--- a
+++ b/src/utils.py
@@ -0,0 +1,134 @@
+import seaborn
+import pandas as pd
+import numpy as np
+import random
+import string
+from pathlib import Path
+# pd.set_option('future.no_silent_downcasting', True) # for future compatibility of .replace method
+
+def inter_len(set1, set2, dfs=False):
+    """
+    Compute the intersection of two sets and return ratio.
+    """
+    if dfs:     
+        n = len(set1.index.intersection(set2.index))
+        p = float(n)/max(len(set1.index), len(set2.index))
+    else:
+        n = len(set1.intersection(set2))
+        p = float(n)/max(len(set1), len(set2))
+    return (n, p)
+
+def plot_pairwise_intersection_ratio(dfs, tech_names):
+    """
+    Creates heatmap with pairwise intersection ratios, i.e., how many indices of different 
+    dataframes are the same. 
+    """
+    n = len(dfs)
+    overlaps = np.zeros((n,n))
+    for i in range(n):
+        for j in range(n):
+            overlaps[i][j] = inter_len(dfs[i].index, dfs[j].index)[1]
+
+    seaborn.heatmap(overlaps, cmap='hot', xticklabels = tech_names+["ids"], yticklabels = tech_names+["ids"])
+
+def join_selected_dfs(dfs, df_names, df_names_to_join):
+    """ 
+    Join list of dfs.
+    """
+    dfs_ids = [df_names.index(df_name) for df_name in df_names_to_join]
+    result = pd.DataFrame()
+
+    for i in dfs_ids:
+        if result.empty:
+            result = dfs[i].copy()
+        else:
+            result = result.join(dfs[i], how="inner")
+
+    return result
+
+def add_prefix_to_cols(df, prefix):
+    """ 
+    Adds prefix to colnames.
+    """
+    df.columns = [prefix+col for col in df.columns]
+
+def generate_sample_id(characters, length=10):
+    """ 
+    Generates random sample id. Same style of id as used in medical datasets. 
+    """
+    rand_string = ''.join(random.choices(characters, k=length))
+
+    return rand_string[:length//2]+'-'+rand_string[length//2:]
+
+def has_duplicates(seq):
+    return len(seq) != len(set(seq))
+
+def gen_unique_ids(n):
+    """ 
+    Generate unique list of random ids for a medical dataframe.
+    """
+    length = 10
+    characters = string.ascii_lowercase
+         
+    ids = [generate_sample_id(characters, length) for i in range(n)]
+    assert(not has_duplicates(ids))
+    return ids
+
+def prepare_and_save_data(df: pd.DataFrame, 
+                          directory_path_: str, 
+                          file_name: str, 
+                          treatment_col: str ='tre_nivolumab', 
+                          directory_name: str ="", 
+                          counterfactual_outcomes = None):
+    """ 
+    Brings data into the format required for the Tumor board and creates directories where necessary.
+    """
+    # Get colnames
+    outcome_cols = [col for col in df.columns if col.startswith("out")]
+
+    # Create directories
+    prepared_path_ = directory_path_ + file_name[:-4] + '/' + directory_name + '/' + treatment_col + '/'
+
+    #Return to this if errors occur
+    #prepared_path_ = directory_path_ + file_name[:-4] + directory_name + '/' + treatment_col + '/'
+
+    outcomes_path_ = prepared_path_+'outcomes/'
+    treatment_path_ = prepared_path_+'treatment/'
+    features_path_ = prepared_path_+'features/'
+
+    # Create directory and subdirectories 
+    Path(prepared_path_).mkdir(parents=True, exist_ok=True)
+    Path(outcomes_path_).mkdir(parents=True, exist_ok=True)
+    Path(treatment_path_).mkdir(parents=True, exist_ok=True)
+    Path(features_path_).mkdir(parents=True, exist_ok=True)
+    
+    # Create treatment file
+    df_treatment_factual = pd.get_dummies(df[treatment_col]).replace({False:0, True:1})
+    df_treatment_factual.columns = ['a'+str(i)+'_'+str(name) for i,name in enumerate(df_treatment_factual.columns)]
+    df_treatment_factual.to_csv(treatment_path_ + 'treatment_factual.csv', index_label='ID')
+
+    if counterfactual_outcomes is not None:
+        df_treatment_data = df_treatment_factual.replace({0:1})
+        df_treatment_data.to_csv(treatment_path_ + 'treatment_data.csv', index_label='ID')
+        counterfactual_outcomes.to_csv(outcomes_path_ + 'outcomes_data.csv', index_label='ID')
+
+    # Create outcome file
+    df_outcomes_factual = df[outcome_cols]
+    df_outcomes_factual.columns = ['y'+str(i)+'_'+str(name) for i,name in enumerate(df_outcomes_factual.columns)]
+    df_outcomes_factual.to_csv(outcomes_path_ + 'outcomes_factual.csv', index_label='ID')
+
+    # Find out how many technologies are used
+    feature_names = df.columns[df.columns.str.match(r'^\d{2}_')]
+    tech_numbers = set([feature[:2] for feature in feature_names])
+
+    # Create features files
+    df_features = pd.DataFrame()
+    for tech in tech_numbers:
+        df_feature_factual = df[df.columns[df.columns.str.startswith(tech)]]
+        df_feature_factual.columns = ['p'+str(int(tech))+'_'+str(i)+'_'+tech_name[3:] for i, tech_name in enumerate(df_feature_factual.columns)]
+        df_features = pd.concat([df_features, df_feature_factual], axis=1)
+        df_feature_factual.to_csv(features_path_ + 'patient_data_'+str(int(tech))+'.csv', index_label='ID')
+    df_features.to_csv(features_path_ + 'patient_data_all.csv', index_label='ID')
+
+    # Return prepared path
+    return prepared_path_
\ No newline at end of file