def get_results(dataset_name, coldstart, cs_type='none', n_entries=0):
# Get dataset
dataset = get_dataset(dataset_name)
models = []
# Hybrid Model
from hybrid_model.hybrid import HybridModel
from hybrid_model.config import hybrid_config
model_type = HybridModel
config = hybrid_config
models.append(EvalModel(model_type.__name__, model_type, config))
# Bias Baseline
from hybrid_model.models import BiasEstimator
model_type = BiasEstimator
config = {}
models.append(EvalModel(model_type.__name__, model_type, config))
# SVD
from hybrid_model.models import SVD
model_type = SVD
config = {}
models.append(EvalModel(model_type.__name__, model_type, config))
results = evaluate_models_xval(dataset,
models,
coldstart=coldstart,
cs_type=cs_type,
n_entries=n_entries)
return results
def analyze(ds_name):
ds = dataset.get_dataset(ds_name)
(inds_u, inds_i, y, users_features, items_features) = ds.data
(users_desc, items_desc, users_features_desc,
items_features_desc) = dataset.get_dataset_desc(ds_name)
n_users = ds.n_users
n_items = ds.n_items
n_users_features = len(users_features_desc)
n_items_features = len(items_features_desc)
# Sanity checks
assert (n_users, n_users_features) == users_features.shape
assert (n_items, n_items_features) == items_features.shape
matrix = np.zeros((n_users, n_items))
for u, i, r in zip(inds_u, inds_i, y):
matrix[u, i] = r
entries = len(matrix.nonzero()[0])
sparsity = float(entries)
sparsity /= (matrix.shape[0] * matrix.shape[1])
print('Number of users {}'.format(n_users))
print('Number of Items {}'.format(n_items))
print('Total valid entries {}'.format(entries))
print('Sparsity {:4.4f}%'.format(sparsity * 100))
items_per_user_avg = np.mean(np.sum((matrix != 0).astype(np.int), 1))
users_per_item_avg = np.mean(np.sum((matrix != 0).astype(np.int), 0))
print('Average number of items per user {}'.format(items_per_user_avg))
print('Average number of users per item {}'.format(users_per_item_avg))
users_witout_items = np.sum(np.sum((matrix != 0), 1) == 0)
print('Users without any items {}'.format(users_witout_items))
items_without_users = np.sum(np.sum((matrix != 0), 0) == 0)
print('Items without any users {}'.format(items_without_users))
# # Sparsify
# # Delete users and items without valid matrix
# sales_sparse = matrix[~np.all(matrix == 0, 1), :]
# sales_sparse = sales_sparse[:, ~np.all(matrix == 0, 0)]
#
# entries_sparse = len(sales_sparse.nonzero()[0])
# sparsity_sparse = float(entries)
# sparsity_sparse /= (sales_sparse.shape[0] * sales_sparse.shape[1])
# print('Removing users and items without matrix:')
# print('Number of users {}'.format(sales_sparse.shape[0]))
# print('Number of Items {}'.format(sales_sparse.shape[1]))
# print('Sparsity after removal of users without matrix {:4.4f}%'.format(sparsity_sparse * 100))
# items_per_user_avg = np.mean(np.sum((sales_sparse != 0), 1))
# users_per_item_avg = np.mean(np.sum((sales_sparse != 0), 0))
# print('Average number of items per user {}'.format(items_per_user_avg))
# print('Average number of users per item {}'.format(users_per_item_avg))
ga = np.mean(y)
y_shift = y - ga
print('Global Average: {:4.4f}'.format(ga))
user_stats = pd.DataFrame([], index=users_features_desc)
user_stats['# users'] = np.sum(users_features, 0)
user_stats['# interactions'] = np.sum(users_features[inds_u, :], 0)
user_stats['avg rating'] = users_features[inds_u, :].T @ y_shift / np.sum(
users_features[inds_u, :], 0)
print(user_stats)
item_stats = pd.DataFrame([], index=items_features_desc)
item_stats['# items'] = np.sum(items_features, 0)
item_stats['# interactions'] = np.sum(items_features[inds_i, :], 0)
item_stats['avg rating'] = items_features[inds_i, :].T @ y_shift / np.sum(
items_features[inds_i, :], 0)
print(item_stats)
user_feature_stats = pd.DataFrame(
(users_features.T @ users_features)[7:, :9],
index=users_features_desc[7:],
columns=users_features_desc[:9])
print(user_feature_stats)
item_feature_stats = pd.DataFrame((items_features.T @ items_features),
index=items_features_desc,
columns=items_features_desc)
print(item_feature_stats)
concat_features_desc = users_features_desc + items_features_desc
concat_features_interactions = np.concatenate(
(users_features[inds_u, :], items_features[inds_i, :]), 1)
feature_interactions = pd.DataFrame(
concat_features_interactions.T @ concat_features_interactions,
index=concat_features_desc,
columns=concat_features_desc)
print(feature_interactions)
feature_corr = pd.DataFrame(
(concat_features_interactions.T * y_shift)
@ concat_features_interactions / feature_interactions.values,
index=concat_features_desc,
columns=concat_features_desc)
print(feature_corr)
import script_chdir
import numpy as np
import results.plots as lplot
import matplotlib.pyplot as plt
from hybrid_model.dataset import get_dataset
from hybrid_model.index_sampler import IndexSamplerUserItembased as IndexSampler
dataset = get_dataset('ml100k')
(inds_u, inds_i, y, users_features, items_features) = dataset.data
# mat = np.zeros((dataset.n_users, dataset.n_items), np.float)
#
# for u, i in zip(inds_u, inds_i):
# mat[u, i] = 1.0
#
# # Get user/item distributions and order
# dist_users = np.sum(mat, axis=1).astype(np.int)
# dist_items = np.sum(mat, axis=0).astype(np.int)
user_dist = np.bincount(inds_u, minlength=dataset.n_users)
item_dist = np.bincount(inds_i, minlength=dataset.n_items)
order_users = np.argsort(-user_dist)
order_items = np.argsort(-item_dist)
dist_users = user_dist[order_users]
dist_items = item_dist[order_items]
inds_u = np.argsort(order_users)[inds_u]
inds_i = np.argsort(order_items)[inds_i]