sequenzo 0.1.21__cp310-cp310-macosx_10_9_x86_64.whl

sequenzo/visualization/plot_mean_time.py

@@ -0,0 +1,194 @@
+"""
+@Author  : Yuqi Liang 梁彧祺
+@File    : plot_mean_time.py
+@Time    : 14/02/2025 10:12
+@Desc    :
+    Implementation of Mean Time Plot for social sequence analysis,
+    closely following ggseqplot's `ggseqmtplot` function,
+    and TraMineR's `plot.stslist.meant.Rd` for mean time calculation.
+"""
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+from typing import Optional
+from sequenzo.define_sequence_data import SequenceData
+from sequenzo.visualization.utils import (
+    save_and_show_results,
+    show_plot_title
+)
+
+
+def _compute_mean_time(seqdata: SequenceData, weights="auto") -> pd.DataFrame:
+    """
+    Compute mean total time spent in each state across all sequences.
+    Optimized version using pandas operations.
+
+    :param seqdata: SequenceData object containing sequence information
+    :param weights: (np.ndarray or "auto") Weights for sequences. If "auto", uses seqdata.weights if available
+    :return: DataFrame with mean time spent and standard error for each state
+    """
+    # Process weights
+    if isinstance(weights, str) and weights == "auto":
+        weights = getattr(seqdata, "weights", None)
+    
+    if weights is not None:
+        weights = np.asarray(weights, dtype=float).reshape(-1)
+        if len(weights) != len(seqdata.values):
+            raise ValueError("Length of weights must equal number of sequences.")
+    
+    # Get data and preprocess
+    seq_df = seqdata.to_dataframe()
+    inv = {v: k for k, v in seqdata.state_mapping.items()}
+    states = list(range(1, len(seqdata.states) + 1))  # Use numerical state indices
+    n = len(seq_df)
+
+    # Get weights
+    if weights is None:
+        w = np.ones(n)
+    else:
+        w = np.asarray(weights, dtype=float)
+
+    # Broadcast weights to each time point
+    W = np.repeat(w[:, None], seq_df.shape[1], axis=1)
+
+    # Convert to long format with weights
+    df_long = seq_df.melt(value_name='state_idx')
+    # Replicate weights for each time point
+    W_long = pd.DataFrame(W, columns=seq_df.columns).melt(value_name='w')['w'].to_numpy()
+    df_long['w'] = W_long
+    # Keep state_idx as numerical for consistent grouping
+    
+    # Calculate mean time spent in each state per sequence
+    # For each sequence, count time spent in each state, then take weighted average
+    seq_state_times = {}
+    for s in states:
+        # Count occurrences of state s in each sequence
+        state_counts = (seq_df == s).sum(axis=1)  # Sum across time for each sequence
+        # Calculate weighted mean across sequences
+        seq_state_times[s] = np.average(state_counts, weights=w) if len(state_counts) > 0 else 0.0
+    
+    mean_times = seq_state_times
+
+    # Calculate standard errors for mean time
+    se = {}
+    n_sequences = len(seq_df)
+    
+    for s in states:
+        if n_sequences > 1:
+            # Count occurrences of state s in each sequence
+            state_counts = (seq_df == s).sum(axis=1)
+            # Calculate weighted standard error
+            weighted_mean = seq_state_times[s]
+            weighted_var = np.average((state_counts - weighted_mean) ** 2, weights=w)
+            # Standard error of the weighted mean
+            se[s] = np.sqrt(weighted_var / n_sequences) if weighted_var >= 0 else 0.0
+        else:
+            se[s] = 0.0
+
+    # Create result DataFrame
+    mean_time_df = pd.DataFrame({
+        'State': [inv[s] for s in states],
+        'MeanTime': [mean_times[s] for s in states],
+        'StandardError': [se[s] for s in states]
+    })
+
+    mean_time_df.sort_values(by='MeanTime', ascending=True, inplace=True)
+
+    return mean_time_df
+
+
+def plot_mean_time(seqdata: SequenceData,
+                   weights="auto",
+                   show_error_bar: bool = True,
+                   title=None,
+                   x_label="Mean Time",
+                   y_label="State",
+                   fontsize: int = 12,
+                   save_as: Optional[str] = None,
+                   dpi: int = 200) -> None:
+    """
+    Plot Mean Time Plot for sequence data with clean white background.
+
+    :param seqdata: SequenceData object containing sequence information
+    :param weights: (np.ndarray or "auto") Weights for sequences. If "auto", uses seqdata.weights if available
+    :param show_error_bar: Boolean flag to show or hide error bars
+    :param title: Optional title for the plot
+    :param x_label: Label for the x-axis
+    :param y_label: Label for the y-axis
+    :param save_as: Optional file path to save the plot
+    :param dpi: Resolution of the saved plot
+    """
+    # Use default style as base
+    plt.style.use('default')
+
+    # Compute all required data at once
+    mean_time_df = _compute_mean_time(seqdata, weights)
+
+    # Create figure and preallocate memory
+    fig = plt.figure(figsize=(12, 7))
+
+    # Create main plot
+    ax = plt.subplot(111)
+
+    # Get color mapping - use original colors without enhancement
+    cmap = seqdata.get_colormap()
+    colors = [cmap.colors[i] for i in range(len(seqdata.states))]
+
+    # Assign colors to states (without enhancing saturation)
+    mean_time_df['Color'] = pd.Categorical(mean_time_df['State']).codes
+    mean_time_df['Color'] = mean_time_df['Color'].map(lambda x: colors[x])
+
+    # Create custom barplot
+    for i, (_, row) in enumerate(mean_time_df.iterrows()):
+        ax.barh(y=i, width=row['MeanTime'], height=0.7,
+                color=row['Color'], edgecolor='white', linewidth=0.5)
+
+    # Set y-axis ticks and labels
+    ax.set_yticks(range(len(mean_time_df)))
+    ax.set_yticklabels(mean_time_df['State'], fontsize=fontsize-2)
+
+    # Add error bars if needed
+    if show_error_bar:
+        ax.errorbar(
+            x=mean_time_df["MeanTime"],
+            y=range(len(mean_time_df)),
+            xerr=mean_time_df["StandardError"],
+            fmt='none',
+            ecolor='black',
+            capsize=3,
+            capthick=1,
+            elinewidth=1.5
+        )
+
+    # Set plot properties
+    if title:
+        show_plot_title(ax, title, show=True, fontsize=fontsize+2, fontweight='bold', pad=20)
+    ax.set_xlabel(x_label, fontsize=fontsize)
+    ax.set_ylabel(y_label, fontsize=fontsize, labelpad=15)
+
+    # Clean white background with light grid
+    ax.set_facecolor('white')
+    ax.grid(axis='x', color='#E0E0E0', linestyle='-', linewidth=0.5)
+    ax.set_axisbelow(True)  # Place grid lines behind the bars
+
+    # Customize borders
+    for spine in ax.spines.values():
+        spine.set_color('#CCCCCC')  # Light gray border
+        spine.set_linewidth(0.5)
+
+    # Adjust layout(1/2)
+    plt.subplots_adjust(left=0.3)
+
+    # Add a note about normalization
+    relative_threshold = 0.01
+    max_val = mean_time_df['MeanTime'].max()
+    too_many_small = np.sum(mean_time_df['MeanTime'] < relative_threshold * max_val) >= 1
+    if too_many_small:
+        norm_note = f"Note: Some bars may appear as zero, but actually have small non-zero values."
+        plt.figtext(0.5, -0.02, norm_note, ha='center', fontsize=fontsize-2, style='italic')
+
+    # Adjust layout(2/2)
+    plt.tight_layout()
+
+    save_and_show_results(save_as, dpi=200)
+

sequenzo/visualization/plot_modal_state.py

@@ -0,0 +1,276 @@
+"""
+@Author  : Yuqi Liang 梁彧祺
+@File    : plot_modal_state.py
+@Time    : 01/03/2025 13:45
+@Desc    : 
+"""
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+from typing import Optional, Union, List
+from sequenzo import SequenceData
+from sequenzo.visualization.utils import (
+    set_up_time_labels_for_x_axis,
+    create_standalone_legend,
+    save_figure_to_buffer,
+    combine_plot_with_legend,
+    save_and_show_results,
+    show_plot_title
+)
+from PIL import Image
+
+
+def plot_modal_state(seqdata: SequenceData,
+                     group_by: Optional[Union[str, pd.Series, np.ndarray]] = None,
+                     group_labels: Optional[List[str]] = None,
+                     weights="auto",
+                     xlabel: str = "Time",
+                     ylabel: str = "Rel. Freq.",
+                     fig_width: int = 12,
+                     fig_height: Optional[int] = None,
+                     show_counts: bool = True,
+                     show_group_titles: bool = True,
+                     fontsize: int = 12,
+                     save_as: Optional[str] = None,
+                     dpi: int = 200) -> None:
+    """
+    Creates a modal state frequency plot showing the most frequent state at each position
+    and its relative frequency, with optional grouping by a categorical variable.
+
+    :param seqdata: SequenceData object containing sequence information
+    :param group_by: Column name or array with grouping variable
+    :param group_labels: Optional custom labels for groups
+    :param weights: (np.ndarray or "auto") Weights for sequences. If "auto", uses seqdata.weights if available
+    :param xlabel: Label for the x-axis
+    :param ylabel: Label for the y-axis
+    :param fig_width: Width of the figure
+    :param fig_height: Height of the figure (auto-calculated based on groups if None)
+    :param show_counts: Whether to show the count of sequences in each group title
+    :param save_as: Optional file path to save the plot
+    :param dpi: Resolution of the saved plot
+
+    :return: None
+    """
+    # Process weights
+    if isinstance(weights, str) and weights == "auto":
+        weights = getattr(seqdata, "weights", None)
+    
+    if weights is not None:
+        weights = np.asarray(weights, dtype=float).reshape(-1)
+        if len(weights) != len(seqdata.values):
+            raise ValueError("Length of weights must equal number of sequences.")
+    
+    # Get sequence data as a DataFrame
+    seq_df = seqdata.to_dataframe()
+
+    # Ensure seq_df has the same index as the original data
+    # This is crucial to align the grouping variable with sequence data
+    seq_df.index = seqdata.data.index
+    
+    # Get weights for all sequences
+    if weights is None:
+        w_all = np.ones(len(seq_df))
+    else:
+        w_all = np.asarray(weights)
+
+    # Create state mapping from numerical values back to state names
+    inv_state_mapping = {v: k for k, v in seqdata.state_mapping.items()}
+
+    # Process grouping variable
+    if group_by is None:
+        # If no grouping, create a single group with all sequences
+        groups = pd.Series(["All Sequences"] * len(seq_df), index=seq_df.index)
+        if group_labels is None:
+            group_labels = ["All Sequences"]
+    elif isinstance(group_by, str):
+        # If grouping by column name from original data
+        if group_by not in seqdata.data.columns:
+            raise ValueError(f"Column '{group_by}' not found in sequence data")
+        groups = seqdata.data[group_by]
+        if group_labels is None:
+            group_labels = sorted(groups.unique())
+    else:
+        # If grouping by external array or Series
+        if len(group_by) != len(seq_df):
+            raise ValueError("Length of group_by must match number of sequences")
+        groups = pd.Series(group_by)
+        if group_labels is None:
+            group_labels = sorted(set(groups))
+
+    # Prepare plotting
+    n_groups = len(group_labels)
+    n_time_points = len(seq_df.columns)
+
+    if fig_height is None:
+        # Auto-calculate height based on number of groups
+        fig_height = max(4, 3 * n_groups)
+
+    # TODO: Title is not very pretty here so I decided to remove it.
+    # But here I keep 1 to keep the space big enough for the distance
+    # between the second subplot and the upper first subplot
+    title_height = 1
+    adjusted_fig_height = fig_height + title_height
+
+    # Create main figure with additional space for title
+    main_fig = plt.figure(figsize=(fig_width, adjusted_fig_height))
+
+    # No title, use whole figure for plots
+    plot_gs = main_fig.add_gridspec(nrows=n_groups, height_ratios=[1] * n_groups, hspace=0.3)
+
+    # Create axes for each group
+    axes = []
+    for i in range(n_groups):
+        axes.append(main_fig.add_subplot(plot_gs[i]))
+
+    # Make sure all axes share x and y scales
+    for ax in axes[1:]:
+        ax.sharex(axes[0])
+        ax.sharey(axes[0])
+
+    # Get colors for states
+    colors = seqdata.color_map_by_label
+
+    # Process each group
+    for i, group in enumerate(group_labels):
+        ax = axes[i]
+
+        # Get indices for this group
+        group_indices = groups == group
+        group_count = group_indices.sum()
+
+        # Skip if no sequences in this group
+        if group_count == 0:
+            continue
+
+        # Subset data for this group and get corresponding weights
+        group_data = seq_df[group_indices]
+        w = w_all[group_indices.to_numpy()]
+
+        # Calculate modal states and their frequencies for each time point
+        modal_states = []
+        modal_freqs = []
+
+        for col in group_data.columns:
+            states_idx = group_data[col].to_numpy()
+            
+            # Calculate weighted counts for each state
+            weighted_sum = {}
+            # Use numerical state indices (1, 2, 3, ...) instead of state labels
+            for s_num in range(1, len(seqdata.states) + 1):  # s_num is the integer encoding
+                weighted_sum[s_num] = float(w[states_idx == s_num].sum())
+            
+            totw = float(w.sum())
+            
+            if totw > 0:
+                # Find the state with maximum weighted count
+                modal_s = max(weighted_sum, key=weighted_sum.get)
+                modal_state = inv_state_mapping[modal_s]
+                modal_freq = weighted_sum[modal_s] / totw
+            else:
+                modal_state, modal_freq = None, 0.0
+            
+            modal_states.append(modal_state)
+            modal_freqs.append(modal_freq)
+
+        # Equal width for all bars
+        x = np.arange(n_time_points)
+        bar_width = 0.8  # Fixed width for all bars
+
+        # Create bars with consistent width
+        for j, (state, freq) in enumerate(zip(modal_states, modal_freqs)):
+            if state is not None:
+                # state is already a label from inv_state_mapping
+                ax.bar(x[j], freq, width=bar_width, color=colors[state],
+                       edgecolor='white', linewidth=0.5)
+
+        # Set group title with count if requested
+        if show_group_titles:
+            if show_counts:
+                if weights is not None and not np.allclose(weights, 1.0):
+                    sum_w = float(w.sum())
+                    title_text = f"{group} (n={group_count}, total weight={sum_w:.1f})"
+                else:
+                    title_text = f"{group} (n={group_count})"
+            else:
+                title_text = group
+            show_plot_title(ax, title_text, show=True, fontsize=fontsize, pad=15)
+
+        # Set y-axis limits and ticks
+        ax.set_ylim(0, 1.0)
+        ax.set_yticks([0, 0.25, 0.5, 0.75, 1.0])
+
+        # Set grid and spines
+        ax.grid(axis='y', color='#E0E0E0', linestyle='-', linewidth=0.5)
+        ax.set_axisbelow(True)
+
+        # Clean up borders
+        for spine in ax.spines.values():
+            spine.set_color('#CCCCCC')
+            spine.set_linewidth(0.5)
+
+        # Add y-label only for the middle subplot
+        if i == n_groups // 2:
+            ax.set_ylabel(ylabel, fontsize=fontsize)
+
+    # Set up X-axis (time) labels on the bottom subplot
+    set_up_time_labels_for_x_axis(seqdata, axes[-1])
+    axes[-1].set_xlabel(xlabel, fontsize=fontsize, labelpad=10)
+
+    # Save main figure to memory
+    main_buffer = save_figure_to_buffer(main_fig, dpi=dpi)
+
+    # Create a legend
+    # Create standalone legend
+    legend_buffer = create_standalone_legend(
+        colors=colors,
+        labels=seqdata.labels,
+        ncol=min(5, len(seqdata.states)),
+        figsize=(fig_width, 1),
+        fontsize=fontsize-2,
+        dpi=dpi
+    )
+
+    if save_as and not save_as.lower().endswith(('.png', '.jpg', '.jpeg', '.pdf')):
+        save_as = save_as + '.png'
+
+    # Combine main plot with legend
+    combined_img = combine_plot_with_legend(
+        main_buffer,
+        legend_buffer,
+        output_path=save_as,
+        dpi=dpi,
+        padding=20  # Increased padding between plot and legend
+    )
+
+    # Display combined image
+    plt.figure(figsize=(fig_width, adjusted_fig_height + 1))
+    plt.imshow(combined_img)
+    plt.axis('off')
+    plt.show()
+    plt.close()
+
+
+
+if __name__ == '__main__':
+    # Import necessary libraries
+    from sequenzo import *  # Social sequence analysis
+    import pandas as pd  # Data manipulation
+
+    # List all the available datasets in Sequenzo
+    print('Available datasets in Sequenzo: ', list_datasets())
+
+    # Load the data that we would like to explore in this tutorial
+    # `df` is the short for `dataframe`, which is a common variable name for a dataset
+    df = load_dataset('country_co2_emissions')
+
+    # Create a SequenceData object from the dataset
+
+    # Define the time-span variable
+    time = list(df.columns)[1:]
+
+    states = ['Very Low', 'Low', 'Middle', 'High', 'Very High']
+
+    sequence_data = SequenceData(df, time=time, time_type="year", id_col="country", states=states)
+
+    plot_modal_state(sequence_data)
+

sequenzo/visualization/plot_most_frequent_sequences.py

@@ -0,0 +1,147 @@
+"""
+@Author  : Yuqi Liang 梁彧祺
+@File    : plot_most_frequent_sequences.py
+@Time    : 12/02/2025 10:40
+@Desc    :
+    Generate sequence frequency plots.
+
+    This script plots the 10 most frequent sequences,
+    similar to `seqfplot` in R's TraMineR package.
+"""
+
+import pandas as pd
+import matplotlib.pyplot as plt
+import numpy as np
+
+from sequenzo.define_sequence_data import SequenceData
+from sequenzo.visualization.utils import (
+    set_up_time_labels_for_x_axis,
+    save_and_show_results,
+    show_plot_title
+)
+
+
+def plot_most_frequent_sequences(seqdata: SequenceData, top_n: int = 10, weights="auto", title=None, fontsize=12, save_as=None, dpi=200, show_title: bool = True):
+    """
+    Generate a sequence frequency plot, similar to R's seqfplot.
+
+    :param seqdata: (SequenceData) A SequenceData object containing sequences.
+    :param top_n: (int) Number of most frequent sequences to display.
+    :param weights: (np.ndarray or "auto") Weights for sequences. If "auto", uses seqdata.weights if available
+    :param title: (str, optional) Title for the plot. If None, no title will be displayed.
+    :param fontsize: (int) Base font size for text elements
+    :param save_as: (str, optional) Path to save the plot.
+    :param dpi: (int) Resolution of the saved plot.
+    """
+    sequences = seqdata.values.tolist()
+    
+    # Process weights
+    if isinstance(weights, str) and weights == "auto":
+        weights = getattr(seqdata, "weights", None)
+    
+    if weights is not None:
+        weights = np.asarray(weights, dtype=float).reshape(-1)
+        if len(weights) != len(seqdata.values):
+            raise ValueError("Length of weights must equal number of sequences.")
+    
+    if weights is None:
+        weights = np.ones(len(sequences))
+
+    # Weighted counting of sequences
+    agg = {}
+    for seq, w in zip(sequences, weights):
+        key = tuple(seq)
+        agg[key] = agg.get(key, 0.0) + float(w)
+
+    # Select Top-N by weighted frequency
+    items = sorted(agg.items(), key=lambda kv: kv[1], reverse=True)[:top_n]
+    df = pd.DataFrame(items, columns=['sequence', 'wcount'])
+    totw = float(np.sum(weights))
+    df['freq'] = df['wcount'] / (totw if totw > 0 else 1.0) * 100.0
+
+    # **Ensure colors match seqdef**
+    # Use numeric color map directly to avoid label/state-name mismatches
+    inv_state_mapping = {v: k for k, v in seqdata.state_mapping.items()}  # Reverse mapping kept if needed elsewhere
+
+    # **Plot settings**
+    fig, ax = plt.subplots(figsize=(10, 6))
+
+    # **Adjust y_positions calculation to ensure sequences fill the entire y-axis**
+    y_positions = df['freq'].cumsum() - df['freq'] / 2  # Center the bars
+
+    for i, (seq, freq) in enumerate(zip(df['sequence'], df['freq'])):
+        left = 0  # Starting x position
+        for t, state_idx in enumerate(seq):
+            # Use numeric-coded color map; if unknown, fall back to gray
+            color = seqdata.color_map.get(int(state_idx), "gray")
+
+            width = 1  # Width of each time slice
+            ax.barh(y=y_positions[i], width=width * 1.01, left=left - 0.005,
+                    height=freq, color=color, linewidth=0,
+                    antialiased=False)
+            left += width  # Move to the next time slice
+
+    # **Formatting**
+    ax.set_xlabel("Time", fontsize=fontsize)
+    # Check if we have effective weights (not all 1.0) and they were provided by user
+    original_weights = getattr(seqdata, "weights", None)
+    if original_weights is not None and not np.allclose(original_weights, 1.0):
+        # Show both count and weighted total if weights are used
+        ax.set_ylabel("Cumulative Frequency (%)\nN={:,}, total weight={:.1f}".format(len(sequences), totw), fontsize=fontsize)
+    else:
+        ax.set_ylabel("Cumulative Frequency (%)\nN={:,}".format(len(sequences)), fontsize=fontsize)
+    if show_title and title is not None:
+        show_plot_title(ax, title, show=True, fontsize=fontsize+2, pad=20)
+
+    # **Optimize X-axis ticks: align to the center of each bar**
+    set_up_time_labels_for_x_axis(seqdata, ax)
+
+    # **Set Y-axis ticks and labels**
+    sum_freq_top_10 = df['freq'].sum()  # Cumulative frequency of top 10 sequences
+    max_freq = df['freq'].max()  # Frequency of the top 1 sequence
+
+    # Set Y-axis ticks: 0%, top1 frequency, top10 cumulative frequency
+    y_ticks = [0, max_freq, sum_freq_top_10]
+    ax.set_yticks(y_ticks)
+    ax.set_yticklabels([f"{ytick:.1f}%" for ytick in y_ticks], fontsize=fontsize-2)
+
+    # **Set Y-axis range to ensure the highest tick is the top10 cumulative frequency**
+    # Force Y-axis range to be from 0 to sum_freq_top_10
+    ax.set_ylim(0, sum_freq_top_10)
+
+    # **Annotate the frequency percentage on the left side of the highest frequency sequence**
+    ax.annotate(f"{max_freq:.1f}%", xy=(-0.5, y_positions.iloc[0]),
+                xycoords="data", fontsize=fontsize, color="black", ha="left", va="center")
+
+    # **Annotate 0% at the bottom of the Y-axis**
+    ax.annotate("0%", xy=(-0.5, 0), xycoords="data", fontsize=fontsize, color="black", ha="left", va="center")
+
+    # **Clean up axis aesthetics like plot_state_distribution**
+    ax.spines['top'].set_visible(False)
+    ax.spines['right'].set_visible(False)
+    ax.spines['left'].set_visible(True)  # Keep the left border like state_distribution
+    ax.spines['bottom'].set_visible(True)  # Show bottom border to connect with left
+    
+    # Style the left spine to match plot_state_distribution
+    ax.spines['left'].set_color('gray')
+    ax.spines['left'].set_linewidth(0.7)
+    ax.spines['bottom'].set_color('gray')
+    ax.spines['bottom'].set_linewidth(0.7)
+    
+    # Style the tick parameters
+    ax.tick_params(axis='y', colors='gray', length=4, width=0.7)
+    ax.tick_params(axis='x', colors='gray', length=4, width=0.7)
+    
+    # Extend the left spine slightly beyond the plot area
+    ax.spines['left'].set_bounds(0, sum_freq_top_10)
+    ax.spines['left'].set_position(('outward', 5))  # Move spine 5 points to the left
+    
+    # Align bottom spine with the left spine position
+    ax.spines['bottom'].set_position(('outward', 5))  # Move bottom spine to align with left
+
+    # Use legend from SequenceData
+    ax.legend(*seqdata.get_legend(), bbox_to_anchor=(1.05, 1), loc='upper left')
+
+    save_and_show_results(save_as, dpi=200)
+
+