key, value = p_str.split('=')
try:
init_args[key] = eval(value)
except:
init_args[key] = value
# convert the column argument to list
if args.columns is not None:
args.columns = args.columns.split(',')
# read the dataset
logger.info('Reading {}'.format(args.train))
train_df, item_to_idx, user_to_idx = read_dataset(args.train,
sep=',',
header=0,
columns=args.columns,
user_key=args.user_key,
item_key=args.item_key,
rating_key=args.rating_key)
logger.info('Reading {}'.format(args.test))
test_df, _, _ = read_dataset(args.test,
sep=',',
header=0,
columns=args.columns,
user_key=args.user_key,
item_key=args.item_key,
rating_key=args.rating_key,
item_to_idx=item_to_idx,
user_to_idx=user_to_idx)
# build reverse maps
try:
init_args[key] = eval(value)
except:
init_args[key] = value
# convert the column argument to list
if args.columns is not None:
args.columns = args.columns.split(',')
# read the dataset
logger.info('Reading {}'.format(args.dataset))
dataset, idx_to_user, idx_to_item = read_dataset(
args.dataset,
header=args.header,
sep=args.sep,
columns=args.columns,
make_implicit=args.make_implicit,
implicit_th=args.implicit_th,
item_key=args.item_key,
user_key=args.user_key,
rating_key=args.rating_key)
nusers, nitems = len(idx_to_user), len(idx_to_item)
logger.info('The dataset has {} users and {} items'.format(nusers, nitems))
# evaluate the recommendation quality with k-fold cross-validation
logger.info('Running {}-fold Cross Validation'.format(args.cv_folds))
roc_auc_, precision_, recall_, map_, mrr_, ndcg_ = np.zeros(
args.cv_folds), np.zeros(args.cv_folds), np.zeros(args.cv_folds), np.zeros(
args.cv_folds), np.zeros(args.cv_folds), np.zeros(args.cv_folds)
at = args.rec_length
nfold = 0
init_args = OrderedDict()
if args.params:
for p_str in args.params.split(','):
key, value = p_str.split('=')
try:
init_args[key] = eval(value)
except:
init_args[key] = value
# convert the column argument to list
if args.columns is not None:
args.columns = args.columns.split(',')
# read the dataset
logger.info('Reading {}'.format(args.train))
train_df = read_dataset(args.train, sep=',', header=0)
logger.info('Reading {}'.format(args.test))
test_df = read_dataset(args.test, sep=',', header=0)
nusers, nitems = train_df.user_idx.max()+1, train_df.item_idx.max()+1
train = df_to_csr(train_df, nrows=nusers, ncols=nitems)
test = df_to_csr(test_df, nrows=nusers, ncols=nitems)
# train the recommender
recommender = RecommenderClass(**init_args)
logger.info('Recommender: {}'.format(recommender))
tic = dt.now()
logger.info('Training started')
recommender.fit(train)
logger.info('Training completed in {}'.format(dt.now() - tic))
parser.add_argument('--item_key', type=str, default='item_id')
parser.add_argument('--rating_key', type=str, default='rating')
parser.add_argument('--rnd_seed', type=int, default=1234)
args = parser.parse_args()
# convert the column argument to list
if args.columns is not None:
args.columns = args.columns.split(',')
# read the dataset
logger.info('Reading {}'.format(args.dataset))
dataset, idx_to_user, idx_to_item = read_dataset(args.dataset,
header=args.header,
sep=args.sep,
columns=args.columns,
make_binary=args.make_binary,
binary_th=args.binary_th,
item_key=args.item_key,
user_key=args.user_key,
rating_key=args.rating_key)
nusers, nitems = len(idx_to_user), len(idx_to_item)
logger.info('The dataset has {} users and {} items'.format(nusers, nitems))
# compute the holdout split
logger.info('Computing the {:.0f}% holdout split'.format(args.holdout_perc *
100))
train_df, test_df = holdout(dataset,
user_key=args.user_key,
item_key=args.item_key,
perc=args.holdout_perc,
import logging
import argparse
import pandas as pd
import numpy as np
from recpy.recommenders.item_knn import ItemKNNRecommender
from recpy.recommenders.non_personalized import TopPop, GlobalEffects
from recpy.utils.data_utils import read_dataset, df_to_csr
from recpy.metrics import roc_auc, precision, recall, map, ndcg, rr
logger = logging.getLogger(__name__)
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s: %(name)s: %(l-evelname)s: %(message)s")
train_df = read_dataset('data/ml100k/train_tuning.csv', sep=',', header=0)
test_df = read_dataset('data/ml100k/valid.csv', sep=',', header=0)
nusers, nitems = train_df.user_idx.max() + 1, train_df.item_idx.max() + 1
train = df_to_csr(train_df, nrows=nusers, ncols=nitems)
test = df_to_csr(test_df, nrows=nusers, ncols=nitems)
shrinkage = np.arange(0, 200, 25)
result = np.zeros_like(shrinkage, dtype=np.float32)
metric = recall
at = 10
for idx, sh in enumerate(shrinkage):
logger.info("Iter {}/{}".format(idx + 1, len(shrinkage)))
recommender = ItemKNNRecommender(shrinkage=sh, similarity='cosine')
recommender.fit(train)
roc_auc_ /= n_eval
precision_ /= n_eval
recall_ /= n_eval
map_ /= n_eval
mrr_ /= n_eval
ndcg_ /= n_eval
return roc_auc_, precision_, recall_, map_, mrr_, ndcg_
metric = roc_auc
cv_folds = 5
at = 10
is_binary = True
train_df = read_dataset('../../data/ml100k/binary_holdout/train.csv',
sep=',',
header=0)
test_df = read_dataset('../../data/ml100k/binary_holdout/test.csv',
sep=',',
header=0)
nusers, nitems = train_df.user_idx.max() + 1, train_df.item_idx.max() + 1
train = df_to_csr(train_df, is_binary=is_binary, nrows=nusers, ncols=nitems)
test = df_to_csr(test_df, is_binary=is_binary, nrows=nusers, ncols=nitems)
#
# TopPop
#
# RecommenderClass = TopPop
# param_space = {}
# # Evaluate all the metrics over the hold out split
# recommender = RecommenderClass()
