df.columns); first = choices[0]
+    df_idx = df.copy(); df_idx["_i"] = np.arange(len(df[first].values))
+    plot = gr.LinePlot(df_idx, x="_i", y=first, title=f"Time Series: {first}")
+    dropdown = gr.Dropdown(choices=choices, value=first, label="Select a Column to Visualize")
+    msg = "Using default time series (TS1 and TS2). Select a series from the dropdown for visualization."
+    return plot, msg, dropdown, True
 
+# ─── INFERENCE ─────────────────────────────────────────────────────────────────
 @spaces.GPU
 def infer_chatts_stream(prompt: str, csv_file, use_default):
-    """
-    Streaming version of ChatTS inference.
-
-    Constructs a full prompt using the uploaded or default time series and yields
-    the assistant's response incrementally.
-    """
     if not prompt.strip():
         yield "Please enter a prompt"
         return
@@ -189,15 +172,12 @@ def infer_chatts_stream(prompt: str, csv_file, use_default):
         yield "Please upload a CSV file first or the file contains errors"
         return
     try:
-        # Prepare time series data
         ts_names, ts_list = [], []
         for name in df.columns:
-            series = df[name]
-            last_valid = series.last_valid_index()
+            series = df[name]; last_valid = series.last_valid_index()
             if last_valid is not None:
                 trimmed = series.loc[:last_valid].to_numpy(dtype=np.float32)
-                ts_names.append(name)
-                ts_list.append(trimmed)
+                ts_names.append(name); ts_list.append(trimmed)
         if not ts_list:
             yield "No valid time series data found. Please upload time series first."
             return
@@ -207,186 +187,74 @@ def infer_chatts_stream(prompt: str, csv_file, use_default):
             prefix += f"The {name} is of length {len(arr)}: <ts><ts/>\n"
         full_prompt = (
             "<|im_start|>system\nYou are a helpful assistant. Your name is ChatTS. "
-            "You can analyze time series data and provide insights. If the user asks who you are, "
-            "you should give your name and capabilities in the language of the prompt. "
-            "If no time series are provided, you should say 'I cannot answer this question as you "
-            "haven't provided the timeseries I need' in the language of the prompt. "
-            "Always check if the user has provided at least one time series data before answering."
+            "You can analyze time series data and provide insights. If user asks who you are, you should give your name and capabilities "
+            "in the language of the prompt. If no time series are provided, you should say 'I cannot answer this question as you haven't provide the timeseries I need' "
+            "in the language of the prompt. Always check if the user has provided at least one time series data before answering."
             "<|im_end|><|im_start|>user\n"
-            f"{prefix}{clean_prompt} "
-            "Please output a step-by-step analysis about the time series attributes that are mentioned "
-            "in the question first, and then give a detailed result about this question. "
-            "Always remember to carefully double check the values before answering the results."
+            f"{prefix}{clean_prompt} Please output a step-by-step analysis about the time series attributes that mentioned in the question first, "
+            "and then give a detailed result about this question. Always remember to carefully double check the values before answer the results."
             "<|im_end|><|im_start|>assistant\n"
         )
-        inputs = processor(
-            text=[full_prompt],
-            timeseries=ts_list,
-            padding=True,
-            return_tensors="pt"
-        )
+        inputs = processor(text=[full_prompt], timeseries=ts_list, padding=True, return_tensors="pt")
         inputs = {k: v.to(model.device) for k, v in inputs.items()}
         streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
-        inputs.update({
-            "max_new_tokens": 512,
-            "streamer": streamer,
-            "temperature": 0.3
-        })
-        thread = Thread(
-            target=model.generate,
-            kwargs=inputs
-        )
-        thread.start()
-        model_output = ""
+        inputs.update({"max_new_tokens": 512, "streamer": streamer, "temperature": 0.3})
+        thread = Thread(target=model.generate, kwargs=inputs); thread.start()
+        out = ""
         for new_text in streamer:
-            model_output += new_text
-            yield model_output
+            out += new_text
+            yield out
     except Exception as e:
         yield f"Error during inference: {str(e)}"
 
-# Example prompts for quick analysis.
-EXAMPLE_PROMPTS = [
-    "Analyze all the given time series and summarize the overall trends.",
-    "Identify and describe any anomalies in each time series.",
-    "Compare the patterns between the first and second time series.",
-    "Describe any seasonal patterns present in the time series.",
-    "Find the maximum and minimum values in each series and comment on them."
-]
-
-# Custom CSS to style the app in a dark theme and create card-like example buttons.
-CSS = """
-/* Dark background and base text color */
-.gradio-container, .gr-block, .gr-row, .gr-column {
-    background-color: #111827 !important;
-    color: #f3f4f6 !important;
-}
-/* Style labels */
-label, .gr-textbox label, .gr-dropdown label {
-    color: #f3f4f6 !important;
-}
-/* Style input elements */
-textarea, select, input[type=text], input[type=number], .gr-dropdown select {
-    background-color: #1f2937 !important;
-    color: #f3f4f6 !important;
-    border: 1px solid #374151 !important;
-}
-/* Buttons styling */
-button, .gr-button {
-    background-color: #374151 !important;
-    color: #f3f4f6 !important;
-    border: 1px solid #4b5563 !important;
-}
-button:hover, .gr-button:hover {
-    background-color: #4b5563 !important;
-}
-/* Example card buttons */
-.example-card {
-    background-color: #1f2937 !important;
-    color: #f3f4f6 !important;
-    border: 1px solid #4b5563 !important;
-    border-radius: 8px;
-    padding: 10px;
-    margin-right: 10px;
-    cursor: pointer;
-    display: flex;
-    justify-content: center;
-    align-items: center;
-    transition: background-color 0.2s ease;
+# ─── EXAMPLES (semantic names) ─────────────────────────────────────────────────
+EXAMPLE_PROMPTS = {
+    "🔍 Detect Spikes": "Identify all spikes in each series <ts><ts/> and report timestamps and magnitudes. Explain briefly.",
+    "📈 Trend & Seasonality": "Describe the trend and seasonality for the provided time series <ts><ts/>. Estimate the dominant period length and amplitude.",
+    "🔗 Compare Metrics": "Compare the two series <ts><ts/>. Are there lagged correlations? Estimate the lag and correlation strength.",
+    "⚡ Local Change Analysis": "Find intervals with >10% rise or drop relative to the prior 20 points for <ts><ts/>. Return intervals and reasons.",
+    "📊 Correlation Strength": "Quantify Pearson correlation between each pair of series <ts><ts/> and highlight the strongest positive/negative pairs."
 }
-.example-card:hover {
-    background-color: #374151 !important;
-}
-"""
 
-# ─── BUILD THE GRADIO APP ─────────────────────────────────────────────────────
+# ─── UI ────────────────────────────────────────────────────────────────────────
+with gr.Blocks(title="ChatTS Demo", css=CSS) as demo:
+    use_default_state = gr.State(value=True)
 
-with gr.Blocks(title="ChatTS Demo (Dark Theme)", css=CSS) as demo:
-    # Create example buttons in a row.
-    with gr.Row():
-        example_buttons = []
-        for idx, prompt_text in enumerate(EXAMPLE_PROMPTS):
-            btn = gr.Button(f"Example {idx+1}", elem_classes="example-card")
-            example_buttons.append((btn, prompt_text))
+    gr.HTML("<h3 style='text-align:center; color:#e8ebfa; font-size:20px; font-weight:700; margin-top:8px;'>💡 Click an example below and click Run ChatTS to try different questions with default time series or upload your own CSV file.</h3>")
 
-    # Hidden state to track whether to use default time series
-    use_default_state = gr.State(value=True)
+    with gr.Row(elem_classes="examples-row"):
+        btns = {}
+        grads = ["grad-blue","grad-purple","grad-green","grad-amber","grad-rose"]
+        for (title, prompt), grad in zip(EXAMPLE_PROMPTS.items(), grads):
+            btns[title] = gr.Button(title, elem_classes=f"cardbtn btn-chip {grad}")
 
-    # Upload, prompt input and run button
     with gr.Row():
         with gr.Column(scale=1):
-            upload = gr.File(
-                label="Upload CSV File",
-                file_types=[".csv"],
-                type="filepath"
-            )
+            upload = gr.File(label="", file_types=[".csv"], type="filepath")
             prompt_input = gr.Textbox(
                 lines=6,
                 placeholder="Enter your question here...",
                 label="Analysis Prompt",
-                value="Please analyze all the given time series and provide insights about the local fluctuations in the time series in detail."
+                value="Find the maximum and minimum values in each series and comment on them."
             )
             run_btn = gr.Button("Run ChatTS", variant="primary")
         with gr.Column(scale=2):
-            series_selector = gr.Dropdown(
-                label="Select a Column to Visualize",
-                choices=[],
-                value=None
-            )
+            series_selector = gr.Dropdown(label="Select a Column to Visualize", choices=[], value=None)
             plot_out = gr.LinePlot(value=pd.DataFrame(), label="Time Series Visualization")
-            file_status = gr.Textbox(
-                label="File Status",
-                interactive=False,
-                lines=2
-            )
+            file_status = gr.Textbox(label="File Status", interactive=False, lines=2)
 
-    # Output area
-    text_out = gr.Textbox(
-        lines=10,
-        label="ChatTS Analysis Results",
-        interactive=False
-    )
+    text_out = gr.Textbox(lines=10, label="ChatTS Analysis Results", interactive=False)
 
-    # Initialize with default data
-    demo.load(
-        fn=initialize_interface,
-        outputs=[plot_out, file_status, series_selector, use_default_state]
-    )
+    demo.load(fn=initialize_interface, outputs=[plot_out, file_status, series_selector, use_default_state])
 
-    # Event handlers for upload and dropdown
-    upload.upload(
-        fn=preview_csv,
-        inputs=[upload, use_default_state],
-        outputs=[plot_out, file_status, series_selector, use_default_state]
-    )
-    upload.clear(
-        fn=clear_csv,
-        inputs=[],
-        outputs=[plot_out, file_status, series_selector]
-    )
-    series_selector.change(
-        fn=update_plot,
-        inputs=[upload, series_selector, use_default_state],
-        outputs=[plot_out]
-    )
+    for title, prompt in EXAMPLE_PROMPTS.items():
+        btns[title].click(fn=lambda p=prompt: p, outputs=prompt_input)
 
-    # Event handler for run button to perform inference
-    run_btn.click(
-        fn=infer_chatts_stream,
-        inputs=[prompt_input, upload, use_default_state],
-        outputs=[text_out]
-    )
-
-    # Example buttons update the prompt input when clicked
-    for btn, prompt_text in example_buttons:
-        # Use closure to capture prompt_text
-        def click_fn(p=prompt_text):
-            return p
-        btn.click(
-            fn=click_fn,
-            inputs=[],
-            outputs=[prompt_input],
-            queue=False  # no queuing required for simple update
-        )
+    upload.upload(fn=preview_csv, inputs=[upload, use_default_state],
+                  outputs=[plot_out, file_status, series_selector, use_default_state])
+    upload.clear(fn=clear_csv, outputs=[plot_out, file_status, series_selector])
+    series_selector.change(fn=update_plot, inputs=[upload, series_selector, use_default_state], outputs=[plot_out])
+    run_btn.click(fn=infer_chatts_stream, inputs=[prompt_input, upload, use_default_state], outputs=[text_out])
 
 if __name__ == '__main__':
     demo.launch()

requirements.txt

@@ -4,4 +4,4 @@ numpy
 pandas
 accelerate
 torch>=2.2.0
-pydantic==2.10.6
+pydantic==2.10.6