sequenzo 0.1.17__cp39-cp39-win_amd64.whl → 0.1.18__cp39-cp39-win_amd64.whl

sequenzo/with_event_history_analysis/sequence_history_analysis.py

@@ -0,0 +1,283 @@
+"""
+@Author  : Yuqi Liang 梁彧祺
+@File    : sequence_history_analysis.py
+@Time    : 30/09/2025 21:08
+@Desc    : Sequence History Analysis - Convert person-level sequence data to person-period format
+"""
+
+import numpy as np
+import pandas as pd
+
+
+def person_level_to_person_period(data, id_col="id", period_col="time", event_col="event"):
+    """
+    Convert person-level data to person-period format.
+    
+    This function expands each person's single row into multiple rows,
+    one for each time period they are observed.
+    
+    Parameters
+    ----------
+    data : pandas.DataFrame
+        Input data with one row per person
+    id_col : str, optional
+        Name of the ID column (default: "id")
+    period_col : str, optional
+        Name of the time period column (default: "time")
+    event_col : str, optional
+        Name of the event indicator column (default: "event")
+    
+    Returns
+    -------
+    pandas.DataFrame
+        Expanded data with one row per person-period
+    
+    Examples
+    --------
+    >>> data = pd.DataFrame({'id': [1, 2], 'time': [3, 2], 'event': [True, False]})
+    >>> person_level_to_person_period(data)
+       id  time  event
+    0   1     1  False
+    1   1     2  False
+    2   1     3   True
+    3   2     1  False
+    4   2     2  False
+    """
+    # Check for missing values in critical columns
+    if data[[id_col, period_col, event_col]].isna().any().any():
+        raise ValueError("Cannot handle missing data in the time or event variables")
+    
+    # Create an index that repeats each row based on the time value
+    # For example, if time=3, that row will be repeated 3 times
+    index = np.repeat(np.arange(len(data)), data[period_col].values)
+    
+    # Find the cumulative sum to identify which rows should have the event
+    idmax = np.cumsum(data[period_col].values) - 1
+    
+    # Expand the data by repeating rows
+    dat = data.iloc[index].copy()
+    dat.reset_index(drop=True, inplace=True)
+    
+    # Create sequential time periods for each ID (1, 2, 3, ...)
+    dat[period_col] = dat.groupby(id_col).cumcount() + 1
+    
+    # Set all events to False initially
+    dat[event_col] = False
+    
+    # Set events to True only at the final period for each person
+    # Convert to bool to avoid dtype incompatibility warning
+    dat.loc[idmax, event_col] = data[event_col].values.astype(bool)
+    
+    return dat
+
+
+def _extract_sequence_dataframe(seqdata):
+    """
+    Extract sequence DataFrame from various input types.
+    
+    Parameters
+    ----------
+    seqdata : SequenceData, pandas.DataFrame, or numpy.ndarray
+        Input sequence data
+    
+    Returns
+    -------
+    pandas.DataFrame
+        Sequence data as a DataFrame
+    """
+    # Check if input is a SequenceData object
+    if hasattr(seqdata, 'seqdata'):
+        # This is a SequenceData object
+        return seqdata.seqdata.copy()
+    elif isinstance(seqdata, pd.DataFrame):
+        return seqdata.copy()
+    else:
+        # Assume it's array-like
+        return pd.DataFrame(seqdata)
+
+
+def seqsha(seqdata, time, event, include_present=False, align_end=False, covar=None):
+    """
+    Sequence History Analysis: Create person-period format with sequence history.
+    
+    This function converts sequence data into a person-period format where each
+    row represents a time point for a person, with columns showing their sequence
+    history up to that point.
+    
+    Parameters
+    ----------
+    seqdata : SequenceData, pandas.DataFrame, or numpy.ndarray
+        Sequence data where each row is a person and each column is a time point.
+        Can be a SequenceData object, DataFrame, or array.
+    time : array-like
+        Duration or time until event for each person. Length should equal the 
+        number of sequences. Each value indicates how many time periods that 
+        person is observed. For example, if all persons are observed for the 
+        full sequence length, use: np.full(n_persons, sequence_length)
+    event : array-like
+        Event indicator for each person (True/False or 1/0). Length should 
+        equal the number of sequences.
+    include_present : bool, optional
+        If True, include the current time point in the history (default: False)
+        If False, only include past time points (recommended for most analyses)
+    align_end : bool, optional
+        If True, align sequences from the end (right-aligned) (default: False)
+        If False, align sequences from the start (left-aligned)
+    covar : pandas.DataFrame or numpy.ndarray, optional
+        Additional covariates to merge with the output (default: None)
+        Should have the same number of rows as seqdata
+    
+    Returns
+    -------
+    pandas.DataFrame
+        Person-period data with the following columns:
+        - id: Person identifier
+        - time: Time period within person
+        - event: Event indicator (True only at the final period for each person)
+        - Sequence history columns (varies based on align_end parameter)
+        - Additional covariate columns (if covar is provided)
+    
+    Raises
+    ------
+    ValueError
+        If maximum time exceeds the length of the longest sequence
+    
+    Examples
+    --------
+    Example 1: Basic usage with DataFrame
+    >>> import pandas as pd
+    >>> import numpy as np
+    >>> seqdata = pd.DataFrame([[1, 2, 3, 4], [1, 1, 2, 2]])
+    >>> time = np.array([3, 2])
+    >>> event = np.array([True, False])
+    >>> result = seqsha(seqdata, time, event)
+    
+    Example 2: Usage with SequenceData object (recommended)
+    >>> from sequenzo import SequenceData, load_dataset
+    >>> df = load_dataset('pairfam_family')
+    >>> time_cols = [str(i) for i in range(1, 265)]
+    >>> seq_data = SequenceData(df, time=time_cols, id_col='id', 
+    ...                          states=list(range(1, 10)))
+    >>> # All persons observed for 264 months
+    >>> time = np.full(len(df), 264)
+    >>> event = df['highschool'].values
+    >>> result = seqsha(seq_data, time, event)
+    
+    Example 3: With covariates
+    >>> covar = df[['sex', 'yeduc', 'east']]
+    >>> result = seqsha(seq_data, time, event, covar=covar)
+    
+    Example 4: Right-aligned sequences
+    >>> result = seqsha(seq_data, time, event, align_end=True)
+    
+    Notes
+    -----
+    - The time parameter represents observation duration, not calendar time
+    - When include_present=False (default), only past states are included
+    - Use align_end=True when analyzing sequences leading up to an event
+    - Missing values in the original sequence are converted to "NA_orig"
+    """
+    # Extract sequence DataFrame from input (handles SequenceData, DataFrame, or array)
+    seq_df = _extract_sequence_dataframe(seqdata)
+    
+    # Convert time and event to numpy arrays for consistency
+    time_array = np.asarray(time)
+    event_array = np.asarray(event)
+    
+    # Check that dimensions match
+    n_sequences = len(seq_df)
+    if len(time_array) != n_sequences:
+        raise ValueError(
+            f"Length of 'time' ({len(time_array)}) must match number of sequences ({n_sequences})"
+        )
+    if len(event_array) != n_sequences:
+        raise ValueError(
+            f"Length of 'event' ({len(event_array)}) must match number of sequences ({n_sequences})"
+        )
+    
+    # Create base time data: one row per person with their time and event
+    basetime = pd.DataFrame({
+        'id': np.arange(1, n_sequences + 1),
+        'time': time_array,
+        'event': event_array
+    })
+    
+    # Convert to person-period format (expand rows)
+    persper = person_level_to_person_period(basetime, "id", "time", "event")
+    
+    # Convert sequence data to matrix and handle missing values
+    sdata = seq_df.values.astype(str)
+    sdata[pd.isna(seq_df.values)] = "NA_orig"
+    
+    # Get the time periods for each row in person-period data
+    age = persper['time'].values
+    ma = int(np.max(age))
+    
+    # Check if time values are valid
+    if ma > seq_df.shape[1]:
+        raise ValueError("Maximum time of event occurrence is higher than the longest sequence!")
+    
+    # Create empty matrix to store past sequence states
+    past = np.full((len(persper), seq_df.shape[1]), np.nan, dtype=object)
+    
+    if align_end:
+        # Right-align the sequences (align from the end)
+        start = 1 if include_present else 2
+        
+        for aa in range(start, ma + 1):
+            # Find rows where time equals aa
+            cond = age == aa
+            # Get the person IDs for these rows
+            ids_a = persper.loc[cond, 'id'].values - 1  # Subtract 1 for 0-based indexing
+            
+            if include_present:
+                # Include current time point: fill from (ncol-aa) to end
+                past[cond, (seq_df.shape[1] - aa):seq_df.shape[1]] = sdata[ids_a, 0:aa]
+            else:
+                # Exclude current time point: fill from (ncol-aa+1) to end
+                past[cond, (seq_df.shape[1] - aa + 1):seq_df.shape[1]] = sdata[ids_a, 0:(aa - 1)]
+        
+        # Create column names counting backwards
+        col_names = [f"Tm{i}" for i in range(seq_df.shape[1], 0, -1)]
+    else:
+        # Left-align the sequences (align from the start)
+        for aa in range(1, ma + 1):
+            if include_present:
+                # Include present: use time > aa
+                cond = age > aa
+            else:
+                # Exclude present: use time >= aa
+                cond = age >= aa
+            
+            # Get the person IDs for these rows
+            ids_a = persper.loc[cond, 'id'].values - 1  # Subtract 1 for 0-based indexing
+            
+            # Fill in the sequence state at position aa-1 (0-based)
+            past[cond, aa - 1] = sdata[ids_a, aa - 1]
+        
+        # Use original column names or create default ones
+        if seq_df.columns is not None and len(seq_df.columns) > 0:
+            col_names = [str(col) for col in seq_df.columns[:ma]]
+            # Pad with additional column names if needed
+            col_names += [f"col_{i}" for i in range(ma, seq_df.shape[1])]
+        else:
+            col_names = [f"col_{i}" for i in range(seq_df.shape[1])]
+    
+    # Convert past matrix to DataFrame
+    past_df = pd.DataFrame(past, columns=col_names)
+    
+    # Combine person-period data with sequence history
+    alldata = pd.concat([persper.reset_index(drop=True), past_df], axis=1)
+    
+    # Add covariates if provided
+    if covar is not None:
+        # Merge covariates based on the ID (subtract 1 for 0-based indexing)
+        if isinstance(covar, pd.DataFrame):
+            covar_subset = covar.iloc[alldata['id'].values - 1].reset_index(drop=True)
+            alldata = pd.concat([alldata, covar_subset], axis=1)
+        else:
+            covar_array = np.array(covar)
+            covar_subset = covar_array[alldata['id'].values - 1]
+            alldata = pd.concat([alldata, pd.DataFrame(covar_subset)], axis=1)
+    
+    return alldata

{sequenzo-0.1.17.dist-info → sequenzo-0.1.18.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sequenzo
-Version: 0.1.17
+Version: 0.1.18
 Summary: A fast, scalable and intuitive Python package for social sequence analysis.
 Author-email: Yuqi Liang <yuqi.liang.1900@gmail.com>, Xinyi Li <1836724126@qq.com>, Jan Heinrich Ernst Meyerhoff-Liang <jan.meyerhoff1@gmail.com>
 License: BSD 3-Clause License
@@ -58,6 +58,7 @@ Requires-Dist: joblib>=1.0.1
 Requires-Dist: docutils>=0.17
 Requires-Dist: tqdm<5.0.0,>=4.62.3
 Requires-Dist: missingno<0.6.0,>=0.5.2
+Requires-Dist: rpy2
 Provides-Extra: dev
 Requires-Dist: pytest>=6.2.5; extra == "dev"
 Requires-Dist: flake8>=3.9.2; extra == "dev"
@@ -123,12 +124,12 @@ Perfect for research, policy, and business, enabling seamless analysis of catego
 
 Sequenzo provides pre-built Python wheels for maximum compatibility — no need to compile from source.
 
-| Platform         | Architecture                  | Python Versions      | Status            |
+| Platform         | Architecture                  | Python Versions       | Status            |
 |------------------|-------------------------------|-----------------------|-------------------|
-| **macOS**        | `universal2` (Intel + Apple Silicon) | 3.9, 3.11       | ✅ Pre-built wheel |
-| **Windows**      | `AMD64` (64-bit)              | 3.9, 3.10, 3.11       | ✅ Pre-built wheel |
-| **Linux (glibc)**| `x86_64` (standard Linux)     | 3.9, 3.10, 3.11       | ✅ Pre-built wheel |
-| **Linux (musl)** | `x86_64` (Alpine Linux)       | 3.9, 3.10, 3.11       | ✅ Pre-built wheel |
+| **macOS**        | `universal2` (Intel + Apple Silicon) | 3.9, 3.10, 3.11, 3.12 | ✅ Pre-built wheel |
+| **Windows**      | `AMD64` (64-bit)              | 3.9, 3.10, 3.11, 3.12 | ✅ Pre-built wheel |
+| **Linux (glibc)**| `x86_64` (standard Linux)     | 3.9, 3.10, 3.11, 3.12 | ✅ Pre-built wheel |
+| **Linux (musl)** | `x86_64` (Alpine Linux)       | 3.9, 3.10, 3.11, 3.12 | ✅ Pre-built wheel |
 
 
 What do these terms mean?