""" Gokul Ramanan panel_creation_predictions.py 8/6/2025 Description: Creates a panel dashboard to predict the number of runs some of the top young cricketers will score in 5 years for a specific format. Minimum Criteria for the prediction is 5 years of international experience in the format. """ import panel as pn from runs_api import RUNSAPI import plotly.graph_objects as go import plotly.colors as pc from statsmodels.nonparametric.smoothers_lowess import lowess import pandas as pd # Loads javascript dependencies and configures Panel (required) pn.extension('plotly', 'tabulator') # WIDGET DECLARATIONS api = RUNSAPI() api.load_runs("add_preds.csv") # Search Widgets format_select = pn.widgets.CheckBoxGroup(name="Format", options=["test", "odi", "t20i"], value = ["test", "odi", "t20i"]) country_select = pn.widgets.MultiSelect(name="Country", options=sorted(api.runs["Country"].unique().tolist()), size=6) year_slider = pn.widgets.IntRangeSlider(name="Year Range", start=api.runs["Year"].min(), end=api.runs["Year"].max(), step=1) top_n_slider = pn.widgets.IntSlider(name="Top N Players", start=1, end=12, value=12) player_select = pn.widgets.MultiChoice( name="Select Players", options=sorted(api.runs["Name"].unique().tolist()), placeholder="Choose players to compare (optional)..." ) career_length_slider = pn.widgets.IntRangeSlider( name="Career Length (Years)", start=1, end=25, step=1, value=(1, 25) ) all_formats_toggle = pn.widgets.Checkbox( name="Only Include Players in All 3 Formats", value=False ) not_formats_toggle = pn.widgets.Checkbox( name="Don't Include Players in All 3 Formats", value=False ) # Plotting widgets width = pn.widgets.IntSlider(name = 'Width', start = 250, end = 2000, step = 250, value = 1500) height = pn.widgets.IntSlider(name = 'Height', start = 200, end = 2500, step = 100, value = 800) color_by = pn.widgets.Select( name="Color By", options=["Name", "Country", "Debut Bin"], value="Name" ) career_align_toggle = pn.widgets.Checkbox( name="Align Careers to Year 0", value=False ) detailed_tooltip = pn.widgets.Checkbox( name="Show Detailed Hover Info", value=True ) plot_metric_select = pn.widgets.RadioButtonGroup( name="Plot Metric", options=["Cumulative Runs", "Cumulative Batting Average"], button_type="primary", value="Cumulative Runs" ) theme_map = { "white": "plotly_white", "dark": "plotly_dark", "gray1": "ggplot2", "gray2": "seaborn", "white2": "simple_white" } theme_select = pn.widgets.RadioButtonGroup( name="Plot Theme", options=list(theme_map.keys()), button_type="success", value="dark" ) # CALLBACK FUNCTIONS def get_plot(format_select, country_select, year_slider, top_n_slider, player_select, career_length_slider, width, height, theme_select_value, plot_metric_select_value, color_by_value, detailed_tooltip_value, align_career_value, all_formats_toggle, not_formats_toggle): """ Generate a cumulative runs line chart based on filter selections. Parameters: format_select (list): Selected match formats (e.g., ['odi', 'test']). country_select (list): List of selected countries. year_slider (list): List of [start_year, end_year] to filter data by year. top_n_slider (int): Number of top players to include based on total runs. width (int): Plot width in pixels. height (int): Plot height in pixels. theme_select_value (str): Theme name corresponding to Plotly templates. Returns: panel.pane.Plotly or panel.pane.Markdown: Plotly pane if data exists, otherwise a message pane. """ if plot_metric_select_value == "Cumulative Runs": y_col = "cumulative_format_runs" hover = "Runs" ranking_metric = "Runs" elif plot_metric_select_value == "Cumulative Batting Average": y_col = "cumulative_format_average" hover = "Average" ranking_metric = "Average" df = api.apply_filters(formats=format_select, countries=country_select, year_range=year_slider, top_n_players=top_n_slider, player_select_value = player_select, ranking_metric=ranking_metric, career_length_slider = career_length_slider, only_all_formats = all_formats_toggle, not_all_formats=not_formats_toggle) if plot_metric_select_value == "Cumulative Strike Rate": df = df[~df["Name"].isin(["Sachin Tendulkar", "Allan Border", "Javed Miandad", "Desmond Haynes", "Aravinda de Silva", "Mohammed Azharuddin", "Viv Richards", "Marvin Atapattu", "Sunil Gavaskar", "Saleem Malik", "Gordon Greenidge", "Arjuna Ranatunga", "Richie Richardson", ])] df = df[df["cumulative_SR"].notnull()] if df.empty: return pn.pane.Markdown("### No data for selected filters.", width=700) fig = go.Figure() grouped = df.groupby("Name") # Choose a color palette (20 vibrant colors) color_palette = ( pc.qualitative.Set3 + pc.qualitative.Set2 + pc.qualitative.Bold + pc.qualitative.Pastel + pc.qualitative.Dark2 + pc.qualitative.Safe ) # Get unique labels from your dataframe unique_labels = df[color_by_value].unique() color_map = {label: color_palette[i % len(color_palette)] for i, label in enumerate(sorted(unique_labels))} x_col = "Career Year" if align_career_value else "Year" x_axis_title = "Career Year" if align_career_value else "Year" for name, group in grouped: color_label = group[color_by_value].iloc[0] color = color_map.get(color_label, "#000000") custom_data = group[["cumulative_innings", "cumulative_matches", "cumulative_100s", "cumulative_50s"]].values if detailed_tooltip_value: hovertemplate = ( f"{name}
" "Year: %{x}
" f"{hover}: %{{y}}
" "Innings: %{customdata[0]}
" "Matches: %{customdata[1]}
" f"{color_by.name}: {color_label}" ) else: hovertemplate = ( f"{name}
" f"Year: %{{x}}
" f"{hover}: %{{y}}
" f"{color_by.name}: {color_label}" ) historical = group[group["Is_Prediction"] == False] predicted = group[group["Is_Prediction"] == True] # Historical (solid) fig.add_trace(go.Scatter( x=historical[x_col], y=historical[y_col], mode='lines+markers', name=name, customdata=historical[["cumulative_innings", "cumulative_matches", "cumulative_100s", "cumulative_50s"]], line=dict(color=color), legendgroup=color_label, hovertemplate=hovertemplate )) # Prediction (dot) if not predicted.empty: fig.add_trace(go.Scatter( x=predicted[x_col], y=predicted[y_col], mode='markers', name=f"{name} (Predicted)", customdata=predicted[["cumulative_innings", "cumulative_matches", "cumulative_100s", "cumulative_50s"]], line=dict(color=color, dash='dot'), marker=dict(size=10, symbol='diamond'), legendgroup=color_label, hovertemplate=hovertemplate.replace("", "[Predicted] ") )) if align_career_value: curve_df = df[[x_col, y_col]].dropna() if not curve_df.empty: smoothed = lowess(endog=curve_df[y_col], exog=curve_df[x_col], frac=0.2) fig.add_trace(go.Scatter( x=smoothed[:, 0], y=smoothed[:, 1], mode='lines', name="Best Fit Curve", line=dict(width=4, color='black', dash='dot'), hoverinfo='skip', showlegend=True )) fig.update_layout( title=f"Cumulative International {hover} Over Time", xaxis_title=x_axis_title, yaxis_title=f"Cumulative {hover}", width=width, height=height, showlegend=True, template= theme_map[theme_select_value] ) return pn.pane.Plotly(fig) def get_catalog(format_select, country_select, year_slider, top_n_slider,player_select,career_length_slider, all_formats_toggle, not_formats_toggle): """ Generate an interactive data table of filtered run statistics. Parameters: format_select (list): Selected match formats (e.g., ['t20i']). country_select (list): List of selected countries. year_slider (tuple): (start_year, end_year) year range filter. top_n_slider (int): Number of top run-scorers to include. Returns: panel.widgets.Tabulator: A paginated and scrollable data table of the filtered DataFrame. """ df = api.apply_filters(formats=format_select, countries=country_select, year_range=year_slider, top_n_players=top_n_slider, player_select_value= player_select, career_length_slider = career_length_slider, only_all_formats = all_formats_toggle, not_all_formats= not_formats_toggle) table = pn.widgets.Tabulator(df[["Name", "Year", "Country", "Matches", "Innings", "Runs", "cumulative_matches", "cumulative_innings", "cumulative_format_runs", "cumulative_format_average", "Career Year", "Debut Year", "Debut Bin"]], selectable=False, pagination = 'local', page_size = 20) return table # CALLBACK BINDINGS (Connecting widgets to callback functions) plot = pn.bind(get_plot, format_select, country_select, year_slider, top_n_slider, player_select, career_length_slider, width, height, theme_select, plot_metric_select, color_by, detailed_tooltip, career_align_toggle, all_formats_toggle, not_formats_toggle) catalog = pn.bind(get_catalog, format_select, country_select, year_slider, top_n_slider, player_select, career_length_slider, all_formats_toggle, not_formats_toggle) # === Player Selection Tab === player_selection_md = """ ### Player Selection Criteria Developed by Gokul Ramanan (August 2025) To ensure a balanced and representative analysis, we selected players using the following criteria: - **Top 15 in ICC Rankings (as of July 8, 2025)** in at least **one** format. - **Minimum 5–10 years of international batting experience** across formats. - If a player has played **11+ years**, they must be **under 35 years of age**. - **Exceptions (4 total)** were made for players with **exceptional potential** or realistic chances to **break all-time records**. - **If a player hasn’t played a format in the last 5 years**, they are considered **retired** in that format. Their past runs in that format are excluded for simplicity. --- ### Players and Format Eligibility | Player | Formats Eligible | Reason for Selection | |---------------|------------------|----------------------| | **Joe Root** | Test, ODI | Likely to challenge Sachin's Test run tally | | **Shubman Gill** | Test, ODI | Next big Indian star; now Indian Test captain | | **Suryakumar Yadav** | T20 | T20 beast; currently India T20I captain | | **Shreyas Iyer** | ODI, T20 | Key to India's white-ball success, CT hero | | **Travis Head** | All Three | Met all criteria requirements | | **Rishabh Pant** | All Three | Met all criteria requirements | | **Babar Azam** | All Three | Met all criteria requirements | | **Kusal Mendis** | All Three | Met all criteria requirements | | **Shai Hope** | All Three | Met all criteria requirements | | **KL Rahul** | All Three | Met all criteria requirements | | **Mohammad Rizwan** | All Three | Met all criteria requirements | | **Aiden Markram** | All Three | Met all criteria requirements | """ selection_tab = pn.pane.Markdown(player_selection_md, width=950) ml_walkthrough_md = """ ### Machine Learning Prediction Model Walkthrough This dashboard’s prediction pipeline begins by analyzing the **run-scoring trajectories of the top 85 batters in cricket history**. From that foundation, a hybrid model was designed to estimate a batter’s future scoring ability by combining three perspectives: their **recent form**, their **career average**, and their **trajectory average** (based on where they are in their career). These features were then weighted in a custom approach that consistently outperformed standard machine learning baselines like XGBoost when backtested across multiple eras. This provided a strong foundation for player-specific averages that realistically capture the effect of career phases, including late-career decline. The next stage focused on predicting **match volume and innings opportunities**, since raw scoring ability means little without chances to bat. Using ICC’s Future Tours Programme (FTP) data and historical schedules back to 2007, the model trained an XGBoost regressor to forecast the number of matches each country will play through 2030. These match forecasts were then combined with historical innings-per-match ratios to build an innings prediction engine. Backtesting this module produced exceptional accuracy, achieving **R² = 0.990** and **MAPE = 0.029**, with nearly 100% of predictions falling within a 20% error margin. Finally, the pipeline ties the pieces together: **Predicted Runs = Predicted Innings × Predicted Batting Average**. Each player’s projected 2030 endpoint is marked on the dashboard with a distinct diamond symbol, allowing users to compare actual versus forecasted career totals in real time. This end-to-end approach — from career form modeling to schedule forecasting — provides a transparent, accurate, and highly interpretable framework for projecting the next generation of international batting greats. It demonstrates how domain knowledge, careful backtesting, and hybrid modeling can create results that not only beat off-the-shelf ML models but also resonate with cricket’s real-world dynamics. """ ml_tab = pn.pane.Markdown(ml_walkthrough_md, width=950) # DASHBOARD WIDGET CONTAINERS ("CARDS") card_width = 320 search_card = pn.Card( pn.Column( # Widget 1 format_select, # Widget 2 country_select, # Widget 3 year_slider, # Widget 4 top_n_slider, # Widget 5 player_select, # Widget 6 career_length_slider, all_formats_toggle, not_formats_toggle ), title="Search", width=card_width, collapsed=False ) plot_card = pn.Card( pn.Column( career_align_toggle, # Default Widget plot_metric_select, # Default Widget 2 color_by, # Default Widget 3 detailed_tooltip, # Widget 1 width, # Widget 2 height, # Widget 3 theme_select ), title="Plot", width=card_width, collapsed=False ) # LAYOUT layout = pn.template.FastListTemplate( title="Top Prospective International Cricket Batters: Predicted Runs by 2030", sidebar=[ search_card, plot_card, ], theme_toggle=False, main=[ pn.Tabs( ("Table", catalog), ("Time Series", plot), ("Selection Criteria", selection_tab), ("ML Model Walkthrough", ml_tab), active=1 # Which tab is active by default? ) ], header_background='#a93226' ).servable()