anomaly-pipeline 0.1.27__py3-none-any.whl

anomaly_pipeline/helpers/evaluation_plots.py

@@ -0,0 +1,546 @@
+import pandas as pd
+import numpy as np
+import plotly.graph_objects as go
+import plotly.express as px
+
+
+def initialize_fig(group, group_columns, variable, date_column, anomaly_detection_model):
+    
+    plot_title = "  --  ".join(list(group[group_columns].values[0])).upper() + "  --  " + anomaly_detection_model
+    
+    fig = go.Figure()
+    
+    # Actuals
+    fig.add_trace(go.Scatter(
+        x=group[date_column],
+        y=group[variable],
+        mode='lines',
+        line=dict(color='seagreen', width=1.5),
+        name=variable if variable == variable.upper() else variable.title(),
+    ))
+
+    fig.update_layout(
+        title=dict(
+                text=plot_title,
+                y=0.96,
+                x=0.5,
+                xanchor='center',
+                yanchor='top',
+                font=dict(size=18, color='black', weight='bold'),
+            ),
+        height=350,
+        width=1200,
+        margin=dict(l=50, r=50, t=40, b=30),
+        plot_bgcolor='snow',
+        paper_bgcolor='whitesmoke',
+        xaxis=dict(
+            range=[group[date_column].min(), group[date_column].max()],
+            showline=True,
+            linewidth=0.5,
+            linecolor='orange',
+            zeroline=False,
+            gridcolor='rgba(255, 165, 0, 0.1)',
+            mirror=True
+            ),
+        yaxis=dict(
+            range=[group[variable].min()*0.9, group[variable].max()*1.06],
+            showline=True,
+            linewidth=0.5,
+            linecolor='orange',
+            zeroline=False,
+            gridcolor='rgba(255, 165, 0, 0.1)',
+            mirror=True
+            ),
+        yaxis_title=dict(
+            text=variable if variable == variable.upper() else variable.title(),
+            font=dict(size=16, weight='bold', color='black')
+            ),
+        legend=dict(
+            orientation="v",
+            yanchor="top",
+            y=1,
+            xanchor="left",
+            x=1.02,
+            )
+        )
+    
+    return fig
+
+
+def add_anomalies(fig, group, date_column, variable):
+    # Add anomalies markers
+    fig.add_trace(go.Scatter(
+        x=group[group['is_Anomaly'] == True][date_column],
+        y=group[group['is_Anomaly'] == True][variable],
+        mode='markers',
+        marker=dict(color='pink', symbol='cross', line=dict(width=1), size=9),
+        name='Anomalies',
+        hoverinfo='skip',
+        ))
+    return fig
+
+
+def add_model_anomalies(fig, group, date_column, variable, model):
+    fig.add_trace(go.Scatter(
+        x=group[group[f'is_{model}_anomaly'] == True][date_column],
+        y=group[group[f'is_{model}_anomaly'] == True][variable],
+        mode='markers',
+        marker=dict(color='palevioletred', symbol='circle', line=dict(width=1), size=9),
+        name=f'{model} Anomalies',
+        customdata=group[group[f'is_{model}_anomaly'] == True][[f'{model}_anomaly']],
+        hovertemplate=(
+            f'Date: %{{x|%Y-%m-%d}}<br>' +
+            f'{variable if variable == variable.upper() else variable.title()}: %{{y:~.2s}}<br>' +
+            f'{model}' + ' Category: %{customdata[0]}<extra></extra>'
+            )
+        ))
+    return fig
+
+
+def add_anomaly_region(fig, region, group, variable, date_column, threshold_column, threshold_label):
+    
+    if region == 'upper':
+        y0 = group[threshold_column].values[0]
+        y1 = group[variable].max()*1.06
+    elif region == 'lower':
+        y0 = 0
+        y1 = group[threshold_column].values[0]
+    
+    # Shading
+    fig.add_shape(
+        type="rect",
+        x0=0, x1=1, xref="paper",
+        y0=y0, y1=y1,
+        yref="y",
+        fillcolor="rgba(255, 0, 0, 0.055)",
+        line=dict(width=0),
+        layer="below"
+    )
+
+    # Upper Percentile line
+    fig.add_trace(go.Scatter(
+        x=group[date_column],
+        y=group[threshold_column],
+        mode='lines',
+        line=dict(color='orangered', width=1, dash='dashdot'),
+        name=threshold_label,
+        showlegend=False
+        ))
+    
+    return fig
+
+
+def add_eval_period_highlight(fig, group, date_column, variable, eval_period):
+    fig.add_trace(go.Scatter(
+        x=group[date_column][-eval_period:],
+        y=group[variable][-eval_period:],
+        mode='lines',
+        line=dict(
+            color='rgba(0, 255, 0, 0.25)', # 'lime' with 0.25 alpha
+            width=10
+        ),
+        name='Evalution Period',
+        hoverinfo='skip',
+    ))
+    return fig
+
+
+def anomaly_eval_plot(group, group_columns, variable, date_column, eval_period=12, show_anomaly_scores_on_main_plot=False):
+    
+    # IS ANOMALY Plot
+    # This is the main plot
+    try:
+        group = group.copy()
+
+        anomaly_cols = []
+        for col in group.columns.to_list():
+            if col.startswith('is_') and col.endswith('_anomaly') and col != 'is_anomaly':
+                anomaly_cols.append(col)
+        group['Anomaly Vote Models'] = group.apply(
+            lambda row: sorted([col.removeprefix('is_').removesuffix('_anomaly')
+                for col in anomaly_cols
+                if pd.notna(row[col]) and row[col] == True]),
+            axis=1)
+        group['Anomaly Vote Models'] = group['Anomaly Vote Models'].apply(lambda x: ', '.join(x))
+        group['Anomaly_Votes'] = group[anomaly_cols].sum(axis=1).astype(int)
+        group['Vote_Cnt'] = group[anomaly_cols].replace(False, True).sum(axis=1).astype(int)
+        group['Anomaly_Votes_Display'] = group['Anomaly_Votes'].astype(str) + ' out of ' + group['Vote_Cnt'].astype(str)
+        group['is_Anomaly'] = np.where(group['Anomaly_Votes']/group['Vote_Cnt'] >= 0.5, True, False)
+        group['Anomaly_Score'] = 2 * (group['Anomaly_Votes']/group['Vote_Cnt'] - 0.5).astype(float)
+        group['Anomaly_Score_Display'] = np.where(group['Anomaly_Score'] < 0, np.floor(100*group['Anomaly_Score']),
+                                                  np.where(group['Anomaly_Score'] > 0, np.ceil(100*group['Anomaly_Score']),
+                                                  1)).astype(float)
+        group['Mean'] = group[variable].mean()
+        group['Median'] = group[variable].median()
+
+        fig = initialize_fig(group, group_columns, variable, date_column, "Anomalies")
+
+        # Mean
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['Mean'],
+            mode='lines',
+            line=dict(color='maroon', width=0.7, dash='dash'),
+            name='Mean',
+            showlegend=True,
+            hoverinfo='skip',
+        ))
+
+        # Median
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['Median'],
+            mode='lines',
+            line=dict(color='darkblue', width=0.7, dash='dot'),
+            name='Median',
+            showlegend=True,
+            hoverinfo='skip',
+        ))
+
+        # Anomalies
+        fig.add_trace(go.Scatter(
+            x=group[group['is_Anomaly'] == True][date_column],
+            y=group[group['is_Anomaly'] == True][variable],
+            mode='markers',
+            marker=dict(color='red', symbol='circle', line=dict(width=1), size=5*(group[group['is_Anomaly'] == True]['Anomaly_Score'] + 2)),
+            name='Anomalies',
+            customdata=group[group['is_Anomaly'] == True][['Anomaly_Votes_Display', 'Anomaly Vote Models', 'Anomaly_Score_Display']],
+            hovertemplate=(
+                f'Date: %{{x|%Y-%m-%d}}<br>' +
+                f'{variable if variable == variable.upper() else variable.title()}: %{{y:,d}}<br>' +
+                'Anomaly Votes: %{customdata[0]}<br>' +
+                'Anomaly Vote Models: %{customdata[1]}<br>' +
+                'Anomaly Score: %{customdata[2]}<extra></extra><br>'
+                )
+            ))
+
+        # Near Anomalies
+        fig.add_trace(go.Scatter(
+            x=group[(group['is_Anomaly'] == False) & (group['Anomaly_Votes'] >= 1)][date_column],
+            y=group[(group['is_Anomaly'] == False) & (group['Anomaly_Votes'] >= 1)][variable],
+            mode='markers',
+            marker=dict(color='orange',
+                        symbol='circle',
+                        line=dict(width=1),
+                        size=5*(group[(group['is_Anomaly'] == False) & (group['Anomaly_Votes'] >= 1)]['Anomaly_Score'] + 2)),
+            name='Not Quite Anomalies',
+            customdata=group[(group['is_Anomaly'] == False) & (group['Anomaly_Votes'] >= 1)][['Anomaly_Votes_Display', 'Anomaly Vote Models', 'Anomaly_Score_Display']],
+            hovertemplate=(
+                f'Date: %{{x|%Y-%m-%d}}<br>' +
+                f'{variable if variable == variable.upper() else variable.title()}: %{{y:,d}}<br>' +
+                'Anomaly Votes: %{customdata[0]}<br>' +
+                'Anomaly Vote Models: %{customdata[1]}<br>' +
+                'Anomaly Score: %{customdata[2]}<extra></extra><br>'
+                )
+            ))
+
+        # Add Anomaly Scores to Secondary Axis
+        if show_anomaly_scores_on_main_plot:
+            fig.add_trace(go.Bar(
+                x=group[date_column],
+                y=group['Anomaly_Score_Display'],
+                name='Anomaly Score',
+                marker=dict(
+                    color=np.where(group['Anomaly_Score_Display'] > 0, 'rgba(255, 0, 0, 0.35)', 'rgba(128, 128, 128, 0.15)'),
+                    # line=dict(width=0.5, color='gray')
+                ),
+                yaxis='y2',
+                # Ensure it doesn't clutter the main hover box
+                hoverinfo='y+name',
+                showlegend=True
+                ))
+
+            fig.update_layout(
+                yaxis2=dict(
+                        title='Anomaly Score',
+                        overlaying='y',
+                        side='right',
+                        range=[-105, 105],
+                        showgrid=False,
+                    ),
+                margin=dict(l=50, r=200, t=50, b=50),
+                legend=dict(
+                    orientation="v",
+                    yanchor="top",
+                    y=1,
+                    xanchor="left",
+                    x=1.07,
+                ))
+
+        fig.show()
+        print("\n")
+        
+    except Exception as e:
+        print(f"Anomaly Plot Failed: {e}")
+
+
+def anomaly_percentile_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # Percentile Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "Percentile Anomaly Detection")
+        # Lower anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'lower', group, variable, date_column, 'Percentile_low', 'Percentile Low')
+        # Upper anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'upper', group, variable, date_column, 'Percentile_high', 'Percentile High')
+        # Percentile Anomalies
+        fig = add_model_anomalies(fig, group, date_column, variable, 'Percentile')
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"Percentile Anomaly Plot Failed: {e}") 
+
+
+def anomaly_sd_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # SD Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "SD Anomaly Detection")
+        # Lower anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'lower', group, variable, date_column, 'SD2_low', 'SD Low')
+        # Upper anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'upper', group, variable, date_column, 'SD2_high', 'SD High')
+        # SD Anomalies
+        fig = add_model_anomalies(fig, group, date_column, variable, 'SD')
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"SD Anomaly Plot Failed: {e}")
+        
+        
+def anomaly_mad_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # MAD Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "MAD Anomaly Detection")
+        # Lower anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'lower', group, variable, date_column, 'MAD_low', 'MAD Low')
+        # Upper anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'upper', group, variable, date_column, 'MAD_high', 'MAD High')
+        # MAD Anomalies
+        fig = add_model_anomalies(fig, group, date_column, variable, 'MAD')
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"MAD Anomaly Plot Failed: {e}")
+
+
+def anomaly_iqr_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # IQR Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "IQR Anomaly Detection")
+        # Lower anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'lower', group, variable, date_column, 'IQR_low', 'IQR Low')
+        # Upper anomaly region shading and threshold line
+        fig = add_anomaly_region(fig, 'upper', group, variable, date_column, 'IQR_high', 'IQR High')
+        # IQR Anomalies
+        fig = add_model_anomalies(fig, group, date_column, variable, 'IQR')
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"IQR Anomaly Plot Failed: {e}")
+
+
+def anomaly_ewma_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # EWMA Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "EWMA Anomaly Detection")
+        # EWMA Forecast
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['EWMA_forecast'],
+            mode='lines',
+            line=dict(color='slateblue', width=0.5, dash='dashdot'),
+            name='EWMA Forecast',
+            showlegend=True
+            ))
+        # EWMA low line
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['EWMA_low'],
+            mode='lines',
+            line=dict(color='orangered', width=1, dash='dashdot'),
+            name='EWMA Low',
+            showlegend=False
+            ))
+        # Lower Shading
+        fig.add_trace(go.Scatter(
+            x=group[group['EWMA_low'].isna()==False][date_column],
+            y=[0] * len(group[group['EWMA_low'].isna()==False]),
+            mode="lines",
+            line=dict(width=0),
+            hoverinfo="skip",
+            showlegend=False,
+            fill="tonexty",
+            fillcolor="rgba(255, 0, 0, 0.07)"
+        ))
+        # EWMA high line
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['EWMA_high'],
+            mode='lines',
+            line=dict(color='orangered', width=1, dash='dashdot'),
+            name='EWMA High',
+            showlegend=False,
+            ))
+        # Upper Shading
+        fig.add_trace(go.Scatter(
+            x=group[group['EWMA_high'].isna()==False][date_column],
+            y=[group[variable].max() * 1.06] * len(group[group['EWMA_high'].isna()==False]),
+            mode="lines",
+            line=dict(width=0),
+            hoverinfo="skip",
+            showlegend=False,
+            fill="tonexty",
+            fillcolor="rgba(255, 0, 0, 0.07)"
+            ))
+        # EWMA Anomalies
+        fig.add_trace(go.Scatter(
+            x=group[group['is_EWMA_anomaly'] == True][date_column],
+            y=group[group['is_EWMA_anomaly'] == True][variable],
+            mode='markers',
+            marker=dict(color='palevioletred', symbol='circle', line=dict(width=1), size=9),
+            name='EWMA Anomalies',
+            ))
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"EWMA Anomaly Plot Failed: {e}")
+
+
+def anomaly_fb_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # FB Prophet Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "FB Prophet Anomaly Detection")    
+        # FB Forecast
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['FB_forecast'],
+            mode='lines',
+            line=dict(color='slateblue', width=0.5, dash='dashdot'),
+            name='FB Forecast',
+            showlegend=True
+            ))
+        # Lower FB line
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['FB_low'],
+            mode='lines',
+            line=dict(color='orangered', width=1, dash='dashdot'),
+            name='FB Low',
+            showlegend=False
+            ))
+        # Lower Shading
+        fig.add_trace(go.Scatter(
+            x=group[group['FB_low'].isna()==False][date_column],
+            y=[0] * len(group[group['FB_low'].isna()==False]),
+            mode="lines",
+            line=dict(width=0),
+            hoverinfo="skip",
+            showlegend=False,
+            fill="tonexty",
+            fillcolor="rgba(255, 0, 0, 0.07)"
+        ))
+        # Upper FB line
+        fig.add_trace(go.Scatter(
+            x=group[date_column],
+            y=group['FB_high'],
+            mode='lines',
+            line=dict(color='orangered', width=1, dash='dashdot'),
+            name='FB High',
+            showlegend=False,
+            ))
+        # Upper Shading
+        fig.add_trace(go.Scatter(
+            x=group[group['FB_high'].isna()==False][date_column],
+            y=[group[variable].max() * 1.06] * len(group[group['FB_high'].isna()==False]),
+            mode="lines",
+            line=dict(width=0),
+            hoverinfo="skip",
+            showlegend=False,
+            fill="tonexty",
+            fillcolor="rgba(255, 0, 0, 0.07)"
+            ))
+        # FB Anomalies
+        fig.add_trace(go.Scatter(
+            x=group[group['is_FB_anomaly'] == True][date_column],
+            y=group[group['is_FB_anomaly'] == True][variable],
+            mode='markers',
+            marker=dict(color='palevioletred', symbol='circle', line=dict(width=1), size=9),
+            name='FB Anomalies',
+            ))
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"EWMA Anomaly Plot Failed: {e}")
+
+
+def anomaly_dbscan_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # DBSCAN Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "DBSCAN Anomaly Detection")
+        # Evaluation Period
+        if eval_period >= 1:
+            fig = add_eval_period_highlight(fig, group, date_column, variable, eval_period)
+        # DBSCAN Anomalies
+        fig.add_trace(go.Scatter(
+            x=group[group['is_DBSCAN_anomaly'] == True][date_column],
+            y=group[group['is_DBSCAN_anomaly'] == True][variable],
+            mode='markers',
+            marker=dict(color='mediumorchid', symbol='circle', line=dict(width=1), size=7),
+            name='DBSCAN Anomalies',
+            ))
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"DBSCAN Anomaly Plot Failed: {e}")
+
+
+def anomaly_isolation_forest_timeseries_plot(group, group_columns, variable, date_column, final_anomalies=True, eval_period=12):
+    # Isolation Forest Model Plot
+    try:
+        group = group.copy()
+        fig = initialize_fig(group, group_columns, variable, date_column, "Isolation Forest Anomaly Detection")
+        # Evaluation Period
+        if eval_period >= 1:
+            fig = add_eval_period_highlight(fig, group, date_column, variable, eval_period)
+        # Isolation Forest Anomalies
+        fig.add_trace(go.Scatter(
+            x=group[group['is_IsolationForest_anomaly_timeseries'] == True][date_column],
+            y=group[group['is_IsolationForest_anomaly_timeseries'] == True][variable],
+            mode='markers',
+            marker=dict(color='mediumorchid', symbol='circle', line=dict(width=1), size=7),
+            name='Isolation Forest Anomalies',
+            ))
+        # Add anomalies markers
+        if final_anomalies:
+            fig = add_anomalies(fig, group, date_column, variable)
+        fig.show()
+        print("\n")
+    except Exception as e:
+        print(f"Isolation Forest Time Series Anomaly Plot Failed: {e}")

anomaly_pipeline/helpers/ewma.py

@@ -0,0 +1,119 @@
+import pandas as pd
+import numpy as np
+import statistics
+
+# # EWMA functions
+
+def ewma_forecast(train, alpha):
+    """Return last EWMA forecast value based on training data."""
+    ewma = train.ewm(alpha=alpha, adjust=False).mean()
+    return ewma.iloc[-1]
+
+
+"""
+def ew_std(series, alpha):
+    
+    Compute exponentially weighted standard deviation.
+    Uses the same alpha as EWMA so recent points get more weight.
+
+    Parameters
+    ----------
+    series : pandas Series of actual values
+    alpha : float in (0,1)
+
+    Returns
+    -------
+    float : exponentially weighted std deviation
+    # mean
+    
+    mean = series.mean()
+    #print(mean)
+
+    # Squared deviation from the mean
+    squared_diff = (series - mean) ** 2
+    
+    #print(squared_diff)
+
+    # EWMA of squared deviation → variance
+    ewma_var = squared_diff.ewm(alpha=alpha, adjust=False).mean()
+    #print(ewma_var)
+    #print(ewma_var.iloc[-1])
+
+    # Std = sqrt(var)
+    return np.sqrt(ewma_var.iloc[-1]) """
+
+
+def ewma_with_anomalies_rolling_group(group, group_columns, variable, date_column, alpha, sigma, eval_period):
+    
+    """
+    Rolling (expanding window) EWMA anomaly detection for a SINGLE GROUP ONLY.
+    Expects `group` to already be filtered to one group.
+    """
+
+    group = group.sort_values(date_column).reset_index(drop=True)
+    n = len(group)
+
+    train_size = n - eval_period  # rolling split
+
+    # Build group key dictionary
+    # group_columns can be list of multiple cols
+    key_dict = {col: group[col].iloc[0] for col in group_columns}
+
+    results = []
+
+    for i in range(train_size, n):
+
+        train = group.loc[:i-1, variable].astype(float)
+        test_value = group.loc[i, variable]
+
+        # --- EWMA + weighted STD ---
+        ewma_train = train.ewm(alpha=alpha, adjust=False).mean()
+        #last_std = ew_std(train, alpha)
+        last_std = np.std(train)
+        forecast = ewma_forecast(train, alpha)
+
+        upper_limit = forecast + sigma * last_std
+        lower_limit = max(forecast - sigma * last_std, 0)
+
+        anomaly = True if (test_value > upper_limit or test_value < lower_limit) else False
+
+        # TRAIN part (added only once)
+        if i == train_size:
+            train_part = pd.concat([
+                group.loc[:i-1, group_columns].reset_index(drop=True),
+                pd.DataFrame({
+                    date_column: group.loc[:i-1, date_column].values,
+                    "alpha": alpha,
+                    "sigma":sigma,
+                    "EWMA_forecast": ewma_train.values,
+                    "STD": last_std,
+                    "EWMA_high": np.nan,
+                    "EWMA_low": np.nan,
+                    "set": "TRAIN",
+                    "is_EWMA_anomaly": pd.NA,
+                })
+            ], axis=1)
+
+            results.append(train_part)
+
+        # TEST row
+        test_part = pd.DataFrame({
+            **{col: [key_dict[col]] for col in key_dict},
+            date_column: [pd.to_datetime(group.loc[i, date_column])],
+            "alpha": [alpha],
+            "sigma":[sigma],
+            "EWMA_forecast": [forecast],
+            "STD": [last_std],
+            "EWMA_high": [upper_limit],
+            "EWMA_low": [lower_limit],
+            "set": ["TEST"],
+            "is_EWMA_anomaly": [anomaly],
+        })
+
+        results.append(test_part)
+
+    final_output = pd.concat(results, ignore_index=True)
+    # Type Safety Check: Ensure the date column is always datetime before returning
+    final_output[date_column] = pd.to_datetime(final_output[date_column])
+    return final_output
+