ET_623_Project / utils.py
vivek9's picture
Update utils.py
ebdbfef verified
import pandas as pd
import numpy as np
from scipy import stats
import csv
def process_csv(input_file, output_file="output.csv",user_id_column="user_id",timestamp_column="timestamp",action_column="actions"):
with open(input_file, 'r', newline='') as csvfile, open(output_file, 'w', newline='') as new_csvfile:
reader = csv.reader(csvfile)
writer = csv.writer(new_csvfile)
# Get the column indices for user_id, timestamp, and action
user_id_index, timestamp_index, action_index = None, None, None
for i, row in enumerate(reader):
if i == 0: # Header row
user_id_index = row.index(user_id_column)
timestamp_index = row.index(timestamp_column)
action_index = row.index(action_column)
break
# Read the rest of the data and store it in a dictionary
user_actions = {}
for row in reader:
user_id = row[user_id_index]
timestamp = row[timestamp_index]
action = row[action_index]
if user_id not in user_actions:
user_actions[user_id] = []
# Sort actions by timestamp in ascending order
user_actions[user_id].append((timestamp, action))
# Write the data to the output CSV file with the new column
writer.writerow([user_id_column, 'actions']) # Write header with new column
for user_id, actions in user_actions.items():
sorted_actions = actions # Sort by timestamp
concatenated_actions = ';'.join([action for timestamp, action in sorted_actions])
writer.writerow([user_id, concatenated_actions])
def generate_sequence_list(sentence, min_gap,max_gap, sliding_window_min=1,sliding_window_max=1):
# Split the sentence into words
words = sentence.split(";")
# Generate n-grams
ngrams = []
for sliding_window in range(sliding_window_min,sliding_window_max+1):
for gram_length in range(min_gap,max_gap + 1): # Loop from 1 to n
for i in range(0, len(words) - gram_length + 1, sliding_window):
ngram = '--->'.join(words[i:i+gram_length])
ngrams.append(ngram)
return ngrams
def create_dict_from_df(df, identifier_column, sequence,min_gap,max_gap,sliding_window_min=1,sliding_window_max=1):
result_dict = {}
unique_values_set = set() # Initialize set to store unique values
for index, row in df.iterrows():
key = row[identifier_column]
values = generate_sequence_list(row[sequence],min_gap,max_gap,sliding_window_min,sliding_window_max)
result_dict[key] = values
unique_values_set.update(values) # Update the set with unique values
return result_dict, unique_values_set
def create_dataframe_from_dict_and_set(result_dict, unique_values_set):
# Initialize an empty dictionary to store counts
counts_dict = {}
# Iterate over the set
for value in unique_values_set:
counts_dict[value] = {}
# Iterate over the keys in the result_dict
for key, values in result_dict.items():
counts_dict[value][key] = values.count(value)
# Create a DataFrame from the counts dictionary
df = pd.DataFrame(counts_dict).fillna(0)
# Transpose the DataFrame so that keys become columns and values become rows
df = df.transpose()
return df
def process_dataframe(df):
# Calculate num_student
num_student = len(df.columns)
# Calculate I-Frequency and S-Frequency
I_Frequency = df.sum(axis=1)
S_Frequency = (df > 0).sum(axis=1)
# Create a dictionary for new data
new_data = {
'I-Frequency': I_Frequency,
'S-Frequency': S_Frequency
}
# Create a DataFrame from the new data
new_df = pd.DataFrame(new_data)
# Calculate I-Support by dividing I-Frequency with num_student
new_df['I-Support (mean)'] = new_df['I-Frequency'] / num_student
new_df['S-Support'] = new_df['S-Frequency'] / num_student
# Calculate standard deviation of each row
new_df['I-Support (sd)'] = df.std(axis=1,skipna=True)
return new_df
def calculate_p_value(test_type, vector_a, vector_b=None, **kwargs):
"""
Calculate the p-value for different types of t-tests.
Parameters:
test_type (str): Type of test to perform.
vector_a (array-like): Data for sample A.
vector_b (array-like, optional): Data for sample B (only required for some tests).
**kwargs: Additional keyword arguments required for specific tests.
Returns:
p_value (float): The p-value obtained from the test.
"""
if test_type == 'poisson_means_test':
# Poisson means test
result = stats.poisson_means_test(vector_a, vector_b, **kwargs)
elif test_type == 'ttest_ind':
# T-test for the means of two independent samples
result = stats.ttest_ind(vector_a, vector_b, **kwargs)
elif test_type == 'mannwhitneyu':
# Mann-Whitney U rank test on two independent samples
result = stats.mannwhitneyu(vector_a, vector_b, **kwargs)
elif test_type == 'bws_test':
# Baumgartner-Weiss-Schindler test on two independent samples
result = stats.bws_test(vector_a, vector_b, **kwargs)
elif test_type == 'ranksums':
# Wilcoxon rank-sum statistic for two samples
result = stats.ranksums(vector_a, vector_b, **kwargs)
elif test_type == 'brunnermunzel':
# Brunner-Munzel test on samples
result = stats.brunnermunzel(vector_a, vector_b, **kwargs)
elif test_type == 'mood':
# Mood's test for equal scale parameters
result = stats.mood(vector_a, vector_b, **kwargs)
elif test_type == 'ansari':
# Ansari-Bradley test for equal scale parameters
result = stats.ansari(vector_a, vector_b, **kwargs)
elif test_type == 'cramervonmises_2samp':
# Two-sample Cram�r-von Mises test for goodness of fit
result = stats.cramervonmises_2samp(vector_a, vector_b, **kwargs)
elif test_type == 'epps_singleton_2samp':
# Epps-Singleton (ES) test statistic
result = stats.epps_singleton_2samp(vector_a, vector_b, **kwargs)
elif test_type == 'ks_2samp':
# Two-sample Kolmogorov-Smirnov test for goodness of fit
result = stats.ks_2samp(vector_a, vector_b, **kwargs)
elif test_type == 'kstest':
# One-sample or two-sample Kolmogorov-Smirnov test for goodness of fit
result = stats.kstest(vector_a, vector_b, **kwargs)
else:
raise ValueError("Invalid test type.")
# Get the p-value
p_value = result.pvalue
return p_value
def SPM_(path_to_csv,dataset_format, identifier_column, sequence_column,sortby="S-Support",min_gap=1,max_gap=1,sliding_window_min=1,sliding_window_max=1,S_support_thresh=0,I_support_thresh=0,timestamp_column="timestamp"):
if dataset_format==1:
process_csv(path_to_csv, output_file="output.csv",user_id_column=identifier_column,timestamp_column=timestamp_column,action_column=sequence_column)
path_to_csv="output.csv"
# Read CSV file
data = pd.read_csv(path_to_csv)
# Create dictionary from DataFrame
data_seq, corpus = create_dict_from_df(data, identifier_column, sequence_column, min_gap,max_gap,sliding_window_min,sliding_window_max)
# Create occurrence matrix
occurence_matrix = create_dataframe_from_dict_and_set(data_seq, corpus)
# Process occurrence matrix
spm_result = process_dataframe(occurence_matrix)
spm_result = spm_result.sort_values(by=sortby, ascending=False)
return spm_result[(spm_result['S-Support'] > S_support_thresh) & (spm_result['I-Support (mean)'] > I_support_thresh)], occurence_matrix
def SPM(config):
path_to_csv = config.get('path_to_csv')
dataset_format = config.get('dataset_format')
identifier_column = config.get('identifier_column')
sequence_column = config.get('sequence_column')
sortby = config.get('sortby', "S-Support")
min_gap = config.get('min_gap', 1)
max_gap = config.get('max_gap', 1)
sliding_window_min = config.get('sliding_window_min', 1)
sliding_window_max = config.get('sliding_window_max', 1)
S_support_thresh = config.get('S_support_thresh', 0)
I_support_thresh = config.get('I_support_thresh', 0)
timestamp_column = config.get('timestamp_column', "timestamp")
return SPM_(path_to_csv,dataset_format, identifier_column, sequence_column,sortby,min_gap,max_gap,sliding_window_min,sliding_window_max,S_support_thresh,I_support_thresh,timestamp_column)
def DSM(config):
path_to_csv_left = config['path_to_csv_left']
dataset_format = config['dataset_format']
path_to_csv_right = config['path_to_csv_right']
identifier_column = config['identifier_column']
sequence_column = config['sequence_column']
sortby = config['sortby']
min_gap = config['min_gap']
max_gap = config['max_gap']
sliding_window_min = config['sliding_window_min']
sliding_window_max = config['sliding_window_max']
S_support_thresh = config['S_support_thresh']
I_support_thresh = config['I_support_thresh']
threshold_pvalue = config['threshold_pvalue']
test_type = config['test_type']
timestamp_column = config.get('timestamp_column', 'timestamp')
if dataset_format == 1:
process_csv(path_to_csv_left, output_file="output_left.csv", user_id_column=identifier_column,
timestamp_column=timestamp_column, action_column=sequence_column)
path_to_csv_left = "output_left.csv"
process_csv(path_to_csv_right, output_file="output_right.csv", user_id_column=identifier_column,
timestamp_column=timestamp_column, action_column=sequence_column)
path_to_csv_left = "output_right.csv"
ptrn_left = []
ptrn_right = []
ptrn_both_left = []
ptrn_both_right = []
spm_result_left, occurence_matrix_left = SPM_(path_to_csv_left, 0, identifier_column, sequence_column, sortby,
min_gap, max_gap, sliding_window_min, sliding_window_max,
S_support_thresh, I_support_thresh)
spm_result_right, occurence_matrix_right = SPM_(path_to_csv_right, 0, identifier_column, sequence_column, sortby,
min_gap, max_gap, sliding_window_min, sliding_window_max,
S_support_thresh, I_support_thresh)
result_data = []
all_ptrn = set(spm_result_left.index)
all_ptrn.update(spm_result_right.index)
left_ptrn_data = set(spm_result_left.index)
right_ptrn_data = set(spm_result_right.index)
for ptrn in all_ptrn:
isupport_left = occurence_matrix_left.loc[ptrn, :].values if ptrn in spm_result_left.index else np.zeros(
occurence_matrix_left.shape[1])
isupport_right = occurence_matrix_right.loc[ptrn, :].values if ptrn in spm_result_right.index else np.zeros(
occurence_matrix_right.shape[1])
p_value = calculate_p_value(test_type, isupport_left, isupport_right)
if p_value < threshold_pvalue:
if (ptrn in left_ptrn_data) and (ptrn in right_ptrn_data):
if isupport_left.mean() > isupport_right.mean():
ptrn_both_left.append(ptrn)
result_data.append((ptrn, p_value, isupport_left.mean(), isupport_right.mean(), "both_left"))
else:
ptrn_both_right.append(ptrn)
result_data.append((ptrn, p_value, isupport_left.mean(), isupport_right.mean(), "both_right"))
else:
if ptrn in left_ptrn_data:
ptrn_left.append(ptrn)
result_data.append((ptrn, p_value, isupport_left.mean(), np.nan, "left"))
else:
ptrn_right.append(ptrn)
result_data.append((ptrn, p_value, np.nan, isupport_right.mean(), "right"))
result_df = pd.DataFrame(result_data,
columns=['ptrn', 'ttest_value', 'isupportleft_mean', 'isupportright_mean', "Group"])
return ptrn_left, ptrn_right, ptrn_both_left, ptrn_both_right, result_df