def __init__(self, config):
self.config = config
self.meta_files = dict()
self.all_asin_to_titles = dict()
for meta in self.config['metas']:
data = self.config[meta]
pp = data['pp']
self.meta_files[meta] = {}
self.meta_files[meta][
'bought'] = pp + '/' + meta + '_meta_id_bought_mapping.pt'
self.meta_files[meta][
'id'] = pp + '/' + meta + '_meta_id_item_mapping.pt'
self.meta_files[meta][
'cat'] = pp + '/' + meta + '_meta_id_cat_map.pt'
self.meta_files[meta]['idx'] = load_dict_output(
pp, 'item_id_map.json')
self.meta_files[meta]['id_item'] = load_dict_output(
pp, 'id_item_map.json')
def create_title_id_meta_mapping(name, pp_data_dir, meta_df):
records = meta_df.to_dict('record')
item_mapping = load_dict_output(pp_data_dir, 'id_item_map.json')
asin_mapping = dict()
name = name
for record in records:
asin_mapping[record['asin']] = record['title']
item_mapping.update(asin_mapping)
with open(pp_data_dir + '/' + name + '_meta_id_item_mapping.pt',
'wb') as f:
pickle.dump(item_mapping, f)
}
print("Reading dataset")
if args.dataset == "movielens":
data_dir = cfg.vals['movielens_dir'] + "/preprocessed_choice/"
elif args.dataset == "amazon":
data_dir = cfg.vals['amazon_dir'] + "/preprocessed_choice/"
else:
raise ValueError("--dataset must be 'amazon' or 'movielens'")
X_train, X_test, y_train, y_test = read_train_test_dir(data_dir)
print("Dataset read complete...")
user_item_rating_map = load_dict_output(data_dir, "user_item_rating.json",
True)
item_rating_map = load_dict_output(data_dir, "item_rating.json", True)
stats = load_dict_output(data_dir, "stats.json")
print("n users: {}".format(stats['n_users']))
print("n items: {}".format(stats['n_items']))
n_test = get_test_sample_size(X_test.shape[0], k=TEST_BATCH_SIZE)
X_test = X_test[:n_test, :]
y_test = y_test[:n_test, :]
mlp = MLP({
'num_users': stats['n_users'],
'num_items': stats['n_items'],
'latent_dim': params["h_dim_size"],
'use_cuda': args.cuda,
import torch
import pandas as pd
import numpy as np
from model.trainer import NeuralUtilityTrainer
from model._loss import loss_mse
from sklearn.preprocessing import OneHotEncoder
from experiments.utils import get_mrs_arr, get_supp_k, get_comp_k
from preprocessing.utils import load_dict_output
# Set model and data paths manuall
# TODO: Ahmad to modify for his application
MODEL_PATH = cfg.vals[
"model_dir"] + "/home_kitchen_ahmad/item_encoder_amazon_utility_done.pt"
data_dir = cfg.vals['amazon_dir'] + "/preprocessed_home_kitchen/"
item_map = load_dict_output(
cfg.vals["amazon_dir"] + "/preprocessed_home_kitchen/", "id_item_map.json",
True)
item_to_idx = load_dict_output(
cfg.vals["amazon_dir"] + "/preprocessed_home_kitchen/", "item_id_map.json",
True)
# Process data
X_train, X_test, y_train, y_test = read_train_test_dir(data_dir)
stats = load_dict_output(data_dir, "stats.json")
one_hot = OneHotEncoder(categories=[range(stats["n_items"])])
items_for_grad = one_hot.fit_transform(
np.arange(stats["n_items"]).reshape(-1, 1)).todense().astype(np.float32)
train = np.concatenate([X_train, y_train], axis=1)
test = np.concatenate([X_test, y_test], axis=1)
print("Reading dataset")
if args.dataset == "movielens":
data_dir = cfg.vals['movielens_dir'] + "/preprocessed/"
elif args.dataset == "amazon":
data_dir = cfg.vals['amazon_dir'] + "/preprocessed/"
else:
raise ValueError("--dataset must be 'amazon' or 'movielens'")
X_train, X_test, y_train, y_test = read_train_test_dir(data_dir)
stats = load_dict_output(data_dir, "stats.json")
print("Dataset read complete...")
n_test = get_test_sample_size(X_test.shape[0], k=TEST_BATCH_SIZE)
X_test = X_test[:n_test, :]
y_test = y_test[:n_test, :]
users_test = X_test[:, 0].reshape(-1,1)
items_test = X_test[:, 1].reshape(-1,1)
y_test = y_test.reshape(-1,1)
predictor = Predictor(model=model, batch_size=TEST_BATCH_SIZE, users=users_test, items=items_test, y=y_test,
use_cuda=args.cuda, n_items=stats["n_items"])
