diffindiff 2.5.4__tar.gz → 2.5.6__tar.gz

{diffindiff-2.5.4 → diffindiff-2.5.6}/PKG-INFO

@@ -1,10 +1,11 @@
 Metadata-Version: 2.1
 Name: diffindiff
-Version: 2.5.4
+Version: 2.5.6
 Summary: diffindiff: Python library for convenient Difference-in-Differences analyses
 Author: Thomas Wieland
 Author-email: geowieland@googlemail.com
 Description-Content-Type: text/markdown
+Provides-Extra: optional
 
 # diffindiff: Python library for convenient Difference-in-Differences analyses
 
@@ -29,7 +30,7 @@ A case study that utilizes the diffindiff library is available on [arXiv](https:
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.4) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.6) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -177,15 +178,11 @@ See the /tests directory for usage examples of most of the included functions.
 
 ## AI Usage Statement
 
-This software was developed without the use of AI-generated code. The Continue Agent in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
+This software was developed without the use of AI-generated code. The GitHub Copilot Chat in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
 
 
-## What's new (v2.5.4)
+## What's new (v2.5.6)
 
 - Bugfixes
-  - Correction of the nonsensical skipping of the treatment group dummy in the case of two observation units in didanalysis.did_analysis()
-  - Automatical drop of duplicates in results dictionaries in didanalysis_helper.extract_model_results()
-  - Inefficient successive data insert in diddata.DiffData.add_synthetic() replaced by efficient concatenating
-- Other
-  - Test script extended by another example
-  - Cleaned dependencies
+  - didtools.is_parallel(): Checking whether there is a pre-treatment period and skipping test if not (additional NOTE) 
+  - Optional installation of XGBoost and LightGBM (more stable if problems with the installation of these packages occur)

{diffindiff-2.5.4 → diffindiff-2.5.6}/README.md

@@ -21,7 +21,7 @@ A case study that utilizes the diffindiff library is available on [arXiv](https:
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.4) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.6) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -169,15 +169,11 @@ See the /tests directory for usage examples of most of the included functions.
 
 ## AI Usage Statement
 
-This software was developed without the use of AI-generated code. The Continue Agent in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
+This software was developed without the use of AI-generated code. The GitHub Copilot Chat in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
 
 
-## What's new (v2.5.4)
+## What's new (v2.5.6)
 
 - Bugfixes
-  - Correction of the nonsensical skipping of the treatment group dummy in the case of two observation units in didanalysis.did_analysis()
-  - Automatical drop of duplicates in results dictionaries in didanalysis_helper.extract_model_results()
-  - Inefficient successive data insert in diddata.DiffData.add_synthetic() replaced by efficient concatenating
-- Other
-  - Test script extended by another example
-  - Cleaned dependencies
+  - didtools.is_parallel(): Checking whether there is a pre-treatment period and skipping test if not (additional NOTE) 
+  - Optional installation of XGBoost and LightGBM (more stable if problems with the installation of these packages occur)

{diffindiff-2.5.4 → diffindiff-2.5.6}/diffindiff/config.py

@@ -4,15 +4,15 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     1.0.25
-# Last update: 2026-08-04 18:17
+# Version:     1.0.27
+# Last update: 2026-10-03 11:03
 # Copyright (c) 2025-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
 # Basic config:
 
 PACKAGE_NAME = "diffindiff"
-PACKAGE_VERSION = "2.5.4"
+PACKAGE_VERSION = "2.5.6"
 
 VERBOSE = False
 

{diffindiff-2.5.4 → diffindiff-2.5.6}/diffindiff/didanalysis.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     2.4.5
-# Last update: 2026-08-03 20:25
+# Version:     2.4.6
+# Last update: 2026-09-11 21:46
 # Copyright (c) 2024-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
@@ -1875,8 +1875,12 @@ class DiffModel:
                 fig, ax = plt.subplots(figsize=(12, 6))
         else:
             fig, ax = plt.subplots(figsize=(plot_size[0], plot_size[1]))       
-    
-        model_data_TG_CG["t"] = pd.to_datetime(model_data_TG_CG["t"])
+
+        date_format = model_config["date_format"]
+        model_data_TG_CG["t"] = pd.to_datetime(
+            model_data_TG_CG["t"],
+            format = date_format
+            )
 
         if not model_config["pre_post"]:
             pre_post_barplot = False
@@ -2022,7 +2026,7 @@ class DiffModel:
             plt.xlabel(x_label)
             plt.ylabel(y_label)
             plt.title(plot_title)
-            ax.xaxis.set_major_formatter(DateFormatter(model_config["date_format"]))
+            ax.xaxis.set_major_formatter(DateFormatter(date_format))
 
         if model_config["pre_post"]:
 
@@ -2222,7 +2226,12 @@ class DiffModel:
             model_data_TG_mean_pred_counterfac.reset_index()
             ],
             axis = 1)
-        model_data_TG_mean[time_col] = pd.to_datetime(model_data_TG_mean[time_col])
+
+        date_format = model_config["date_format"]
+        model_data_TG_mean[time_col] = pd.to_datetime(
+            model_data_TG_mean[time_col],
+            format=date_format
+            )
 
         fig, ax = plt.subplots(figsize=(plot_size[0], plot_size[1]))   
         
@@ -2725,13 +2734,15 @@ def did_analysis(
         print(f"NOTE: Input is {config.MULTIPERIOD_PANELDATA_DESCRIPTION}. Consider including {config.EFFECTS_TYPES['FE']['description']}.")
         
     if log_outcome:
-        
-        if missing_replace_by_zero:
-            data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"] = np.log(data[outcome_col]+log_outcome_add)
-        
+
+        if any(data[outcome_col] <= 0):
+            data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"] = np.log(data[outcome_col]+log_outcome_add)        
         else:
             data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"] = np.log(data[outcome_col])
-        
+
+        if not np.isfinite(data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"]).all():
+            print(f"WARNING: Outcome column '{outcome_col}' in input data contains values <= 0, which is not fully compensated by log_outcome_add = {log_outcome_add}. This may lead to NaN Inf values in the log-transformed outcome. Consider using a larger constant for log_outcome_add.")
+
         outcome_col = f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"
 
     if not ITE and not GTE:
@@ -3320,12 +3331,14 @@ def ddd_analysis(
         pre_post = True
         
     if log_outcome:
-        
-        if missing_replace_by_zero:
-            data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"] = np.log(data[outcome_col]+log_outcome_add)
-        
+
+        if any(data[outcome_col] <= 0):
+            data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"] = np.log(data[outcome_col]+log_outcome_add)        
         else:
             data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"] = np.log(data[outcome_col])
+
+        if not np.isfinite(data[f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"]).all():
+            print(f"WARNING: Outcome column '{outcome_col}' in input data contains values <= 0, which is not fully compensated by log_outcome_add = {log_outcome_add}. This may lead to NaN Inf values in the log-transformed outcome. Consider using a larger constant for log_outcome_add.")
         
         outcome_col = f"{config.LOG_PREFIX}{config.DELIMITER}{outcome_col}"
 

{diffindiff-2.5.4 → diffindiff-2.5.6}/diffindiff/didtools.py

@@ -4,8 +4,8 @@
 # Author:      Thomas Wieland 
 #              ORCID: 0000-0001-5168-9846
 #              mail: geowieland@googlemail.com              
-# Version:     2.2.5
-# Last update: 2026-07-23 19:34
+# Version:     2.2.6
+# Last update: 2026-10-03 11:02
 # Copyright (c) 2025-2026 Thomas Wieland
 #-----------------------------------------------------------------------
 
@@ -22,11 +22,20 @@ from sklearn.svm import SVR
 from sklearn.neighbors import KNeighborsRegressor
 from sklearn.pipeline import Pipeline
 from sklearn.preprocessing import StandardScaler
-from xgboost import XGBRegressor
-from lightgbm import LGBMRegressor
 from sklearn.linear_model import LinearRegression
 from sklearn.model_selection import train_test_split
 from sklearn.neural_network import MLPRegressor
+
+try:
+    from xgboost import XGBRegressor
+except ImportError:
+    XGBRegressor = None
+try:
+    from lightgbm import LGBMRegressor
+except ImportError:
+    LGBMRegressor = None
+
+
 import diffindiff.config as config
 
 
@@ -826,83 +835,105 @@ def is_parallel(
         treatment_col = treatment_col,
         verbose = False
         )
-    
-    if verbose:
-        print(f"Testing outcome '{outcome_col}' for parallel time trends", end = " ... ")
-    
-    if pre_post or not modeldata_isnotreatment:
-        parallel = "not_tested"
-        test_ols_model = None
-    
-    treatment_group = modeldata_isnotreatment[1]
 
-    if config.ACCEPT_CONTINUOUS_TREATMENTS:
-        
-        if len(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] > 0)]) > 0:
-        
-            first_day_of_treatment = min(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] > 0)][time_col])
-            
-            data_test = data[data[time_col] < first_day_of_treatment].copy()
-            data_test[config.TG_COL] = 0
-            data_test.loc[data_test[unit_col].isin(treatment_group), config.TG_COL] = 1
-            
-            if config.TIME_COUNTER_COL not in data_test.columns:
-                data_test = date_counter(
-                    df = data_test,
-                    date_col = time_col, 
-                    new_col = config.TIME_COUNTER_COL,
-                    verbose = False
-                    )
-            data_test[f"{config.TG_COL}_x_{config.TIME_COL}"] = data_test[config.TG_COL]*data_test[config.TIME_COUNTER_COL]        
+    parallel = "not_tested"
+    test_ols_model = None
 
-            test_ols_model = ols(f'{outcome_col} ~ {config.TG_COL} + {config.TIME_COUNTER_COL} + {config.TG_COL}_x_{config.TIME_COL}', data = data_test).fit()
-            coef_TG_x_t_p = test_ols_model.pvalues[f"{config.TG_COL}_x_{config.TIME_COL}"]
+    if not pre_post:
 
-            if coef_TG_x_t_p < alpha:
-                parallel = False
-            else:
-                parallel = True
+        no_pre_period = False
         
-        else:
-            parallel = "not_tested"
-            test_ols_model = None
+        if verbose:
+            print(f"Testing outcome '{outcome_col}' for parallel time trends", end = " ... ")
         
-    else:
+        treatment_group = modeldata_isnotreatment[1]
 
-        if len(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] == 1)]) > 0:
-            
-            first_day_of_treatment = min(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] == 1)][time_col])
+        if config.ACCEPT_CONTINUOUS_TREATMENTS:
             
-            data_test = data[data[time_col] < first_day_of_treatment].copy()
-            data_test[config.TG_COL] = 0
-            data_test.loc[data_test[unit_col].isin(treatment_group), config.TG_COL] = 1
+            if len(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] > 0)]) > 0:
             
-            if config.TIME_COUNTER_COL not in data_test.columns:
-                data_test = date_counter(
-                    df = data_test,
-                    date_col = time_col, 
-                    new_col = config.TIME_COUNTER_COL,
-                    verbose = False
-                    )
-            data_test[f"{config.TG_COL}_x_{config.TIME_COL}"] = data_test[config.TG_COL]*data_test[config.TIME_COUNTER_COL]        
+                first_day_of_treatment = min(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] > 0)][time_col])
+                
+                data_test = data[data[time_col] < first_day_of_treatment].copy()
+                data_test[config.TG_COL] = 0
+                data_test.loc[data_test[unit_col].isin(treatment_group), config.TG_COL] = 1
+                
+                if config.TIME_COUNTER_COL not in data_test.columns:
+                    data_test = date_counter(
+                        df = data_test,
+                        date_col = time_col, 
+                        new_col = config.TIME_COUNTER_COL,
+                        verbose = False
+                        )
+                data_test[f"{config.TG_COL}_x_{config.TIME_COL}"] = data_test[config.TG_COL]*data_test[config.TIME_COUNTER_COL]
+
+                if len(data_test) > 0:
+
+                    test_ols_model = ols(f'{outcome_col} ~ {config.TG_COL} + {config.TIME_COUNTER_COL} + {config.TG_COL}_x_{config.TIME_COL}', data = data_test).fit()
+                    coef_TG_x_t_p = test_ols_model.pvalues[f"{config.TG_COL}_x_{config.TIME_COL}"]
 
-            test_ols_model = ols(f'{outcome_col} ~ {config.TG_COL} + {config.TIME_COUNTER_COL} + {config.TG_COL}_x_{config.TIME_COL}', data = data_test).fit()
-            coef_TG_x_t_p = test_ols_model.pvalues[f"{config.TG_COL}_x_{config.TIME_COL}"]
+                    if coef_TG_x_t_p < alpha:
+                        parallel = False
+                    else:
+                        parallel = True
 
-            if coef_TG_x_t_p < alpha:
-                parallel = False
+                else:
+                    no_pre_period = True
+            
             else:
-                parallel = True
+                parallel = "not_tested"
+                test_ols_model = None
             
         else:
-            parallel = "not_tested"
-            test_ols_model = None
-    
-    if verbose:
-        print("OK")
 
-    if parallel == "not_tested":
-        print("WARNING: Data could not be tested for parallel time trends.")
+            if len(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] == 1)]) > 0:
+                
+                first_day_of_treatment = min(data[(data[unit_col].isin(treatment_group)) & (data[treatment_col] == 1)][time_col])
+                
+                data_test = data[data[time_col] < first_day_of_treatment].copy()
+                data_test[config.TG_COL] = 0
+                data_test.loc[data_test[unit_col].isin(treatment_group), config.TG_COL] = 1
+                
+                if config.TIME_COUNTER_COL not in data_test.columns:
+                    data_test = date_counter(
+                        df = data_test,
+                        date_col = time_col, 
+                        new_col = config.TIME_COUNTER_COL,
+                        verbose = False
+                        )
+                data_test[f"{config.TG_COL}_x_{config.TIME_COL}"] = data_test[config.TG_COL]*data_test[config.TIME_COUNTER_COL]
+
+                if len(data_test) > 0:
+
+                    test_ols_model = ols(f'{outcome_col} ~ {config.TG_COL} + {config.TIME_COUNTER_COL} + {config.TG_COL}_x_{config.TIME_COL}', data = data_test).fit()
+                    coef_TG_x_t_p = test_ols_model.pvalues[f"{config.TG_COL}_x_{config.TIME_COL}"]
+
+                    if coef_TG_x_t_p < alpha:
+                        parallel = False
+                    else:
+                        parallel = True
+
+                else:
+                    no_pre_period = True
+                
+            else:
+                parallel = "not_tested"
+                test_ols_model = None
+        
+        if verbose:
+            print("OK")
+
+    if not pre_post:
+
+        if parallel == "not_tested":
+            print("WARNING: Data could not be tested for parallel time trends.")
+        if no_pre_period:
+            print("WARNING: Data could not be tested for parallel time trends because there is no pre-treatment period.")
+
+    else:
+
+        if verbose:
+            print("NOTE: Data is pre-post data and parallel trends are not tested.")
       
     return [
         parallel, 
@@ -1549,18 +1580,28 @@ def model_wrapper(
             model = SVR(kernel=svr_kernel)
             
         elif model_type == "xgb":
-            model = XGBRegressor(
-                learning_rate = xgb_learning_rate,
-                n_estimators = gb_iterations,
-                random_state = random_state
-            )
+
+            if XGBRegressor is not None:
+                model = XGBRegressor(
+                    learning_rate = xgb_learning_rate,
+                    n_estimators = gb_iterations,
+                    random_state = random_state
+                )
+            else:
+                model_estimation_error = True
+                model_estimation_error_text = "XGBRegressor is not available. Please install xgboost to use this model type."
             
         elif model_type == "lgbm":
-            model = LGBMRegressor(
-                learning_rate = lgbm_learning_rate,
-                n_estimators = gb_iterations,
-                random_state = random_state
-            )
+
+            if LGBMRegressor is not None:
+                model = LGBMRegressor(
+                    learning_rate = lgbm_learning_rate,
+                    n_estimators = gb_iterations,
+                    random_state = random_state
+                )
+            else:
+                model_estimation_error = True
+                model_estimation_error_text = "LGBMRegressor is not available. Please install lightgbm to use this model type."
             
         elif model_type == "mlp":
             model = Pipeline(

{diffindiff-2.5.4 → diffindiff-2.5.6}/diffindiff.egg-info/PKG-INFO

@@ -1,10 +1,11 @@
 Metadata-Version: 2.1
 Name: diffindiff
-Version: 2.5.4
+Version: 2.5.6
 Summary: diffindiff: Python library for convenient Difference-in-Differences analyses
 Author: Thomas Wieland
 Author-email: geowieland@googlemail.com
 Description-Content-Type: text/markdown
+Provides-Extra: optional
 
 # diffindiff: Python library for convenient Difference-in-Differences analyses
 
@@ -29,7 +30,7 @@ A case study that utilizes the diffindiff library is available on [arXiv](https:
 
 If you use this software, please cite:
 
-Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.4) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
+Wieland, T. (2026). diffindiff: A Python library for convenient difference-in-differences analyses (Version 2.5.6) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.18656820
 
 
 ## Installation
@@ -177,15 +178,11 @@ See the /tests directory for usage examples of most of the included functions.
 
 ## AI Usage Statement
 
-This software was developed without the use of AI-generated code. The Continue Agent in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
+This software was developed without the use of AI-generated code. The GitHub Copilot Chat in Microsoft Visual Studio Code using the GPT-5 mini model (by OpenAI) was used solely to assist in drafting and refining docstrings for documentation. The corresponding guidelines and constraints defined by the author are documented in `AGENTS-docstrings.md` in the [public GitHub repository](https://github.com/geowieland/diffindiff_official).
 
 
-## What's new (v2.5.4)
+## What's new (v2.5.6)
 
 - Bugfixes
-  - Correction of the nonsensical skipping of the treatment group dummy in the case of two observation units in didanalysis.did_analysis()
-  - Automatical drop of duplicates in results dictionaries in didanalysis_helper.extract_model_results()
-  - Inefficient successive data insert in diddata.DiffData.add_synthetic() replaced by efficient concatenating
-- Other
-  - Test script extended by another example
-  - Cleaned dependencies
+  - didtools.is_parallel(): Checking whether there is a pre-treatment period and skipping test if not (additional NOTE) 
+  - Optional installation of XGBoost and LightGBM (more stable if problems with the installation of these packages occur)

{diffindiff-2.5.4 → diffindiff-2.5.6}/diffindiff.egg-info/requires.txt

@@ -3,8 +3,10 @@ numpy
 statsmodels>=0.14.5
 scipy>=1.17
 scikit-learn
-xgboost
-lightgbm
 openpyxl
 matplotlib
 patsy
+
+[optional]
+lightgbm
+xgboost

{diffindiff-2.5.4 → diffindiff-2.5.6}/setup.py

@@ -7,7 +7,7 @@ def read_README():
     
 setup(
     name='diffindiff',
-    version='2.5.4',
+    version='2.5.6',
     description='diffindiff: Python library for convenient Difference-in-Differences analyses',
     packages=find_packages(include=["diffindiff", "diffindiff.tests"]),
     include_package_data=True,
@@ -25,11 +25,15 @@ setup(
         'statsmodels>=0.14.5',
         'scipy>=1.17',
         'scikit-learn',
-        'xgboost',
-        'lightgbm',
         'openpyxl',
         'matplotlib',
         'patsy',
     ],
+    extras_require={
+        "optional": [
+        'lightgbm',
+        'xgboost',
+        ]
+    },
     test_suite='tests',
 )