likelihood 2.2.0.dev1__cp310-cp310-manylinux_2_28_x86_64.whl

likelihood/tools/cat_embed.py

@@ -0,0 +1,212 @@
+import logging
+import os
+from typing import List
+
+import numpy as np
+import pandas as pd
+
+os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
+logging.getLogger("tensorflow").setLevel(logging.ERROR)
+import tensorflow as tf
+from sklearn.preprocessing import LabelEncoder
+
+tf.get_logger().setLevel("ERROR")
+
+
+class CategoricalEmbedder:
+    def __init__(self, embedding_dim=32):
+        self.embedding_dim = embedding_dim
+        self.label_encoders = {}
+        self.embeddings = {}
+
+    def fit(self, df: pd.DataFrame, categorical_cols: List):
+        """
+        Fit the embeddings on the given data.
+
+        Parameters
+        ----------
+        df : `pd.DataFrame`
+            Pandas DataFrame containing the tabular data.
+        categorical_cols : `List`
+            List of column names representing categorical features.
+
+        Returns
+        -------
+        `None`
+        """
+        df_processed = df.copy()
+        for col in categorical_cols:
+            if col not in df_processed.columns:
+                raise ValueError(f"Column {col} not found in DataFrame")
+
+        for col in categorical_cols:
+            mode_val = df_processed[col].mode()
+            if not mode_val.empty:
+                df_processed[col] = df_processed[col].fillna(mode_val[0])
+
+        for col in categorical_cols:
+            le = LabelEncoder()
+            df_processed[col] = le.fit_transform(df_processed[col])
+            self.label_encoders[col] = le
+
+            vocab_size = len(le.classes_)
+            embedding_matrix = np.random.rand(vocab_size, self.embedding_dim)
+            self.embeddings[col] = tf.Variable(embedding_matrix, dtype=tf.float32)
+
+    def transform(self, df: pd.DataFrame, categorical_cols: List[str]):
+        """
+        Transform the data using the fitted embeddings.
+
+        Parameters
+        ----------
+        df : `pd.DataFrame`
+            Pandas DataFrame containing the tabular data.
+        categorical_cols : `List[str]`
+            List of column names representing categorical features.
+
+        Returns
+        -------
+        Transformed Pandas DataFrame with original columns except `categorical_cols` replaced by their embedding representations.
+        """
+
+        df_processed = df.copy()
+
+        for col in categorical_cols:
+            if col not in self.label_encoders:
+                raise ValueError(
+                    f"Column {col} has not been fitted. Please call fit() on this column first."
+                )
+            mode_val = df_processed[col].mode()
+            if not mode_val.empty:
+                df_processed[col] = df_processed[col].fillna(mode_val[0])
+            le = self.label_encoders[col]
+            df_processed[col] = le.transform(df_processed[col])
+
+        for col in categorical_cols:
+            indices_tensor = tf.constant(df_processed[col], dtype=tf.int32)
+            embedding_layer = tf.nn.embedding_lookup(
+                params=self.embeddings[col], ids=indices_tensor
+            )
+            if len(embedding_layer.shape) == 1:
+                embedding_layer = tf.expand_dims(embedding_layer, axis=0)
+
+            for i in range(self.embedding_dim):
+                df_processed[f"{col}_embed_{i}"] = embedding_layer[:, i]
+            df_processed.drop(columns=[col], inplace=True)
+
+        return df_processed
+
+    def inverse_transform(self, df: pd.DataFrame, categorical_cols: List[str]):
+        """
+        Inverse transform the data using the fitted embeddings.
+
+        Parameters
+        ----------
+        df : `pd.DataFrame`
+            Pandas DataFrame containing the tabular data with embedded representations.
+        categorical_cols : `List[str]`
+            List of column names representing categorical features.
+
+        Returns
+        -------
+        Transformed Pandas DataFrame with original columns replaced by their categorical labels.
+        """
+
+        df_processed = df.copy()
+
+        for col in categorical_cols:
+            if col not in self.label_encoders:
+                raise ValueError(
+                    f"Column {col} has not been fitted. Please call fit() on this column first."
+                )
+
+            embedding_matrix = self.embeddings[col].numpy()
+            label_encoder = self.label_encoders[col]
+
+            embedded_columns = [f"{col}_embed_{i}" for i in range(self.embedding_dim)]
+            embeddings = df_processed[embedded_columns].values
+
+            distances = np.linalg.norm(embedding_matrix - embeddings[:, np.newaxis], axis=2)
+            original_indices = np.argmin(distances, axis=1)
+            original_labels = label_encoder.inverse_transform(original_indices)
+
+            df_processed[col] = original_labels
+            df_processed.drop(columns=embedded_columns, inplace=True)
+
+        return df_processed
+
+    def save_embeddings(self, path: str):
+        """
+        Save the embeddings to a directory.
+
+        Parameters
+        ----------
+        path : `str`
+            Path to the directory where embeddings will be saved.
+        """
+
+        os.makedirs(path, exist_ok=True)
+        for col, embedding in self.embeddings.items():
+            np.save(os.path.join(path, f"{col}_embedding.npy"), embedding.numpy())
+
+    def load_embeddings(self, path: str):
+        """
+        Load the embeddings from a directory.
+
+        Parameters
+        ----------
+        path : `str`
+            Path to the directory where embeddings are saved.
+        """
+
+        for col in self.label_encoders.keys():
+            embedding_path = os.path.join(path, f"{col}_embedding.npy")
+            if not os.path.exists(embedding_path):
+                raise FileNotFoundError(f"Embedding file {embedding_path} not found.")
+            embedding_matrix = np.load(embedding_path)
+            self.embeddings[col] = tf.Variable(embedding_matrix, dtype=tf.float32)
+
+
+if __name__ == "__main__":
+    data = {
+        "color": ["red", "blue", None, "green", "blue"],
+        "size": ["S", "M", "XL", "XS", None],
+        "price": [10.99, 25.50, 30.00, 8.75, 12.25],
+    }
+    df = pd.DataFrame(data)
+
+    # Initialize the embedder
+    embedder = CategoricalEmbedder(embedding_dim=3)
+
+    # Fit the embeddings on the data
+    embedder.fit(df, categorical_cols=["color", "size"])
+
+    # Transform the data using the fitted embeddings
+    processed_df = embedder.transform(df, categorical_cols=["color", "size"])
+
+    print("Processed DataFrame:")
+    print(processed_df.head())
+
+    # Save the embeddings to disk
+    embedder.save_embeddings("./embeddings")
+
+    # Load the embeddings from disk
+    new_embedder = CategoricalEmbedder(embedding_dim=3)
+    new_embedder.label_encoders = (
+        embedder.label_encoders
+    )  # Assuming label encodings are consistent across runs
+    new_embedder.load_embeddings("./embeddings")
+
+    # Transform the data using the loaded embeddings
+    processed_df_loaded = new_embedder.transform(df, categorical_cols=["color", "size"])
+    print("\nProcessed DataFrame with Loaded Embeddings:")
+    print(processed_df_loaded.head())
+
+    # Inverse transform the data
+    df_loaded = new_embedder.inverse_transform(
+        processed_df_loaded, categorical_cols=["color", "size"]
+    )
+    print("\nOriginal DataFrame:")
+    print(df.head())
+    print("\nProcessed DataFrame with Inverse Transform:")
+    print(df_loaded.head())

likelihood/tools/figures.py

@@ -0,0 +1,348 @@
+import os
+import warnings
+from typing import Optional
+
+import matplotlib.pyplot as plt
+import numpy as np
+from matplotlib.ticker import AutoMinorLocator
+from scipy import stats
+
+plt.rcParams.update({"font.size": 14})
+
+
+def act_pred(
+    y_act: np.ndarray,
+    y_pred: np.ndarray,
+    name: str = "example",
+    x_hist: bool = True,
+    y_hist: bool = True,
+    reg_line: bool = True,
+    save_dir: Optional[str] = None,
+) -> None:
+    """
+    Creates a scatter plot of actual vs predicted values along with histograms and a regression line.
+
+    Parameters
+    ----------
+    y_act : `np.ndarray`
+        The actual values (ground truth) as a 1D numpy array.
+    y_pred : `np.ndarray`
+        The predicted values as a 1D numpy array.
+    name : `str`, optional
+        The name for saving the plot. Default is "example".
+    x_hist : `bool`, optional
+        Whether to display the histogram for the actual values (y_act). Default is True.
+    y_hist : `bool`, optional
+        Whether to display the histogram for the predicted values (y_pred). Default is True.
+    reg_line : `bool`, optional
+        Whether to plot a regression line (best-fit line) in the scatter plot. Default is True.
+    save_dir : `Optional[str]`, optional
+        The directory to save the figure. If None, the figure will not be saved. Default is None.
+
+    Returns
+    -------
+    `None` : The function doesn't return anything. It generates and optionally saves a plot.
+    """
+
+    y_pred, y_act = y_pred.flatten(), y_act.flatten()
+
+    if not isinstance(y_act, np.ndarray) or not isinstance(y_pred, np.ndarray):
+        raise ValueError("y_act and y_pred must be numpy arrays.")
+    if y_act.shape != y_pred.shape:
+        raise ValueError("y_act and y_pred must have the same shape.")
+
+    mec = "#2F4F4F"
+    mfc = "#C0C0C0"
+
+    fig = plt.figure(figsize=(6, 6))
+
+    left, width = 0.1, 0.65
+    bottom, height = 0.1, 0.65
+    bottom_h = left + width
+    left_h = left + width + 0.05
+
+    ax2 = fig.add_axes([left, bottom, width, height])
+    ax2.tick_params(direction="in", length=7, top=True, right=True)
+    ax2.xaxis.set_minor_locator(AutoMinorLocator(2))
+    ax2.yaxis.set_minor_locator(AutoMinorLocator(2))
+
+    ax2.scatter(y_act, y_pred, color=mfc, edgecolor=mec, alpha=0.5, s=35, lw=1.2)
+    ax2.plot(
+        [y_act.min(), y_act.max()], [y_act.min(), y_act.max()], "k--", alpha=0.8, label="Ideal"
+    )
+
+    ax2.set_xlabel("Actual value")
+    ax2.set_ylabel("Predicted value")
+    ax2.set_xlim([y_act.min() * 1.05, y_act.max() * 1.05])
+    ax2.set_ylim([y_act.min() * 1.05, y_act.max() * 1.05])
+
+    ax1 = fig.add_axes([left, bottom_h, width, 0.15])
+    ax1.hist(y_act, bins=31, density=True, color=mfc, edgecolor=mec, alpha=0.6)
+    ax1.set_xticks([])
+    ax1.set_yticks([])
+    ax1.set_xlim(ax2.get_xlim())
+
+    if x_hist:
+        ax1.set_alpha(1.0)
+
+    ax3 = fig.add_axes([left_h, bottom, 0.15, height])
+    ax3.hist(
+        y_pred, bins=31, density=True, color=mfc, edgecolor=mec, orientation="horizontal", alpha=0.6
+    )
+    ax3.set_xticks([])
+    ax3.set_yticks([])
+    ax3.set_ylim(ax2.get_ylim())
+
+    if y_hist:
+        ax3.set_alpha(1.0)
+
+    if reg_line:
+        polyfit = np.polyfit(y_act, y_pred, deg=1)
+        reg_line_vals = np.poly1d(polyfit)(np.unique(y_act))
+        ax2.plot(np.unique(y_act), reg_line_vals, "r-", label="Regression Line", alpha=0.8)
+
+    ax2.legend(loc="upper left", framealpha=0.35, handlelength=1.5)
+
+    plt.tight_layout()
+
+    if save_dir is not None:
+        os.makedirs(save_dir, exist_ok=True)
+        fig_name = os.path.join(save_dir, f"{name}_act_pred.png")
+        plt.savefig(fig_name, bbox_inches="tight", dpi=300)
+
+    plt.show()
+    plt.close(fig)
+
+
+def residual(
+    y_act: np.ndarray, y_pred: np.ndarray, name: str = "example", save_dir: str = None
+) -> None:
+    """
+    Plots the residual errors between the actual and predicted values.
+
+    This function generates a residual plot by calculating the difference between the
+    actual values (y_act) and predicted values (y_pred). The plot shows the residuals
+    (y_pred - y_act) against the actual values. Optionally, the plot can be saved to a file.
+
+    Parameters
+    ----------
+    y_act : `np.ndarray`
+        The actual values, typically the ground truth values.
+    y_pred : `np.ndarray`
+        The predicted values that are compared against the actual values.
+    name : `str`, optional
+        The name of the plot file (without extension) used when saving the plot. Default is "example".
+    save_dir : `str`, optional
+        The directory where the plot will be saved. If None, the plot is not saved. Default is None.
+
+    Returns
+    -------
+    `None` : This function does not return any value. It generates and optionally saves a plot.
+
+    Notes
+    -----
+    - The plot is shown with the residuals (y_pred - y_act) on the y-axis and the actual values (y_act)
+      on the x-axis. The plot includes a horizontal line representing the ideal case where the residual
+      is zero (i.e., perfect predictions).
+    - The plot will be saved as a PNG image if a valid `save_dir` is provided.
+    """
+
+    mec = "#2F4F4F"
+    mfc = "#C0C0C0"
+
+    y_act = np.array(y_act)
+    y_pred = np.array(y_pred)
+
+    xmin = np.min([y_act]) * 0.9
+    xmax = np.max([y_act]) / 0.9
+    y_err = y_pred - y_act
+    ymin = np.min([y_err]) * 0.9
+    ymax = np.max([y_err]) / 0.9
+
+    fig, ax = plt.subplots(figsize=(4, 4))
+
+    ax.plot(y_act, y_err, "o", mec=mec, mfc=mfc, alpha=0.5, label=None, mew=1.2, ms=5.2)
+    ax.plot([xmin, xmax], [0, 0], "k--", alpha=0.8, label="ideal")
+
+    ax.set_ylabel("Residual error")
+    ax.set_xlabel("Actual value")
+    ax.legend(loc="lower right")
+
+    minor_locator_x = AutoMinorLocator(2)
+    minor_locator_y = AutoMinorLocator(2)
+    ax.get_xaxis().set_minor_locator(minor_locator_x)
+    ax.get_yaxis().set_minor_locator(minor_locator_y)
+
+    ax.tick_params(right=True, top=True, direction="in", length=7)
+    ax.tick_params(which="minor", right=True, top=True, direction="in", length=4)
+
+    ax.set_xlim(xmin, xmax)
+    ax.set_ylim(ymin, ymax)
+
+    if save_dir is not None:
+        fig_name = f"{save_dir}/{name}_residual.png"
+        os.makedirs(save_dir, exist_ok=True)
+        plt.savefig(fig_name, bbox_inches="tight", dpi=300)
+
+    plt.draw()
+    plt.pause(0.001)
+
+    plt.close()
+
+
+def residual_hist(
+    y_act: np.ndarray, y_pred: np.ndarray, name: str = "example", save_dir: Optional[str] = None
+) -> None:
+    """
+    Generates a residual error histogram with kernel density estimate (KDE) for the given true and predicted values.
+    Optionally saves the plot to a specified directory.
+
+    Parameters
+    ----------
+    y_act : `np.ndarray`
+        Array of true (actual) values.
+
+    y_pred : `np.ndarray`
+        Array of predicted values.
+
+    name : `str`, optional, default="example"
+        The name used for the saved plot filename.
+
+    save_dir : `str`, optional, default=None
+        Directory path to save the generated plot. If None, the plot is not saved.
+
+    Returns
+    --------
+    `None` : This function generates and optionally saves a plot but does not return any value.
+
+    Raises
+    -------
+    `UserWarning` : If the data has high correlation among variables, suggesting dimensionality reduction.
+    """
+    mec = "#2F4F4F"
+    mfc = "#C0C0C0"
+    y_pred, y_act = y_pred.flatten(), y_act.flatten()
+
+    fig, ax = plt.subplots(figsize=(4, 4))
+    y_err = y_pred - y_act
+    x_range = np.linspace(min(y_err), max(y_err), 1000)
+
+    try:
+        kde_act = stats.gaussian_kde(y_err)
+        ax.plot(x_range, kde_act(x_range), "-", lw=1.2, color="k", label="kde")
+    except np.linalg.LinAlgError as e:
+        warnings.warn(
+            "The data has very high correlation among variables. Consider dimensionality reduction.",
+            UserWarning,
+        )
+
+    ax.hist(y_err, color=mfc, bins=35, alpha=1, edgecolor=mec, density=True)
+
+    ax.set_xlabel("Residual error")
+    ax.set_ylabel("Relative frequency")
+    plt.legend(loc=2, framealpha=0.35, handlelength=1.5)
+
+    ax.tick_params(direction="in", length=7, top=True, right=True)
+
+    minor_locator_x = AutoMinorLocator(2)
+    minor_locator_y = AutoMinorLocator(2)
+    ax.get_xaxis().set_minor_locator(minor_locator_x)
+    ax.get_yaxis().set_minor_locator(minor_locator_y)
+    plt.tick_params(which="minor", direction="in", length=4, right=True, top=True)
+
+    if save_dir is not None:
+        fig_name = f"{save_dir}/{name}_residual_hist.png"
+        os.makedirs(save_dir, exist_ok=True)
+        plt.savefig(fig_name, bbox_inches="tight", dpi=300)
+
+    plt.draw()
+    plt.pause(0.001)
+    plt.close()
+
+
+def loss_curve(
+    x_data: np.ndarray,
+    train_err: np.ndarray,
+    val_err: np.ndarray,
+    name: str = "example",
+    save_dir: Optional[str] = None,
+) -> None:
+    """
+    Plots the loss curve for both training and validation errors over epochs,
+    and optionally saves the plot as an image.
+
+    Parameters
+    ----------
+    x_data : `np.ndarray`
+        Array of x-values (usually epochs) for the plot.
+    train_err : `np.ndarray`
+        Array of training error values.
+    val_err : `np.ndarray`
+        Array of validation error values.
+    name : `str`, optional
+        The name to use when saving the plot. Default is "example".
+    save_dir : `Optional[str]`, optional
+        Directory where the plot should be saved. If None, the plot is not saved. Default is None.
+
+    Returns
+    -------
+    `None` : This function does not return any value. It generates and optionally saves a plot.
+    """
+    mec1 = "#2F4F4F"
+    mfc1 = "#C0C0C0"
+    mec2 = "maroon"
+    mfc2 = "pink"
+
+    fig, ax = plt.subplots(figsize=(4, 4))
+
+    ax.plot(
+        x_data,
+        train_err,
+        "-",
+        color=mec1,
+        marker="o",
+        mec=mec1,
+        mfc=mfc1,
+        ms=4,
+        alpha=0.5,
+        label="train",
+    )
+
+    ax.plot(
+        x_data,
+        val_err,
+        "--",
+        color=mec2,
+        marker="s",
+        mec=mec2,
+        mfc=mfc2,
+        ms=4,
+        alpha=0.5,
+        label="validation",
+    )
+
+    max_val_err = max(val_err)
+    ax.axhline(max_val_err, color="b", linestyle="--", alpha=0.3)
+
+    ax.set_xlabel("Number of training epochs")
+    ax.set_ylabel("Loss (Units)")
+    ax.set_ylim(0, 2 * np.mean(val_err))
+
+    ax.legend(loc=1, framealpha=0.35, handlelength=1.5)
+
+    minor_locator_x = AutoMinorLocator(2)
+    minor_locator_y = AutoMinorLocator(2)
+    ax.get_xaxis().set_minor_locator(minor_locator_x)
+    ax.get_yaxis().set_minor_locator(minor_locator_y)
+
+    ax.tick_params(right=True, top=True, direction="in", length=7)
+    ax.tick_params(which="minor", right=True, top=True, direction="in", length=4)
+
+    if save_dir is not None:
+        fig_name = f"{save_dir}/{name}_loss_curve.png"
+        os.makedirs(save_dir, exist_ok=True)
+        plt.savefig(fig_name, bbox_inches="tight", dpi=300)
+
+    plt.draw()
+    plt.pause(0.001)
+    plt.close()