#Mice Analysis

##Descriptive Statistics ###Script Overview

The script (descriptive_statistics.py) processes behavioral data from a CSV file and generates four types of visualizations to analyze performance across subjects (mice) in a reward-based task. The main functions are:

plot_performance_comparison

• Purpose: Compares performance between good and poor performers using a boxplot for good performers and scattered points for poor performers.
• Inputs:
  • df_data: DataFrame with columns subject (subject identifier) and perf (performance values, 0-1).
  • chance_level: Threshold to classify good/poor performers (default: 0.55).
• Features:
  • Boxplot for good performers (performance ≥ chance_level) with styled whiskers, caps, and medians.
  • Jittered scatter points for both good and poor performers, with distinct colors (blue for good, red for poor).
  • Reference line at the chance level.
  • Custom legend and statistical annotation (Mann-Whitney U test p-value) if both groups exist.
  • Large fonts and high contrast for presentation readability.

plot_performance_psychometric

• Purpose: Plots psychometric curves showing performance as a function of reward probability for each subject and the group average.
• Inputs:
  • combined_data: DataFrame with columns subject, perf, and probability_r (probability of reward for right choice).
• Features:
  • Individual subject curves with performance plotted against reward probability, excluding the 0.5 probability condition.
  • Group average curve in black for emphasis.
  • Ideal performance line (y=x) and chance level references (horizontal and vertical at 0.5).
  • viridis color palette for subjects, with a custom legend placed outside the plot.
  • Handles invalid performance values (-1) and transforms left-side probabilities to right-equivalent for consistency.

outcome_block_change

• Purpose: Analyzes and visualizes performance around block transitions (changes in reward probability) or choice switches, showing trial-by-trial correctness.
• Inputs:
  • df: DataFrame with columns subject, choice (0=left, 1=right), and probability_r.
  • window: Number of trials to show before/after the event.
  • switch_type: 0 for block transitions, 1 for choice switches.
• Features:
  • Plots performance (fraction correct) aligned to switch events for each subject, with SEM error bars.
  • Identifies switch points based on changes in probability (switch_type=0) or choice (switch_type=1).
  • Includes reference lines for the event (vertical at lag 0) and chance level (horizontal at 0.5).
  • Uses viridis colormap for subjects, with a legend placed outside the plot.
  • Returns an aggregated DataFrame for further analysis.

plot_blocks

• Purpose: Visualizes the reward probability for the right choice across trials for a specific session and subject.
• Inputs:
  • df: DataFrame with columns session, subject, trial, and probability_r.
• Features:
  • Plots reward probability dynamics for a single session (hardcoded: session 129, subject 'B1').
  • Includes a chance level reference line at 0.5.
  • Optimized for slides with a clean design: Arial font, thicker lines, visible grid, and no top/right spines.
  • High-contrast colors and large fonts for readability.

Main Execution

• Data Loading:
  • Reads a CSV file (global_trials1.csv) with behavioral data, using ; as the separator and specifying iti_duration as float.
  • Applies a custom parsing function to filter data (e.g., for trained mice, trained=1).
• Data Preprocessing:
  • Creates a choice column based on outcome and side (correct/incorrect and left/right).
  • Computes a perf column (1 for correct choices based on probability_r, 0 otherwise).
• Visualization:
  • Calls all four plotting functions in sequence: plot_blocks, plot_performance_comparison, plot_performance_psychometric, and outcome_block_change (with switch_type=1 and window=10).

Usage

1. Ensure the required CSV file (global_trials1.csv) is available at the specified path.
2. Verify that the extra_plotting, model_avaluation, and parsing modules are accessible.
3. Update the data_path variable if needed.
4. The script will generate and display four plots:
   • Reward probability dynamics for a specific session/subject.
   • Performance comparison between good and poor performers.
   • Psychometric curves for all subjects.
   • Performance around choice switches.

Notes

• The script assumes a specific CSV structure with columns like subject, session, outcome, side, probability_r, and iti_duration. Ensure your data matches this format.
• The plot_blocks function is hardcoded for session 129 and subject 'B1'; modify the filtering condition (df[(df['session'] == 129) & (df['subject'] == 'B1')]) for other sessions/subjects.
• The parsing function is not provided; it must handle data filtering for trained/untrained mice.
• Visualization parameters (e.g., font sizes, figure dimensions) are optimized for presentations but can be adjusted in the params dictionaries.

##Model Performance Visualization ###Script overview The script (metrics_plots.py) is designed to load previously stored model performance data from CSV files and create comparative visualizations grouped by models. It includes two main functions:

####plot_metrics_comparison_grouped_by_models

• Purpose: Generates five separate plots to compare model performance across different metrics: Log Likelihood, Number of Trials, BIC (Bayesian Information Criterion), Log Likelihood per Observation, and Accuracy.
• Inputs:
  • blocks: A list of experimental conditions (e.g., probability blocks like [[0.2,0.8], [0.3,0.7]]).
  • metrics_data: A nested dictionary containing metric values for each model (e.g., {model_name: {metric_name: [values_across_blocks]}}).
  • model_names: A list of model identifiers (e.g., ['glm_prob_switch', 'glm_prob_r']).
• Features:
  • Uses a consistent color scheme (viridis palette) across all plots.
  • Employs large, readable fonts suitable for presentations.
  • Automatically adjusts layout to prevent overlap.
  • Each plot visualizes one metric, grouping data by models and coloring by experimental conditions.

plot_metric_grouped_by_models

• Purpose: Creates a single grouped comparison plot for a specific metric, using boxplots and stripplots to show distribution and individual data points.
• Inputs:
  • blocks: Same as above.
  • metrics_data: Same as above.
  • model_names: Same as above.
  • metric: The specific metric to plot (e.g., 'log_likelihood').
  • ylabel: Y-axis label for the plot.
  • palette: Color palette for different blocks.
  • Font size parameters for title, labels, legend, and ticks.
• Features:
  • Boxplots show the distribution of metric values per model and block.
  • Stripplots overlay individual data points with slight jitter for clarity.
  • Custom legend placement outside the plot to avoid overlap.
  • Special handling for model-specific block labeling (e.g., different block indices for inference_based vs. other models).
  • Grid and despined axes for clean visualization.

####Main Execution

• Data Loading:
  • Reads CSV files containing model metrics (e.g., all_subjects_glm_metrics_{model}_{block}.csv).
  • Aggregates data by taking the median across subjects for each metric.
  • Handles missing files or errors by appending empty arrays.
• Model Configuration:
  • Supports multiple models (e.g., glm_prob_switch, glm_prob_r, inference_based).
  • Uses different block configurations for standard and special models.
• Visualization:
  • Calls plot_metrics_comparison_grouped_by_models to generate plots for selected models.
  • Includes error handling to report issues during plot generation.

Usage

1. Place the script in a directory containing the required CSV files with model metrics.

2. Update the main_folder and results_base paths in the script to match your file locations.

3. Modify models_to_analyze and models_to_plot to include the desired models.

4. The script will generate and display five plots comparing the specified metrics across models.

Notes

• Ensure CSV files follow the expected format with columns for subject, log_likelihood, log_likelihood_per_obs, BIC, AIC, and accuracy.
• The script assumes median aggregation per subject; modify the df.groupby('subject').median() line to use all data points if needed.

Font sizes and figure dimensions are optimized for presentations but can be adjusted in the function parameters.

##GLM-Right

Script Overview

The script (glm_right.py) processes behavioral data from a CSV file and includes functions to compute choice-related regressors, fit GLMs, and generate plots to visualize model coefficients. The main functions are:

obt_regressors

• Purpose: Processes trial-by-trial data to compute regressors for the GLM and generates the regression formula.
• Inputs:
  • df: DataFrame with columns like session, outcome, side, iti_duration, probability_r.
  • n: Number of previous trials to consider (n_back).
• Features:
  • Filters sessions with >50 trials.
  • Encodes choices (left, right, other) based on outcome and side.
  • Creates choice_num (1=right, 0=left) for the dependent variable.
  • Computes regressors:
    • r_plus: 1 for rewarded right choice, -1 for rewarded left choice, 0 otherwise.
    • r_minus: 1 for unrewarded right choice, -1 for unrewarded left choice, 0 otherwise.
  • Builds regressors for past trials (1 to n-1): r_plus_{i}, r_minus_{i}.
  • Constructs the GLM formula (e.g., r_plus_1 + r_minus_1 + r_plus_2 + r_minus_2).
• Returns:
  • Processed DataFrame with regressors.
  • Regression formula string.

plot_all_mice_glm_combined

• Purpose: Creates a single plot showing GLM coefficients for all mice, optimized for A0 poster presentation.
• Inputs:
  • df: DataFrame with behavioral data.
  • n_back: Number of previous trials to consider.
  • figsize: Figure size (default: A0 poster size, 46.8x33.1 inches).
• Features:
  • Excludes mouse 'A10'.
  • Uses 5-fold cross-validation to fit logistic regression models (choice_num ~ regressors).
  • Computes coefficients, p-values, and evaluation metrics (log-likelihood, AIC, BIC, pseudo R-squared, accuracy, precision, recall, F1, ROC AUC, Brier score).
  • Saves metrics to a CSV file (all_subjects_glm_metrics_glm_prob_r_{n_back}.csv).
  • Plots coefficients for:
    • r_plus (red, solid line with circles).
    • r_minus (blue, dashed line with squares).
    • Intercept (green, circle at x=0).
  • Uses an alpha gradient to differentiate mice.
  • Adds significance markers (***, **, *, ns) based on Fisher’s combined p-values across mice.
  • Includes a custom legend and grid for readability.
  • Optimized for posters with large fonts and thick lines.

glm

• Purpose: Plots GLM coefficients separately for each mouse in a subplot grid, predicting right-side choices.
• Inputs:
  • df: DataFrame with behavioral data.
• Features:
  • Creates a subplot grid (2 rows, dynamic columns) for individual mice.
  • Filters sessions with >50 trials.
  • Uses obt_regressors with n_back=10 to compute regressors.
  • Fits logistic regression (choice_num ~ regressors) for each mouse using 5-fold cross-validation.
  • Plots median coefficients for:
    • Intercept (green).
    • r_plus (red).
    • r_minus (blue).

Main Execution

• Data Loading:
  • Reads global_trials1.csv with ; separator, specifying iti_duration as float.
  • Prints unique tasks for reference.
• Data Preprocessing:
  • Applies a custom parsing function for trained mice (trained=1, opto_yes=0).
• Visualization:
  • Defaults to separate plots for each mouse (separate_mice=True) by calling glm.
  • Optionally generates combined plots for all mice with varying n_back values ([2,3,4,7,10]) if separate_mice=False.

Usage

1. Ensure the required CSV file (global_trials1.csv) is available at the specified path.
2. Verify that the extra_plotting and parsing modules are accessible.
3. Update the data_path variable if needed.
4. The script will generate and display:
   • A subplot grid with GLM coefficients for each mouse (default, separate_mice=True).
   • Optionally, A0-sized combined plots for all mice with different n_back values (if separate_mice=False).

Notes

• The script assumes a specific CSV structure with columns like subject, session, outcome, side, iti_duration, probability_r, and task.
• The parsing and select_train_sessions functions are not provided; they must handle data filtering and cross-validation split creation.
• The n_back=10 is hardcoded in glm; adjust as needed.
• Visualization parameters (e.g., font sizes, figure dimensions) are optimized for presentations/posters but can be modified in plt.rcParams or function arguments.
• The script excludes mouse 'A10' and sessions with ≤50 trials.
• Metrics are saved to a CSV file, overwriting any existing file with the same name.

##GLM-Switch ###Script Overview

The script (glm_switch.py) processes behavioral data from a CSV file and includes functions to compute switch-related regressors, fit GLMs, and generate plots to visualize model coefficients. The main functions are:

obt_regressors

• Purpose: Processes trial-by-trial behavioral data to compute regressors for the GLM and generates the regression formula.
• Inputs:
  • df: DataFrame with columns like session, outcome, side, iti_duration, probability_r.
  • n: Number of previous trials to consider (n_back).
• Features:
  • Filters sessions with >50 trials.
  • Encodes choices (left, right, other) based on outcome and side.
  • Computes switch_num (0=same choice as previous trial, 1=different).
  • Creates last_trial regressor (reward outcome of the previous trial).
  • Builds regressors for past trials (2 to n):
    • rss_plus{i}: 1 if same choice as previous trial and rewarded at lag i.
    • rss_minus{i}: 1 if same choice as previous trial and unrewarded at lag i.
    • rds_plus{i}: 1 if different choice from previous trial and rewarded at lag i.
  • Constructs the GLM formula string (e.g., rss_plus2 + rss_minus2 + rds_plus2 + last_trial).
• Returns:
  • Processed DataFrame with regressors.
  • Regression formula string.

plot_all_mice_glm_combined

• Purpose: Creates a single plot showing GLM coefficients for all mice, optimized for A0 poster presentation.
• Inputs:
  • df: DataFrame with behavioral data.
  • n_back: Number of previous trials to consider.
  • figsize: Figure size (default: A0 poster size, 46.8x33.1 inches).
• Features:
  • Excludes mouse 'A10'.
  • Uses 5-fold cross-validation to fit logistic regression models (switch_num ~ regressors).
  • Computes coefficients, p-values, and evaluation metrics (log-likelihood, AIC, BIC, pseudo R-squared, accuracy, precision, recall, F1, ROC AUC, Brier score).
  • Saves metrics to a CSV file (all_subjects_glm_metrics_glm_prob_switch_{n_back}.csv).
  • Plots coefficients for:
    • rss_plus (red, solid line with circles).
    • rss_minus (blue, dashed line with squares).
    • rds_plus (orange, dashed line with circles).
    • last_trial (gray, circle).
    • Intercept (green, circle).
  • Uses an alpha gradient to differentiate mice.
  • Adds significance markers (***, **, *, ns) based on Fisher’s combined p-values across mice.
  • Includes a custom legend with regressor types and a grid for readability.
  • Optimized for posters with large fonts and thick lines.

glm

• Purpose: Plots GLM coefficients either combined across all mice or separately for each mouse, focusing on switch behavior.
• Inputs:
  • df: DataFrame with behavioral data.
• Features:
  • Creates a subplot grid (2 rows, dynamic columns) for individual mice.
  • Filters sessions with >50 trials.
  • Uses obt_regressors with n_back=10 to compute regressors.
  • Fits logistic regression (switch_num ~ regressors) for each mouse using 5-fold cross-validation.
  • Plots median coefficients for:
    • Intercept (green).
    • rss_plus (red).
    • rss_minus (blue).
    • rds_plus (orange).
    • Other regressors (gray).
  • Adds significance markers (***, **, *, ns) based on median p-values.
  • Adjusts x-tick positions and labels for clarity (e.g., special handling for lag 10).
  • Shares y-axis across subplots for consistency.
  • Optimized for readability with dynamic subplot sizing.

Main Execution

• Data Loading:
  • Reads global_trials1.csv with ; separator, specifying iti_duration as float.
  • Prints unique tasks for reference.
• Data Preprocessing:
  • Applies a custom parsing function for trained mice (trained=1, opto_yes=0).
• Visualization:
  • Defaults to separate plots for each mouse (separate_mice=True) by calling glm.
  • Optionally generates combined plots for all mice with varying n_back values ([2,3,4,7,10]) if separate_mice=False.

Usage

1. Ensure the required CSV file (global_trials1.csv) is available at the specified path.
2. Verify that the extra_plotting, model_avaluation, and parsing modules are accessible.
3. Update the data_path variable if needed.
4. The script will generate and display:
   • A subplot grid with GLM coefficients for each mouse (default, separate_mice=True).
   • Optionally, A0-sized combined plots for all mice with different n_back values (if separate_mice=False).

Notes

• The script assumes a specific CSV structure with columns like subject, session, outcome, side, iti_duration, probability_r, and task.
• The parsing and select_train_sessions functions are not provided; they must handle data filtering and cross-validation split creation.
• The n_back=10 is hardcoded in glm; adjust as needed.
• Visualization parameters (e.g., font sizes, figure dimensions) are optimized for presentations/posters but can be modified in plt.rcParams or function arguments.
• The script excludes mouse 'A10' and sessions with ≤50 trials.
• Metrics are saved to a CSV file, overwriting any existing file with the same name.

##Inference-Based Script

Script Overview

The script (inference-based.py) processes behavioral data from a CSV file and includes functions to compute inference-based features (V_t) and visualize logistic regression coefficients for predicting mouse choices. The main functions are:

manual_computation

• Purpose: Processes trial-by-trial behavioral data to compute value differences (V_t) and choice-reward sequences based on past trials.
• Inputs:
  • df: DataFrame with columns like subject, session, outcome, side, probability_r.
  • n_back: Number of previous trials to consider for sequence patterns.
  • hist: Boolean flag to plot a histogram of computed V_t values.
• Features:
  • Encodes choices (0=left, 1=right) based on outcome and side.
  • Creates choice-reward codes (00, 01, 10, 11) combining choice and reward outcome.
  • Builds sequences of n_back previous choice-reward pairs.
  • Computes V_t as the difference between the mean probability of the right side being active (prob_right) and the left side being active (prob_left) for each sequence.
  • Optionally plots a histogram of V_t values.
  • Prepares next-trial choice data for modeling.
• Returns: Processed DataFrame with computed features (choice, sequence, V_t, etc.).

manual_computation_v2

• Purpose: Computes V_t using recursive equations based on reward history and given probabilities, suitable for an alternative inference model.
• Inputs:
  • df: DataFrame with trial data.
  • p_SW: Probability of switching from active to inactive state.
  • p_RWD: Probability of reward in the active state.
  • hist: Boolean flag to plot a histogram of V_t values.
• Features:
  • Encodes choices similarly to manual_computation.
  • Tracks same-site choices across trials.
  • Computes R_t (reward history) recursively:
    • Resets to 0 on reward.
    • Updates using rho = 1 / ((1 - p_SW) * (1 - p_RWD)) for same-site unrewarded trials.
    • Maintains previous R_t for exploratory switches.
  • Computes V_t based on R_t, side_num, and p_RWD, with warnings for values exceeding 1.
  • Optionally plots a histogram of V_t values.
• Returns: DataFrame with computed R_t, V_t, and other intermediate columns.

plot_all_mice_correct_inf_combined

• Purpose: Creates a single plot showing logistic regression coefficients for all mice, optimized for A0 poster presentation.
• Inputs:
  • df: DataFrame with behavioral data for all mice.
  • n_back: Number of previous trials for sequence patterns.
  • figsize: Figure size (default: A0 poster size, 46.8x33.1 inches).
• Features:
  • Excludes mouse 'A10' from analysis.
  • Uses 5-fold cross-validation to fit logistic regression models (choice ~ V_t + side_num or choice ~ side_num for v2=2).
  • Computes coefficients, standard errors, p-values, and confidence intervals.
  • Calculates comprehensive metrics (log-likelihood, AIC, BIC, pseudo R-squared, accuracy, precision, recall, F1, ROC AUC, Brier score).
  • Saves metrics to a CSV file (path based on v2 and n_back).
  • Plots coefficients (β^V, β^S, Intercept) with distinct colors and significance markers (***, **, *, ns) based on Stouffer's combined p-values.
  • Uses an alpha gradient to differentiate mice in the legend.
  • Optimized for poster presentation with large fonts and thick lines.

inference_plot

• Purpose: Plots logistic regression coefficients either combined across all mice or separately for each mouse.
• Inputs:
  • df: DataFrame with behavioral data.
• Features:
  • Supports two modes:
    • Combined: Calls plot_all_mice_correct_inf_combined with n_back=3 (hardcoded).
    • Separate: Creates a subplot grid (2 rows, dynamic columns) for individual mice.
  • For separate plots:
    • Filters sessions with >50 trials.
    • Uses manual_computation with n_back=3 to compute V_t.
    • Fits logistic regression (choice ~ V_t + side_num) for each mouse using 5-fold cross-validation.
    • Plots median coefficients (Intercept, side_num, V_t) with significance markers.
    • Uses distinct colors for regressors and rotates x-labels if needed.
  • Shares y-axis across subplots for consistency.
  • Optimized for readability with dynamic subplot sizing.

Main Execution

• Data Loading:
  • Reads global_trials1.csv with ; separator, specifying iti_duration as float.
  • Shifts iti_duration to the next trial and excludes subject='manual'.
• Data Preprocessing:
  • Selects relevant columns (subject, session, outcome, side, iti_duration, probability_r, task, date).
  • Encodes outcome_bool (1 for correct, 0 otherwise).
  • Applies a custom parsing function for trained mice (trained=1, opto_yes=0).
• Visualization:
  • Calls inference_plot to generate plots (default: separate plots for each mouse).

Usage

1. Ensure the required CSV file (global_trials1.csv) is available at the specified path.
2. Verify that the extra_plotting, model_avaluation, and parsing modules are accessible.
3. Update the data_path variable if needed.
4. The script will generate and display either:
   • A single A0-sized plot with combined coefficients for all mice (if separate_mice=False).
   • A subplot grid with coefficients for each mouse (if separate_mice=True).

Notes

• The script assumes a specific CSV structure with columns like subject, session, outcome, side, probability_r, iti_duration, task, and date.
• The parsing and select_train_sessions functions are not provided; they must handle data filtering and cross-validation split creation.
• The v2 flag in plot_all_mice_correct_inf_combined controls the computation method (manual_computation or manual_computation_v2) and model formula.
• The n_back=3 is hardcoded in inference_plot for combined plots; adjust as needed.
• Visualization parameters (e.g., font sizes, figure dimensions) are optimized for presentations/posters but can be modified in plt.rcParams or function arguments.
• The script excludes mouse 'A10' and sessions with ≤50 trials in inference_plot.