Mercurial > repos > bgruening > sklearn_feature_selection
view association_rules.py @ 32:a7c667ff83fe draft
"planemo upload for repository https://github.com/bgruening/galaxytools/tree/master/tools/sklearn commit d970dae980fbe349414fc0889f719d875d999c5b"
author | bgruening |
---|---|
date | Fri, 27 Aug 2021 10:19:00 +0000 |
parents | 5773e98921fc |
children | 61edd9e5c17f |
line wrap: on
line source
import argparse import json import warnings import pandas as pd from mlxtend.frequent_patterns import association_rules, fpgrowth from mlxtend.preprocessing import TransactionEncoder def main(inputs, infile, outfile, min_support=0.5, min_confidence=0.5, min_lift=1.0, min_conviction=1.0, max_length=None): """ Parameter --------- input : str File path to galaxy tool parameter infile : str File paths of input vector outfile : str File path to output matrix min_support: float Minimum support min_confidence: float Minimum confidence min_lift: float Minimum lift min_conviction: float Minimum conviction max_length: int Maximum length """ warnings.simplefilter('ignore') with open(inputs, 'r') as param_handler: params = json.load(param_handler) input_header = params['header0'] header = 'infer' if input_header else None with open(infile) as fp: lines = fp.read().splitlines() if header is not None: lines = lines[1:] dataset = [] for line in lines: line_items = line.split("\t") dataset.append(line_items) # TransactionEncoder learns the unique labels in the dataset and transforms the # input dataset (a Python list of lists) into a one-hot encoded NumPy boolean array te = TransactionEncoder() te_ary = te.fit_transform(dataset) # Turn the encoded NumPy array into a DataFrame df = pd.DataFrame(te_ary, columns=te.columns_) # Extract frequent itemsets for association rule mining # use_colnames: Use DataFrames' column names in the returned DataFrame instead of column indices frequent_itemsets = fpgrowth(df, min_support=min_support, use_colnames=True, max_len=max_length) # Get association rules, with confidence larger than min_confidence rules = association_rules(frequent_itemsets, metric="confidence", min_threshold=min_confidence) # Filter association rules, keeping rules with lift and conviction larger than min_liftand and min_conviction rules = rules[(rules['lift'] >= min_lift) & (rules['conviction'] >= min_conviction)] # Convert columns from frozenset to list (more readable) rules['antecedents'] = rules['antecedents'].apply(list) rules['consequents'] = rules['consequents'].apply(list) # The next 3 steps are intended to fix the order of the association # rules generated, so tests that rely on diff'ing a desired output # with an expected output can pass # 1) Sort entry in every row/column for columns 'antecedents' and 'consequents' rules['antecedents'] = rules['antecedents'].apply(lambda row: sorted(row)) rules['consequents'] = rules['consequents'].apply(lambda row: sorted(row)) # 2) Create two temporary string columns to sort on rules['ant_str'] = rules['antecedents'].apply(lambda row: " ".join(row)) rules['con_str'] = rules['consequents'].apply(lambda row: " ".join(row)) # 3) Sort results so they are re-producable rules.sort_values(by=['ant_str', 'con_str'], inplace=True) del rules['ant_str'] del rules['con_str'] rules.reset_index(drop=True, inplace=True) # Write association rules and metrics to file rules.to_csv(outfile, sep="\t", index=False) if __name__ == '__main__': aparser = argparse.ArgumentParser() aparser.add_argument("-i", "--inputs", dest="inputs", required=True) aparser.add_argument("-y", "--infile", dest="infile", required=True) aparser.add_argument("-o", "--outfile", dest="outfile", required=True) aparser.add_argument("-s", "--support", dest="support", default=0.5) aparser.add_argument("-c", "--confidence", dest="confidence", default=0.5) aparser.add_argument("-l", "--lift", dest="lift", default=1.0) aparser.add_argument("-v", "--conviction", dest="conviction", default=1.0) aparser.add_argument("-t", "--length", dest="length", default=5) args = aparser.parse_args() main(args.inputs, args.infile, args.outfile, min_support=float(args.support), min_confidence=float(args.confidence), min_lift=float(args.lift), min_conviction=float(args.conviction), max_length=int(args.length))