pelican-nlp 0.1.0__py3-none-any.whl → 0.1.2__py3-none-any.whl

pelican_nlp/Nils_backup/extract_acoustic_features.py

@@ -0,0 +1,274 @@
+#import pandas as pd
+#import numpy as np
+#from pydub import AudioSegment
+#from pyannote.audio import Model, Inference
+
+class AudioFeatureExtractor:
+    def __init__(self, model_name, token, device="cpu"):
+        """
+        Initializes the AudioFeatureExtractor class.
+
+        Parameters:
+        - model_name: str, name of the pretrained model_instance from pyannote
+        - token: str, the Hugging Face authentication token for downloading the model_instance
+        - device: str, device to run the model_instance on (default is "cpu")
+        """
+        self.model = Model.from_pretrained(model_name, use_auth_token=token).to(device)
+
+    def extract_audio_window(self, audio, start_time=0, duration=60000):
+        """
+        Extract a segment from the audio starting at `start_time` with a specified duration.
+        
+        Parameters:
+        - audio: AudioSegment object, the input audio
+        - start_time: int, starting point of the window in milliseconds (default is 0)
+        - duration: int, duration of the window to extract in milliseconds (default is 60000)
+        
+        Returns:
+        - AudioSegment object of the extracted window
+        """
+        end_time = start_time + duration
+        return audio[start_time:end_time]
+
+    def extract_embeddings(self, inference, file_path):
+        """
+        Extract embeddings from an audio file using the inference model_instance.
+        
+        Parameters:
+        - inference: Inference object from pyannote
+        - file_path: str, path to the audio file
+        
+        Returns:
+        - numpy array of embeddings
+        """
+        embeddings = inference(file_path)
+        return np.asarray(embeddings)
+
+    def process_audio(self, file_path, mode="whole", start_time=0, duration=60000, window_step=None):
+        """
+        Process an audio file, extracting either whole or windowed embeddings based on mode.
+        
+        Parameters:
+        - file_path: str, path to the audio file
+        - mode: str, "whole" for whole file extraction or "window" for windowed extraction (default is "whole")
+        - start_time: int, start time for the audio segment in milliseconds (only for window mode, default is 0)
+        - duration: int, duration for the audio segment in milliseconds (default is 60000)
+        - window_step: int, step size for window extraction in milliseconds (only for "window" mode)
+        
+        Returns:
+        - numpy array of embeddings
+        """
+        audio = AudioSegment.from_file(file_path)
+        if mode == "whole":
+            inference = Inference(self.model, window="whole")
+            embeddings = self.extract_embeddings(inference, file_path)
+        elif mode == "window":
+            # If window mode is specified, we extract in a sliding window fashion
+            embeddings = []
+            inference = Inference(self.model, window="sliding", duration=duration, step=window_step)
+            
+            # Split audio into windows and extract embeddings for each window
+            for i in range(0, len(audio), window_step):
+                window_audio = self.extract_audio_window(audio, start_time=i, duration=duration)
+                temp_path = f"temp_window_{i}.wav"
+                window_audio.export(temp_path, format="wav")
+                window_embeddings = self.extract_embeddings(inference, temp_path)
+                embeddings.append(window_embeddings)
+                os.remove(temp_path)
+            embeddings = np.vstack(embeddings)  # Stack all window embeddings
+        else:
+            raise ValueError("Invalid mode. Use 'whole' or 'window'.")
+
+        return embeddings
+
+    def save_embeddings(self, embeddings, output_path):
+        """
+        Save the embeddings to a CSV file.
+        
+        Parameters:
+        - embeddings: numpy array of embeddings
+        - output_path: str, path to save the CSV file
+        """
+        df = pd.DataFrame(embeddings)
+        df.to_csv(output_path, index=False)
+
+# Example usage:
+if __name__ == "__main__":
+    # Initialize the extractor
+    extractor = AudioFeatureExtractor(
+        model_name="pyannote/embedding",
+        token="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",
+        device="mps"
+    )
+
+    # Process a whole file
+    whole_embeddings = extractor.process_audio(
+        file_path="path/to/audio_file.wav",
+        mode="whole"
+    )
+
+    # Process a file using sliding window extraction
+    window_embeddings = extractor.process_audio(
+        file_path="path/to/audio_file.wav",
+        mode="window",
+        start_time=0,
+        duration=10000,  # e.g., 10 seconds window
+        window_step=5000  # e.g., 5 seconds step
+    )
+
+    # Save the embeddings
+    extractor.save_embeddings(whole_embeddings, "path/to/output_whole.csv")
+    extractor.save_embeddings(window_embeddings, "path/to/output_window.csv")
+    
+    
+
+'''import os
+import numpy as np
+from pydub import AudioSegment
+from pyannote.audio import Model, Inference'''
+
+
+class AudioFeatureExtractor:
+    def __init__(self, model_name_or_instance, device="cpu", use_auth_token=None):
+        """
+        Initializes the AudioFeatureExtractor class.
+
+        Parameters:
+        - model_name_or_instance: str or Model, the name of the pretrained model_instance from pyannote or an instance of Model
+        - device: str, device to run the model_instance on (default is "cpu")
+        - use_auth_token: str, Hugging Face authentication token if required
+        """
+        if isinstance(model_name_or_instance, str):
+            self.model = Model.from_pretrained(
+                model_name_or_instance, use_auth_token=use_auth_token
+            ).to(device)
+        else:
+            self.model = model_name_or_instance.to(device)
+        self.device = device
+
+    def extract_audio_window(self, audio, start_time=0, duration=None):
+        """
+        Extract a segment from the audio starting at 'start_time' with a specified 'duration'.
+
+        Parameters:
+        - audio: AudioSegment object, the input audio
+        - start_time: int, starting point of the window in milliseconds (default is 0)
+        - duration: int, duration of the window to extract in milliseconds (default is None, till the end)
+
+        Returns:
+        - AudioSegment object of the extracted window
+        """
+        if duration is None:
+            duration = len(audio) - start_time
+        end_time = start_time + duration
+        return audio[start_time:end_time]
+
+    def extract_embeddings(self, inference, file_path):
+        """
+        Extract embeddings from the audio file using the specified inference model_instance.
+
+        Parameters:
+        - inference: Inference object from pyannote
+        - file_path: str, path to the audio file
+
+        Returns:
+        - numpy array of embeddings
+        """
+        embeddings = inference(file_path)
+        return np.asarray(embeddings)
+
+    def process_audio(self, file_path, mode="whole", window_duration=None, window_step=None, start_time=0, end_time=None):
+        """
+        Process an audio file, extracting embeddings based on the specified mode.
+
+        Parameters:
+        - file_path: str, path to the audio file
+        - mode: str, "whole" for whole file extraction or "windowed" for windowed extraction (default is "whole")
+        - window_duration: int, duration of the window in milliseconds (required for "windowed" mode)
+        - window_step: int, step size in milliseconds between windows (required for "windowed" mode)
+        - start_time: int, start time in milliseconds for processing (default is 0)
+        - end_time: int, end time in milliseconds for processing (default is None, till the end)
+
+        Returns:
+        - numpy array of embeddings
+        """
+        # Load and optionally trim the audio file
+        audio = AudioSegment.from_file(file_path)
+        if end_time is None or end_time > len(audio):
+            end_time = len(audio)
+        audio = audio[start_time:end_time]
+
+        # Export the (possibly trimmed) audio to a temporary file
+        temp_dir = "temp_audio"
+        os.makedirs(temp_dir, exist_ok=True)
+        temp_path = os.path.join(temp_dir, "temp_audio.wav")
+        audio.export(temp_path, format="wav")
+
+        if mode == "whole":
+            inference = Inference(self.model, window="whole")
+            embeddings = self.extract_embeddings(inference, temp_path)
+        elif mode == "windowed":
+            if window_duration is None or window_step is None:
+                raise ValueError("window_duration and window_step must be specified for 'windowed' mode.")
+            # Convert milliseconds to seconds for pyannote
+            window_duration_sec = window_duration / 1000.0
+            window_step_sec = window_step / 1000.0
+            inference = Inference(
+                self.model,
+                window="sliding",
+                duration=window_duration_sec,
+                step=window_step_sec
+            )
+            embeddings = self.extract_embeddings(inference, temp_path)
+        else:
+            os.remove(temp_path)
+            raise ValueError("Invalid mode. Use 'whole' or 'windowed'.")
+
+        # Clean up temporary file
+        os.remove(temp_path)
+        return embeddings
+
+    def save_embeddings(self, embeddings, output_path):
+        """
+        Save the embeddings to a file.
+
+        Parameters:
+        - embeddings: numpy array, the embeddings to save
+        - output_path: str, the path where embeddings will be saved
+        """
+        np.save(output_path, embeddings)
+        # Alternatively, to save as CSV:
+        # np.savetxt(output_path, embeddings, delimiter=",")
+
+
+# Example usage:
+if __name__ == "__main__":
+    # Initialize the extractor with a model_instance name and token if required
+    extractor = AudioFeatureExtractor(
+        model_name_or_instance="pyannote/embedding",
+        device="cpu",
+        use_auth_token="YOUR_HUGGING_FACE_TOKEN"  # Replace with your token if necessary
+    )
+
+    # Path to your audio file
+    audio_file_path = "path/to/your/audio_file.wav"
+
+    # Extract embeddings from the whole audio file
+    whole_embeddings = extractor.process_audio(
+        file_path=audio_file_path,
+        mode="whole"
+    )
+
+    # Save the embeddings
+    extractor.save_embeddings(whole_embeddings, "whole_embeddings.npy")
+
+    # Extract embeddings using sliding windows
+    windowed_embeddings = extractor.process_audio(
+        file_path=audio_file_path,
+        mode="windowed",
+        window_duration=5000,  # Window duration in milliseconds (e.g., 5000 ms = 5 seconds)
+        window_step=1000       # Window step in milliseconds (e.g., 1000 ms = 1 second)
+    )
+
+    # Save the windowed embeddings
+    extractor.save_embeddings(windowed_embeddings, "windowed_embeddings.npy")

pelican_nlp/Nils_backup/fluency/aggregate_fluency_results.py

@@ -0,0 +1,186 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Created on Tue Jun 18 12:28:18 2024
+
+@author: nilsl
+"""
+import os
+import pandas as pd
+import numpy as np
+from config import CONFIG
+
+
+def add_prefix_if_short(text, length, prefix):
+    if len(text) < length:
+        return prefix + text
+    return text
+
+
+def parse_array_from_string(s):
+    """Parses a numpy array from a string representation."""
+    cleaned = s.strip("[]")
+    return np.array([float(x) for x in cleaned.split()])
+
+def load_csvs_to_dataframe(directory, match_str, indeces):
+    df_list = []
+    # Loop through all files in the specified directory
+    for filename in os.listdir(directory):
+        # Check if the file is a CSV and contains the match_str in the filename
+        if filename.endswith(".csv") and match_str in filename:
+            # Construct the full path to the file
+            file_path = os.path.join(directory, filename)
+            # Read the CSV file into a DataFrame
+            df = pd.read_csv(file_path, index_col=0, dtype=str)
+            # Remove the file extension and use the filename as the key in the dictionary
+            for i, index in enumerate(indeces):
+                df[f"attr_{i}"] = filename.split(".")[0].split("_")[index]
+            df_list.append(df)
+
+    return pd.concat(df_list, axis=0)
+
+
+def pivot_df(df, index_cols, pivot_cols):
+    # Perform the pivot operation
+    df_pivot = df.pivot_table(
+        index=index_cols, columns=pivot_cols, aggfunc="first"
+    )
+
+    # Flatten the MultiIndex columns by joining level names with corresponding values
+    df_pivot.columns = [
+        "{}_{}".format("_".join(map(str, col[:-1])), col[-1])
+        for col in df_pivot.columns
+    ]
+
+    # Reset index to make index columns regular columns again
+    df_pivot.reset_index(inplace=True)
+
+    # Handling the columns after resetting index
+    all_cols = df_pivot.columns.tolist()
+    non_index_cols = [col for col in all_cols if col not in index_cols]
+
+    # Sorting non-index columns by base name and context number, modified to handle multiple pivots
+    sorted_cols = sorted(
+        non_index_cols, key=lambda x: (x.split("_")[0], int(x.split("_")[-1]))
+    )
+
+    # Reordering DataFrame columns including index and sorted other columns
+    df_pivot = df_pivot[index_cols + sorted_cols]
+
+    return df_pivot
+
+def main():
+    """Main execution function."""
+    lower = CONFIG["shared"]["preprocessing"]["lower"]
+    
+    behav = pd.read_csv(CONFIG["aggregation"]["paths"]["behav_agg"], dtype=str)
+    behav["study_id"] = (
+        behav["study_id"]
+        .apply(add_prefix_if_short, length=4, prefix="0")
+        .apply(add_prefix_if_short, length=4, prefix="0")
+    )
+    behav_scores = [col for col in behav.columns if not col in ["study_id", "group", "gender", "first_language", "diagnosis"]]
+    behav[behav_scores] = behav[behav_scores].astype(float)
+
+    demo = pd.read_csv(
+        CONFIG["aggregation"]["paths"]["demo_clinical"], 
+        dtype=str
+    )[CONFIG["aggregation"]["demo_columns"]]
+
+    demo_scores = [col for col in demo.columns if not col in ['study_id', 'group', 'gender', 'first_language', 'diagnosis']]
+    demo[demo_scores] = demo[demo_scores].astype(float)
+
+    behav_mss = pd.read_csv(CONFIG["aggregation"]["paths"]["questionnaires"], dtype=str)
+    behav_mss[behav_mss.columns.drop("study_id")] = behav_mss[behav_mss.columns.drop("study_id")].astype(float)
+
+    behav_mss["study_id"] = (
+        behav_mss["study_id"]
+        .apply(add_prefix_if_short, length=4, prefix="0")
+        .apply(add_prefix_if_short, length=4, prefix="0")
+    )
+
+    fluency_optimality = (
+        load_csvs_to_dataframe(
+            CONFIG["optimality"]["paths"]["results_dir"],
+            "lower" if lower else "upper",
+            [3, 4, 5],
+        )
+        .rename(
+            columns={
+                "attr_0": "min_length",
+                "attr_1": "index_0_shuffle",
+                "attr_2": "index_-1_shuffle",
+            }
+        )
+        .drop("task", axis=1)
+    )
+
+    fluency_optimality["z_Real"] = (
+        fluency_optimality["actual_dist"].astype(float)
+        - fluency_optimality["average_dist"].astype(float)
+    ) / fluency_optimality["std_dist"].astype(float)
+    fluency_optimality["z_Real"] = fluency_optimality["z_Real"].astype(float)
+    fluency_optimality["all_pairs_average"] = fluency_optimality[
+        "all_pairs_average"
+    ].astype(float)
+    fluency_optimality["actual_dist"] = fluency_optimality["actual_dist"].astype(
+        float
+    )
+    fluency_optimality["min_length"] = fluency_optimality["min_length"].astype(
+        int
+    )
+
+    fluency_optimality = (
+        fluency_optimality.drop("window_index", axis = 1).groupby(
+            [
+                "min_length",
+                "index_0_shuffle",
+                "index_-1_shuffle",
+                "analysis_mode",
+                "study_id",
+                "sub_task",
+            ]
+        )["z_Real"]
+        .mean()
+        .reset_index()
+    )
+
+    fluency_optimality_pivot = pivot_df(
+        fluency_optimality,
+        ["study_id", "sub_task"],
+        ["analysis_mode", "index_0_shuffle", "index_-1_shuffle", "min_length"],
+    )
+
+    fluency_coherence = pd.read_csv(
+        os.path.join(CONFIG["coherence"]["paths"]["results_dir"], 
+                    f"coherence_results{'_lower' if lower else '_upper'}.csv"), 
+        dtype=str
+    )
+    
+    print(fluency_coherence.columns)
+    print(fluency_coherence.head())
+    
+    fluency_coherence[fluency_coherence.columns.drop(["study_id", "sub_task"])] = fluency_coherence[fluency_coherence.columns.drop(["study_id", "sub_task"])].astype(float)
+
+
+    index = demo.merge(behav).merge(behav_mss)
+    metrics = fluency_coherence.merge(fluency_optimality_pivot, how = "outer")
+
+    paper_df = index.merge(metrics)
+
+    paper_df["task"] = ""
+        
+    paper_df.loc[paper_df["sub_task"] == "b", "task"] = "phonetic"
+    paper_df.loc[paper_df["sub_task"] == "k", "task"] = "phonetic"
+    paper_df.loc[paper_df["sub_task"] == "m", "task"] = "phonetic"
+        
+    paper_df.loc[paper_df["sub_task"] == "animals", "task"] = "semantic"
+    paper_df.loc[paper_df["sub_task"] == "clothes", "task"] = "semantic"
+    paper_df.loc[paper_df["sub_task"] == "food", "task"] = "semantic"
+
+    paper_df.to_csv(CONFIG["aggregation"]["paths"]["output"])
+    
+    return True
+
+if __name__ == "__main__":
+    main()

pelican_nlp/Nils_backup/fluency/behavioral_data.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Process VELAS behavioral data.
+
+This script:
+1. Loads the VELAS behavioral master data
+2. Renames cognitive test variables for clarity
+3. Selects relevant columns
+4. Outputs cleaned CSV
+"""
+import pandas as pd
+from utils import ensure_output_dir
+from config import BEHAVIORAL_CONFIG
+
+def load_behavioral_data(filepath):
+    """Load behavioral data from CSV."""
+    return pd.read_csv(filepath)
+
+def rename_cognitive_variables(df):
+    """Rename cognitive test variables for clarity."""
+    return df.rename(columns=BEHAVIORAL_CONFIG["cognitive_variable_mapping"])
+
+def save_aggregated_data(df, output_path):
+    """Save relevant columns to CSV."""
+    ensure_output_dir(output_path)
+    df[BEHAVIORAL_CONFIG["columns_to_save"]].to_csv(output_path, index=False)
+
+def main():
+    # Get paths from config
+    paths = BEHAVIORAL_CONFIG["paths"]
+    
+    # Process data
+    df = load_behavioral_data(paths["input"])
+    print(df.columns)
+    df = rename_cognitive_variables(df)
+    save_aggregated_data(df, paths["output"])
+    
+    print(f"Processed behavioral data saved to: {paths['output']}")
+
+if __name__ == "__main__":
+    main()

pelican_nlp/Nils_backup/fluency/check_duplicates.py

@@ -0,0 +1,169 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Process VELAS fluency transcripts by cleaning duplicates and hyphenated words.
+
+This script:
+1. Analyzes text files for duplicates and hyphenated words
+2. Cleans transcripts by removing duplicates and hyphens
+3. Saves processed transcripts to output directory
+"""
+import os
+import re
+from collections import Counter
+from pathlib import Path
+from typing import Dict, List, Tuple
+from utils import ensure_output_dir, validate_input_data
+from config import DUPLICATES_CONFIG
+
+
+#implemented in fluency_cleaner==================================
+def analyze_transcript(content: str) -> Tuple[int, int]:
+    """
+    Count duplicates and hyphenated words in a transcript.
+    
+    Args:
+        content: Semicolon-separated transcript content
+    
+    Returns:
+        Tuple of (duplicate_count, hyphenated_word_count)
+    """
+    words = content.split(';')
+    word_counter = Counter(words)
+    
+    duplicates = sum(count - 1 for count in word_counter.values() if count > 1)
+    hyphenated = sum(1 for word in words if '-' in word)
+    
+    return duplicates, hyphenated
+#=============================================================================
+
+def analyze_directory(directory: str) -> Dict[str, int]:
+    """
+    Analyze all transcripts in directory for duplicates and hyphenated words.
+    
+    Args:
+        directory: Path to transcript directory
+    
+    Returns:
+        Dictionary with total counts of duplicates and hyphenated words
+    """
+    total_duplicates = 0
+    total_hyphenated = 0
+
+    for filename in os.listdir(directory):
+        if filename.endswith('.txt') and DUPLICATES_CONFIG["file_filter"] in filename:
+            filepath = os.path.join(directory, filename)
+            with open(filepath, 'r') as file:
+                content = file.read()
+                duplicates, hyphenated = analyze_transcript(content)
+                total_duplicates += duplicates
+                total_hyphenated += hyphenated
+
+    return {
+        'duplicates': total_duplicates,
+        'hyphenated': total_hyphenated
+    }
+
+#=======================================================
+#implemented in fluency_cleaner
+def clean_transcript(content: str) -> str:
+
+    # Remove whitespace and decorators
+    content = re.sub(r'\s+', '', content).strip()
+    
+    # Split and clean words
+    words = [word for word in content.split(';') if word]
+    words = [word.replace('-', '') for word in words]
+    
+    # Remove duplicate words while preserving order
+    word_counter = Counter(words)
+    seen = set()
+    cleaned_words = []
+    
+    for word in words:
+        if word in seen and word_counter[word] > 1:
+            word_counter[word] -= 1
+        else:
+            cleaned_words.append(word)
+            seen.add(word)
+    
+    return ';'.join(cleaned_words)
+#===========================================================
+
+def process_directory(input_dir: str, output_dir: str) -> None:
+    """
+    Process all transcripts in directory, cleaning and saving to output directory.
+    
+    Args:
+        input_dir: Directory containing raw transcripts
+        output_dir: Directory for cleaned transcripts
+    """
+    # Create the output directory and any necessary parent directories
+    print(f"Creating output directory: {output_dir}")
+    print(f"Output directory exists before ensure_output_dir? {os.path.exists(output_dir)}")
+    ensure_output_dir(output_dir)
+    print(f"Output directory exists after ensure_output_dir? {os.path.exists(output_dir)}")
+
+    for filename in os.listdir(input_dir):
+        if filename.endswith('.txt') and DUPLICATES_CONFIG["file_filter"] in filename:
+            input_path = os.path.join(input_dir, filename)
+            output_path = os.path.join(output_dir, filename)
+            
+            print(f"\nProcessing file: {filename}")
+            print(f"Input path: {input_path}")
+            print(f"Output path: {output_path}")
+            print(f"Output dir exists? {os.path.exists(os.path.dirname(output_path))}")
+            
+            # Ensure the directory for this specific file exists
+            ensure_output_dir(os.path.dirname(output_path))
+            print(f"Output dir exists after ensure? {os.path.exists(os.path.dirname(output_path))}")
+            
+            with open(input_path, 'r') as infile:
+                content = infile.read()
+                cleaned_content = clean_transcript(content)
+                
+            with open(output_path, 'w') as outfile:
+                outfile.write(cleaned_content)
+
+def print_analysis_results(results: Dict[str, int], stage: str) -> None:
+    """Print analysis results in a formatted way."""
+    print(f"\nAnalysis results ({stage}):")
+    print(f"- Total duplicates: {results['duplicates']}")
+    print(f"- Total hyphenated words: {results['hyphenated']}")
+
+def main():
+    # Get paths from config
+    paths = DUPLICATES_CONFIG["paths"]
+    
+    # Validate input paths and create output directories
+    input_errors = validate_input_data({"transcripts": paths["input"]})
+    if input_errors:
+        for desc, error in input_errors.items():
+            print(f"Error with {desc}: {error}")
+        return
+    
+    print(f"\nInput directory: {paths['input']}")
+    print(f"Output directory: {paths['output']}")
+    print(f"Input directory exists? {os.path.exists(paths['input'])}")
+    
+    ensure_output_dir(paths["output"])
+    print(f"Output directory exists after ensure? {os.path.exists(paths['output'])}")
+    
+    # Analyze original transcripts
+    print("\nAnalyzing original transcripts...")
+    original_results = analyze_directory(paths["input"])
+    print_analysis_results(original_results, "before cleaning")
+    
+    # Process transcripts
+    print("\nCleaning transcripts...")
+    process_directory(paths["input"], paths["output"])
+    
+    # Analyze cleaned transcripts
+    print("\nAnalyzing cleaned transcripts...")
+    cleaned_results = analyze_directory(paths["output"])
+    print_analysis_results(cleaned_results, "after cleaning")
+    
+    print(f"\nCleaned transcripts saved to: {paths['output']}")
+
+if __name__ == "__main__":
+    main()