likelihood 1.2.22__tar.gz → 1.2.24__tar.gz

{likelihood-1.2.22 → likelihood-1.2.24}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: likelihood
-Version: 1.2.22
+Version: 1.2.24
 Summary: A package that performs the maximum likelihood algorithm.
 Home-page: https://github.com/jzsmoreno/likelihood/
 Author: J. A. Moreno-Guerra

{likelihood-1.2.22 → likelihood-1.2.24}/likelihood/graph/graph.py

@@ -74,3 +74,20 @@ class DynamicGraph(FeatureSelection):
             nx_graph.add_edges_from([(source, target, edge)])
 
         return nx_graph
+
+
+# -------------------------------------------------------------------------
+if __name__ == "__main__":
+    import numpy as np
+    import pandas as pd
+
+    # Generate data
+    x = np.random.rand(3, 100)
+    y = 0.1 * x[0, :] + 0.4 * x[1, :] + 0.5 * x[2, :] + 0.1
+    # Create a DataFrame
+    df = pd.DataFrame(x.T, columns=["x1", "x2", "x3"])
+    df["y"] = y
+    # Instantiate DynamicGraph
+    fs = DynamicGraph(df, n_importances=2)
+    print(fs.fit())
+    fs.draw()

{likelihood-1.2.22 → likelihood-1.2.24}/likelihood/graph/nn.py

@@ -1,6 +1,8 @@
 import os
 
 os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
+# Suppress TensorFlow INFO logs
+os.environ["TF_CPP_MIN_LOG_LEVEL"] = "2"
 import logging
 import warnings
 from typing import List, Tuple
@@ -9,7 +11,6 @@ import numpy as np
 import pandas as pd
 import tensorflow as tf
 from IPython.display import clear_output
-from numpy import ndarray
 from pandas.core.frame import DataFrame
 from sklearn.metrics import f1_score
 from sklearn.model_selection import train_test_split
@@ -21,7 +22,7 @@ logging.getLogger("tensorflow").setLevel(logging.ERROR)
 tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
 
 
-def compare_similarity(arr1: ndarray, arr2: ndarray) -> int:
+def compare_similarity(arr1: np.ndarray, arr2: np.ndarray) -> int:
     """Compares the similarity between two arrays of categories.
 
     Parameters
@@ -44,9 +45,9 @@ def compare_similarity(arr1: ndarray, arr2: ndarray) -> int:
     return count
 
 
-def cal_adjency_matrix(
+def cal_adjacency_matrix(
     df: DataFrame, exclude_subset: List[str] = [], sparse: bool = True, **kwargs
-) -> Tuple[dict, ndarray]:
+) -> Tuple[dict, np.ndarray]:
     """Calculates the adjacency matrix for a given DataFrame.
     The adjacency matrix is a matrix that represents the similarity between each pair of categories.
     The similarity is calculated using the `compare_similarity` function.
@@ -133,7 +134,7 @@ class Data:
         target: str | None = None,
         exclude_subset: List[str] = [],
     ):
-        _, adjacency = cal_adjency_matrix(df, exclude_subset=exclude_subset, sparse=True)
+        _, adjacency = cal_adjacency_matrix(df, exclude_subset=exclude_subset, sparse=True)
         if target is not None:
             X = df.drop(columns=[target] + exclude_subset)
         else:

likelihood-1.2.24/likelihood/models/hmm.py

@@ -0,0 +1,163 @@
+import logging
+import os
+import pickle
+from typing import List, Tuple
+
+import numpy as np
+from IPython.display import clear_output
+
+
+class HMM:
+    def __init__(self, n_states: int, n_observations: int):
+        self.n_states = n_states
+        self.n_observations = n_observations
+
+        # Initialize parameters with random values
+        self.pi = np.random.dirichlet(np.ones(n_states), size=1)[0]
+        self.A = np.random.dirichlet(np.ones(n_states), size=n_states)
+        self.B = np.random.dirichlet(np.ones(n_observations), size=n_states)
+
+    def save_model(self, filename: str = "./hmm") -> None:
+        filename = filename if filename.endswith(".pkl") else filename + ".pkl"
+        with open(filename, "wb") as f:
+            pickle.dump(self, f)
+
+    @staticmethod
+    def load_model(filename: str = "./hmm") -> "HMM":
+        filename = filename + ".pkl" if not filename.endswith(".pkl") else filename
+        with open(filename, "rb") as f:
+            return pickle.load(f)
+
+    def forward(self, sequence: List[int]) -> np.ndarray:
+        T = len(sequence)
+        alpha = np.zeros((T, self.n_states))
+
+        # Add a small constant (smoothing) to avoid log(0)
+        epsilon = 1e-10  # Small value to avoid taking log(0)
+
+        # Initialization (log-space)
+        alpha[0] = np.log(self.pi + epsilon) + np.log(self.B[:, sequence[0]] + epsilon)
+        alpha[0] -= np.log(np.sum(np.exp(alpha[0])))  # Normalization (log-space)
+
+        # Recursion (log-space)
+        for t in range(1, T):
+            for i in range(self.n_states):
+                alpha[t, i] = np.log(
+                    np.sum(np.exp(alpha[t - 1] + np.log(self.A[:, i] + epsilon)))
+                ) + np.log(self.B[i, sequence[t]] + epsilon)
+            alpha[t] -= np.log(np.sum(np.exp(alpha[t])))  # Normalization
+
+        return alpha
+
+    def backward(self, sequence: List[int]) -> np.ndarray:
+        T = len(sequence)
+        beta = np.ones((T, self.n_states))
+
+        # Backward recursion
+        for t in range(T - 2, -1, -1):
+            for i in range(self.n_states):
+                beta[t, i] = np.sum(self.A[i] * self.B[:, sequence[t + 1]] * beta[t + 1])
+
+        return beta
+
+    def viterbi(self, sequence: List[int]) -> np.ndarray:
+        T = len(sequence)
+        delta = np.zeros((T, self.n_states))
+        psi = np.zeros((T, self.n_states), dtype=int)
+
+        # Initialization
+        delta[0] = self.pi * self.B[:, sequence[0]]
+
+        # Recursion
+        for t in range(1, T):
+            for i in range(self.n_states):
+                delta[t, i] = np.max(delta[t - 1] * self.A[:, i]) * self.B[i, sequence[t]]
+                psi[t, i] = np.argmax(delta[t - 1] * self.A[:, i])
+
+        # Reconstruct the most probable path
+        state_sequence = np.zeros(T, dtype=int)
+        state_sequence[T - 1] = np.argmax(delta[T - 1])
+        for t in range(T - 2, -1, -1):
+            state_sequence[t] = psi[t + 1, state_sequence[t + 1]]
+
+        return state_sequence
+
+    def baum_welch(
+        self, sequences: List[List[int]], n_iterations: int, verbose: bool = False
+    ) -> None:
+        for iteration in range(n_iterations):
+            # Initialize accumulators
+            A_num = np.zeros((self.n_states, self.n_states))
+            B_num = np.zeros((self.n_states, self.n_observations))
+            pi_num = np.zeros(self.n_states)
+
+            for sequence in sequences:
+                T = len(sequence)
+                alpha = self.forward(sequence)
+                beta = self.backward(sequence)
+
+                # Update pi
+                gamma = (alpha * beta) / np.sum(alpha * beta, axis=1, keepdims=True)
+                pi_num += gamma[0]
+
+                # Update A and B
+                for t in range(T - 1):
+                    xi = np.zeros((self.n_states, self.n_states))
+                    denom = np.sum(alpha[t] * self.A * self.B[:, sequence[t + 1]] * beta[t + 1])
+
+                    for i in range(self.n_states):
+                        for j in range(self.n_states):
+                            xi[i, j] = (
+                                alpha[t, i]
+                                * self.A[i, j]
+                                * self.B[j, sequence[t + 1]]
+                                * beta[t + 1, j]
+                            ) / denom
+                        A_num[i] += xi[i]
+
+                    B_num[:, sequence[t]] += gamma[t]
+
+                # For the last step of the sequence
+                B_num[:, sequence[-1]] += gamma[-1]
+
+            # Normalize and update parameters
+            self.pi = pi_num / len(sequences)
+            self.A = A_num / np.sum(A_num, axis=1, keepdims=True)
+            self.B = B_num / np.sum(B_num, axis=1, keepdims=True)
+
+            # Logging parameters every 10 iterations
+            if iteration % 10 == 0 and verbose:
+                os.system("cls" if os.name == "nt" else "clear")
+                clear_output(wait=True)
+                logging.info(f"Iteration {iteration}:")
+                logging.info("Pi: %s", self.pi)
+                logging.info("A:\n%s", self.A)
+                logging.info("B:\n%s", self.B)
+
+    def decoding_accuracy(self, sequences: List[List[int]], true_states: List[List[int]]) -> float:
+        correct_predictions = 0
+        total_predictions = 0
+
+        for sequence, true_state in zip(sequences, true_states):
+            predicted_states = self.viterbi(sequence)
+            correct_predictions += np.sum(predicted_states == true_state)
+            total_predictions += len(sequence)
+
+        accuracy = (correct_predictions / total_predictions) * 100
+        return accuracy
+
+    def state_probabilities(self, sequence: List[int]) -> np.ndarray:
+        """
+        Returns the smoothed probabilities of the hidden states at each time step.
+        This is done by using both forward and backward probabilities.
+        """
+        alpha = self.forward(sequence)
+        beta = self.backward(sequence)
+
+        # Compute smoothed probabilities (gamma)
+        smoothed_probs = (alpha * beta) / np.sum(alpha * beta, axis=1, keepdims=True)
+
+        return smoothed_probs
+
+    def sequence_probability(self, sequence: List[int]) -> np.ndarray:
+        return self.state_probabilities(sequence)[-1]

likelihood-1.2.24/likelihood/models/simulation.py

@@ -0,0 +1,222 @@
+import pickle
+import warnings
+from typing import List, Tuple, Union
+
+import matplotlib.pyplot as plt
+import numpy as np
+import pandas as pd
+from pandas.core.frame import DataFrame
+
+from likelihood.tools import DataScaler, FeatureSelection, OneHotEncoder, cdf, check_nan_inf
+
+# Suppress RankWarning
+warnings.simplefilter("ignore", np.RankWarning)
+
+
+# --------------------------------------------------------------------------------------------------------------------------------------
+def categories_by_quartile(df: DataFrame, column: str) -> Tuple[str, str]:
+    # Count the frequency of each category in the column
+    freq = df[column].value_counts()
+
+    # Calculate the 25th percentile (Q1) and 75th percentile (Q3)
+    q1 = freq.quantile(0.25)
+    q3 = freq.quantile(0.75)
+
+    # Filter categories that are below the 25th percentile and above the 75th percentile
+    least_frequent = freq[freq <= q1]
+    most_frequent = freq[freq >= q3]
+
+    # Get the least frequent category (25th percentile) and the most frequent category (75th percentile)
+    least_frequent_category = least_frequent.idxmin() if not least_frequent.empty else None
+    most_frequent_category = most_frequent.idxmax() if not most_frequent.empty else None
+
+    return least_frequent_category, most_frequent_category
+
+
+class SimulationEngine(FeatureSelection):
+    """
+    This class implements a predictive model that utilizes multiple linear regression for numerical target variables
+    and multiple logistic regression for categorical target variables.
+
+    The class provides methods for training the model on a given dataset, making predictions,
+    and evaluating the model's performance.
+
+    Key features:
+    - Supports both numerical and categorical target variables, automatically selecting the appropriate regression method.
+    - Includes methods for data preprocessing, model fitting, prediction, and evaluation metrics.
+    - Designed to be flexible and user-friendly, allowing for easy integration with various datasets.
+
+    Usage:
+    - Instantiate the class with the training data and target variable.
+    - Call the fit method to train the model.
+    - Use the predict method to generate predictions on new data.
+    - Evaluate the model using built-in metrics for accuracy and error.
+
+    This class is suitable for applications in data analysis and machine learning, enabling users to leverage regression techniques
+    for both numerical and categorical outcomes efficiently.
+    """
+
+    def __init__(self, use_scaler: bool = False, **kwargs):
+
+        self.df = pd.DataFrame()
+        self.n_importances = None
+        self.use_scaler = use_scaler
+        self.proba_dict = {}
+
+        super().__init__(**kwargs)
+
+    def predict(self, df: DataFrame, column: str) -> np.ndarray | list:
+        # Let us assign the dictionary entries corresponding to the column
+        w, quick_encoder, names_cols, dfe, numeric_dict = self.w_dict[column]
+
+        df = df[names_cols].copy()
+        # Change the scale of the DataFrame
+        dataset = self.df.copy()
+        dataset.drop(columns=column, inplace=True)
+        numeric_df = dataset.select_dtypes(include="number")
+        if self.use_scaler:
+            scaler = DataScaler(numeric_df.copy().to_numpy().T, n=None)
+            _ = scaler.rescale()
+            dataset_ = df.copy()
+            numeric_df = dataset_.select_dtypes(include="number")
+            numeric_scaled = scaler.rescale(dataset_=numeric_df.to_numpy())
+            numeric_df = pd.DataFrame(numeric_scaled.T, columns=numeric_df.columns)
+            for col in numeric_df.columns:
+                df[col] = numeric_df[col].values
+
+        # Encoding the DataFrame
+        for num, colname in enumerate(dfe._encode_columns):
+            if df[colname].dtype == "object":
+                encode_dict = dfe.encoding_list[num]
+                df[colname] = df[colname].apply(
+                    dfe._code_transformation_to, dictionary_list=encode_dict
+                )
+
+        # Prediction
+        y = df.to_numpy() @ w
+
+        # Categorical column
+        if quick_encoder != None:
+
+            one_hot = OneHotEncoder()
+            y = one_hot.decode(y)
+            encoding_dic = quick_encoder.decoding_list[0]
+            y = [encoding_dic[item] for item in y]
+        # Numeric column
+        else:
+            if self.use_scaler:
+                # scale output
+                y += 1
+                y /= 2
+                y = y * (self.df[column].max() - self.df[column].min())
+
+        return y[:]
+
+    def _encode(self, df: DataFrame) -> np.ndarray | list:
+        df = df.copy()
+        column = df.columns[0]
+        frec = df[column].value_counts() / len(df)
+        df.loc[:, "frec"] = df[column].map(frec)
+        df.sort_values("frec", inplace=True)
+        keys = df[column].to_list()
+        values = df["frec"].to_list()
+        return dict(zip(keys, values))
+
+    def fit(self, df: DataFrame, n_importances: int, **kwargs) -> None:
+        self.df = df
+        self.n_importances = n_importances
+        # We run the feature selection algorithm
+        self.get_digraph(self.df, self.n_importances, self.use_scaler)
+        proba_dict_keys = list(self.w_dict.keys())
+        self.proba_dict = dict(zip(proba_dict_keys, [i for i in range(len(proba_dict_keys))]))
+        for key in proba_dict_keys:
+            x = (
+                self.df[key].values,
+                None if self.df[key].dtype != "object" else self._encode(self.df[[key]]),
+            )
+            poly = kwargs.get("poly", 9)
+            plot = kwargs.get("plot", False)
+            if not x[1]:
+                media = self.df[key].mean()
+                desviacion_estandar = self.df[key].std()
+                cota_inferior = media - 1.5 * desviacion_estandar
+                cota_superior = media + 1.5 * desviacion_estandar
+                if plot:
+                    print(f"Cumulative Distribution Function ({key})")
+                f, cdf_, ox = cdf(x[0].flatten(), poly=poly, plot=plot)
+            else:
+                f, ox = None, None
+                least_frequent_category, most_frequent_category = categories_by_quartile(
+                    self.df[[key]], key
+                )
+                cota_inferior = x[1].get(least_frequent_category, 0)
+                cota_superior = x[1].get(most_frequent_category, 0)
+            self.proba_dict[key] = (
+                f if f else None,
+                x[1],
+                (np.mean(np.abs(np.diff(ox))) / 2.0 if isinstance(ox, np.ndarray) else None),
+                f(cota_inferior) if f else cota_inferior,
+                f(cota_superior) if f else cota_superior,
+            )
+
+    def get_proba(self, value: Union[Union[float, int], str] | list, colname: str) -> List[float]:
+        value = (
+            value
+            if isinstance(value, list)
+            else value.flatten().tolist() if isinstance(value, np.ndarray) else [value]
+        )
+        return [
+            (
+                self.proba_dict[colname][0](val)
+                - self.proba_dict[colname][0](val - self.proba_dict[colname][2])
+                if (isinstance(val, float) or isinstance(val, int))
+                else self.proba_dict[colname][1].get(val, 0)
+            )
+            for val in value
+        ]
+
+    def pred_outliers(self, value: Union[Union[float, int], str] | list, colname: str) -> List[str]:
+        return [
+            (
+                "inlier"
+                if (self.proba_dict[colname][3] < val < self.proba_dict[colname][4])
+                else "outlier"
+            )
+            for val in self.get_proba(value, colname)
+        ]
+
+    def _clean_data(self, df: DataFrame) -> DataFrame:
+
+        df.replace([np.inf, -np.inf], np.nan, inplace=True)
+        df.replace(" ", np.nan, inplace=True)
+        df = check_nan_inf(df)
+        df = df.reset_index()
+        df = df.drop(columns=["index"])
+
+        return df
+
+    def save(self, filename: str = "./simulation_model") -> None:
+        """
+        Save the state of the SimulationEngine to a file.
+
+        Parameters:
+            filename (str): The name of the file where the object will be saved.
+        """
+        filename = filename if filename.endswith(".pkl") else filename + ".pkl"
+        with open(filename, "wb") as f:
+            pickle.dump(self, f)
+
+    @staticmethod
+    def load(filename: str = "./simulation_model"):
+        """
+        Load the state of a SimulationEngine from a file.
+
+        Parameters:
+            filename (str): The name of the file containing the saved object.
+
+        Returns:
+            SimulationEngine: A new instance of SimulationEngine with the loaded state.
+        """
+        filename = filename + ".pkl" if not filename.endswith(".pkl") else filename
+        with open(filename, "rb") as f:
+            return pickle.load(f)