Skip to content
Extraits de code Groupes Projets
Valider 9823af64 rédigé par Adrien Payen's avatar Adrien Payen
Parcourir les fichiers

update eval block

parent c53e2367
Aucune branche associée trouvée
Aucune étiquette associée trouvée
Aucune requête de fusion associée trouvée
Masquer les modifications d'espaces
En ligne Côte à côte
Affichage de
avec 104 ajouts et 26 suppressions
evaluator.ipynb
+ 104
− 26
Voir le fichier @ 9823af64
%% Cell type:markdown id:a665885b tags: %% Cell type:markdown id:a665885b tags:
# Evaluator Module # Evaluator Module
The Evaluator module creates evaluation reports. The Evaluator module creates evaluation reports.
Reports contain evaluation metrics depending on models specified in the evaluation config. Reports contain evaluation metrics depending on models specified in the evaluation config.
%% Cell type:code id:6aaf9140 tags: %% Cell type:code id:6aaf9140 tags:
``` python ``` python
# reloads modules automatically before entering the execution of code # reloads modules automatically before entering the execution of code
%load_ext autoreload %load_ext autoreload
%autoreload 2 %autoreload 2
# imports # imports
import numpy as np import numpy as np
import pandas as pd import pandas as pd
# local imports # local imports
from configs import EvalConfig from configs import EvalConfig
from constants import Constant as C from constants import Constant as C
from loaders import export_evaluation_report from loaders import export_evaluation_report
from loaders import load_ratings from loaders import load_ratings
# New imports # New imports
from surprise.model_selection import train_test_split from surprise.model_selection import train_test_split
from surprise import accuracy from surprise import accuracy
from surprise.model_selection import LeaveOneOut from surprise.model_selection import LeaveOneOut
from collections import Counter from collections import Counter
``` ```
%% Output %% Output
The autoreload extension is already loaded. To reload it, use: The autoreload extension is already loaded. To reload it, use:
%reload_ext autoreload %reload_ext autoreload
%% Cell type:markdown id:d47c24a4 tags: %% Cell type:markdown id:d47c24a4 tags:
# 1. Model validation functions # 1. Model validation functions
Validation functions are a way to perform crossvalidation on recommender system models. Validation functions are a way to perform crossvalidation on recommender system models.
%% Cell type:code id:d6d82188 tags: %% Cell type:code id:d6d82188 tags:
``` python ``` python
# -- implement the function generate_split_predictions -- # -- implement the function generate_split_predictions --
def generate_split_predictions(algo, ratings_dataset, eval_config): def generate_split_predictions(algo, ratings_dataset, eval_config):
"""Generate predictions on a random test set specified in eval_config""" """Generate predictions on a random test set specified in eval_config"""
# Spliting the data into train and test sets # Spliting the data into train and test sets
trainset, testset = train_test_split(ratings_dataset, test_size=eval_config.test_size) trainset, testset = train_test_split(ratings_dataset, test_size=eval_config.test_size)
# Training the algorithm on the train data set # Training the algorithm on the train data set
algo.fit(trainset) algo.fit(trainset)
# Predict ratings for the testset # Predict ratings for the testset
predictions = algo.test(testset) predictions = algo.test(testset)
return predictions return predictions
# -- implement the function generate_loo_top_n -- # -- implement the function generate_loo_top_n --
def generate_loo_top_n(algo, ratings_dataset, eval_config): def generate_loo_top_n(algo, ratings_dataset, eval_config):
"""Generate top-n recommendations for each user on a random Leave-one-out split (LOO)""" """Generate top-n recommendations for each user on a random Leave-one-out split (LOO)"""
# Create a LeaveOneOut split # Create a LeaveOneOut split
loo = LeaveOneOut(n_splits=1) loo = LeaveOneOut(n_splits=1)
for trainset, testset in loo.split(ratings_dataset): for trainset, testset in loo.split(ratings_dataset):
algo.fit(trainset) # Train the algorithm on the training set algo.fit(trainset) # Train the algorithm on the training set
anti_testset = trainset.build_anti_testset() # Build the anti test-set anti_testset = trainset.build_anti_testset() # Build the anti test-set
predictions = algo.test(anti_testset) # Get predictions on the anti test-set predictions = algo.test(anti_testset) # Get predictions on the anti test-set
top_n = {} top_n = {}
for uid, iid, _, est, _ in predictions: for uid, iid, _, est, _ in predictions:
if uid not in top_n: if uid not in top_n:
top_n[uid] = [] top_n[uid] = []
top_n[uid].append((iid, est)) top_n[uid].append((iid, est))
for uid, user_ratings in top_n.items(): for uid, user_ratings in top_n.items():
user_ratings.sort(key=lambda x: x[1], reverse=True) user_ratings.sort(key=lambda x: x[1], reverse=True)
top_n[uid] = user_ratings[:eval_config.top_n_value] # Get top-N recommendations top_n[uid] = user_ratings[:eval_config.top_n_value] # Get top-N recommendations
anti_testset_top_n = top_n anti_testset_top_n = top_n
return anti_testset_top_n, testset return anti_testset_top_n, testset
def generate_full_top_n(algo, ratings_dataset, eval_config): def generate_full_top_n(algo, ratings_dataset, eval_config):
"""Generate top-n recommendations for each user with full training set (LOO)""" """Generate top-n recommendations for each user with full training set (LOO)"""
full_trainset = ratings_dataset.build_full_trainset() # Build the full training set full_trainset = ratings_dataset.build_full_trainset() # Build the full training set
algo.fit(full_trainset) # Train the algorithm on the full training set algo.fit(full_trainset) # Train the algorithm on the full training set
anti_testset = full_trainset.build_anti_testset() # Build the anti test-set anti_testset = full_trainset.build_anti_testset() # Build the anti test-set
predictions = algo.test(anti_testset) # Get predictions on the anti test-set predictions = algo.test(anti_testset) # Get predictions on the anti test-set
top_n = {} top_n = {}
for uid, iid, _, est, _ in predictions: for uid, iid, _, est, _ in predictions:
if uid not in top_n: if uid not in top_n:
top_n[uid] = [] top_n[uid] = []
top_n[uid].append((iid, est)) top_n[uid].append((iid, est))
for uid, user_ratings in top_n.items(): for uid, user_ratings in top_n.items():
user_ratings.sort(key=lambda x: x[1], reverse=True) user_ratings.sort(key=lambda x: x[1], reverse=True)
top_n[uid] = user_ratings[:eval_config.top_n_value] # Get top-N recommendations top_n[uid] = user_ratings[:eval_config.top_n_value] # Get top-N recommendations
anti_testset_top_n = top_n anti_testset_top_n = top_n
return anti_testset_top_n return anti_testset_top_n
def precomputed_information(movie_data): def precomputed_information(movie_data):
""" Returns a dictionary that precomputes relevant information for evaluating in full mode """ Returns a dictionary that precomputes relevant information for evaluating in full mode
Dictionary keys: Dictionary keys:
- precomputed_dict["item_to_rank"] : contains a dictionary mapping movie ids to rankings - precomputed_dict["item_to_rank"] : contains a dictionary mapping movie ids to rankings
- (-- for your project, add other relevant information here -- ) - (-- for your project, add other relevant information here -- )
""" """
# Initialize an empty dictionary to store item_id to rank mapping # Initialize an empty dictionary to store item_id to rank mapping
item_to_rank = {} item_to_rank = {}
# Calculate popularity rank for each movie # Calculate popularity rank for each movie
ratings_count = movie_data.groupby('movieId').size().sort_values(ascending=False) ratings_count = movie_data.groupby('movieId').size().sort_values(ascending=False)
# Assign ranks to movies based on their popularity # Assign ranks to movies based on their popularity
for rank, (movie_id, _) in enumerate(ratings_count.items(), start=1): for rank, (movie_id, _) in enumerate(ratings_count.items(), start=1):
item_to_rank[movie_id] = rank item_to_rank[movie_id] = rank
# Create the precomputed dictionary # Create the precomputed dictionary
precomputed_dict = {} precomputed_dict = {}
precomputed_dict["item_to_rank"] = item_to_rank precomputed_dict["item_to_rank"] = item_to_rank
return precomputed_dict return precomputed_dict
def create_evaluation_report(eval_config, sp_ratings, precomputed_dict, available_metrics): def create_evaluation_report(eval_config, sp_ratings, precomputed_dict, available_metrics):
""" Create a DataFrame evaluating various models on metrics specified in an evaluation config. """ Create a DataFrame evaluating various models on metrics specified in an evaluation config.
""" """
evaluation_dict = {} evaluation_dict = {}
for model_name, model, arguments in eval_config.models: for model_name, model, arguments in eval_config.models:
print(f'Handling model {model_name}') print(f'Handling model {model_name}')
algo = model(**arguments) algo = model(**arguments)
evaluation_dict[model_name] = {} evaluation_dict[model_name] = {}
# Type 1 : split evaluations # Type 1 : split evaluations
if len(eval_config.split_metrics) > 0: if len(eval_config.split_metrics) > 0:
print('Training split predictions') print('Training split predictions')
predictions = generate_split_predictions(algo, sp_ratings, eval_config) predictions = generate_split_predictions(algo, sp_ratings, eval_config)
for metric in eval_config.split_metrics: for metric in eval_config.split_metrics:
print(f'- computing metric {metric}') print(f'- computing metric {metric}')
assert metric in available_metrics['split'] assert metric in available_metrics['split']
evaluation_function, parameters = available_metrics["split"][metric] evaluation_function, parameters = available_metrics["split"][metric]
evaluation_dict[model_name][metric] = evaluation_function(predictions, **parameters) evaluation_dict[model_name][metric] = evaluation_function(predictions, **parameters)
# Type 2 : loo evaluations # Type 2 : loo evaluations
if len(eval_config.loo_metrics) > 0: if len(eval_config.loo_metrics) > 0:
print('Training loo predictions') print('Training loo predictions')
anti_testset_top_n, testset = generate_loo_top_n(algo, sp_ratings, eval_config) anti_testset_top_n, testset = generate_loo_top_n(algo, sp_ratings, eval_config)
for metric in eval_config.loo_metrics: for metric in eval_config.loo_metrics:
assert metric in available_metrics['loo'] assert metric in available_metrics['loo']
evaluation_function, parameters = available_metrics["loo"][metric] evaluation_function, parameters = available_metrics["loo"][metric]
evaluation_dict[model_name][metric] = evaluation_function(anti_testset_top_n, testset, **parameters) evaluation_dict[model_name][metric] = evaluation_function(anti_testset_top_n, testset, **parameters)
# Type 3 : full evaluations # Type 3 : full evaluations
if len(eval_config.full_metrics) > 0: if len(eval_config.full_metrics) > 0:
print('Training full predictions') print('Training full predictions')
anti_testset_top_n = generate_full_top_n(algo, sp_ratings, eval_config) anti_testset_top_n = generate_full_top_n(algo, sp_ratings, eval_config)
for metric in eval_config.full_metrics: for metric in eval_config.full_metrics:
assert metric in available_metrics['full'] assert metric in available_metrics['full']
evaluation_function, parameters = available_metrics["full"][metric] evaluation_function, parameters = available_metrics["full"][metric]
evaluation_dict[model_name][metric] = evaluation_function( evaluation_dict[model_name][metric] = evaluation_function(
anti_testset_top_n, anti_testset_top_n,
**precomputed_dict, **precomputed_dict,
**parameters **parameters
) )
return pd.DataFrame.from_dict(evaluation_dict).T return pd.DataFrame.from_dict(evaluation_dict).T
``` ```
%% Cell type:markdown id:f7e83d1d tags: %% Cell type:markdown id:f7e83d1d tags:
# 2. Evaluation metrics # 2. Evaluation metrics
Implement evaluation metrics for either rating predictions (split metrics) or for top-n recommendations (loo metric, full metric) Implement evaluation metrics for either rating predictions (split metrics) or for top-n recommendations (loo metric, full metric)
%% Cell type:code id:f1849e55 tags: %% Cell type:code id:f1849e55 tags:
``` python ``` python
# -- implement the function get_hit_rate -- # -- implement the function get_hit_rate --
def get_hit_rate(anti_testset_top_n, testset): def get_hit_rate(anti_testset_top_n, testset):
"""Compute the average hit over the users (loo metric) """Compute the average hit over the users (loo metric)
A hit (1) happens when the movie in the testset has been picked by the top-n recommender A hit (1) happens when the movie in the testset has been picked by the top-n recommender
A fail (0) happens when the movie in the testset has not been picked by the top-n recommender A fail (0) happens when the movie in the testset has not been picked by the top-n recommender
""" """
hits = 0 hits = 0
total_users = len(testset) total_users = len(testset)
for uid, true_iid, _ in testset: for uid, true_iid, _ in testset:
if uid in anti_testset_top_n and true_iid in {iid for iid, _ in anti_testset_top_n[uid]}: if uid in anti_testset_top_n and true_iid in {iid for iid, _ in anti_testset_top_n[uid]}:
hits += 1 hits += 1
hit_rate = hits / total_users hit_rate = hits / total_users
return hit_rate return hit_rate
# -- implement the function get_novelty -- # -- implement the function get_novelty --
def get_novelty(anti_testset_top_n, item_to_rank): def get_novelty(anti_testset_top_n, item_to_rank):
"""Compute the average novelty of the top-n recommendation over the users (full metric) """Compute the average novelty of the top-n recommendation over the users (full metric)
The novelty is defined as the average ranking of the movies recommended The novelty is defined as the average ranking of the movies recommended
""" """
total_rank_sum = 0 total_rank_sum = 0
total_recommendations = 0 total_recommendations = 0
for uid, recommendations in anti_testset_top_n.items(): for uid, recommendations in anti_testset_top_n.items():
for iid, _ in recommendations: for iid, _ in recommendations:
if iid in item_to_rank: if iid in item_to_rank:
total_rank_sum += item_to_rank[iid] total_rank_sum += item_to_rank[iid]
total_recommendations += 1 total_recommendations += 1
if total_recommendations == 0: if total_recommendations == 0:
return 0 # Avoid division by zero return 0 # Avoid division by zero
average_rank_sum = total_rank_sum / total_recommendations average_rank_sum = total_rank_sum / total_recommendations
return average_rank_sum return average_rank_sum
``` ```
%% Cell type:markdown id:1a9855b3 tags: %% Cell type:markdown id:1a9855b3 tags:
# 3. Evaluation workflow # 3. Evaluation workflow
Load data, evaluate models and save the experimental outcomes Load data, evaluate models and save the experimental outcomes
%% Cell type:code id:704f4d2a tags: %% Cell type:code id:704f4d2a tags:
``` python ``` python
AVAILABLE_METRICS = { AVAILABLE_METRICS = {
"split": { "split": {
"mae": (accuracy.mae, {'verbose': False}), "mae": (accuracy.mae, {'verbose': False}),
"rmse": (accuracy.rmse, {'verbose': False}) "rmse": (accuracy.rmse, {'verbose': False})
}, },
"loo": { "loo": {
"hit_rate": (get_hit_rate, {}), "hit_rate": (get_hit_rate, {}),
}, },
"full": { "full": {
"novelty": (get_novelty, {}), "novelty": (get_novelty, {}),
} }
} }
sp_ratings = load_ratings(surprise_format=True) sp_ratings = load_ratings(surprise_format=True)
precomputed_dict = precomputed_information(pd.read_csv("data/tiny/evidence/ratings.csv")) precomputed_dict = precomputed_information(pd.read_csv("data/tiny/evidence/ratings.csv"))
evaluation_report = create_evaluation_report(EvalConfig, sp_ratings, precomputed_dict, AVAILABLE_METRICS) evaluation_report = create_evaluation_report(EvalConfig, sp_ratings, precomputed_dict, AVAILABLE_METRICS)
export_evaluation_report(evaluation_report) export_evaluation_report(evaluation_report)
``` ```
%% Output %% Output
Handling model baseline_1 Handling model baseline_1
Training split predictions Training split predictions
- computing metric mae - computing metric mae
- computing metric rmse - computing metric rmse
Training loo predictions Training loo predictions
Training full predictions Training full predictions
Handling model baseline_2 Handling model baseline_2
Training split predictions Training split predictions
- computing metric mae - computing metric mae
- computing metric rmse - computing metric rmse
Training loo predictions Training loo predictions
Training full predictions Training full predictions
Handling model baseline_3 Handling model baseline_3
Training split predictions Training split predictions
- computing metric mae - computing metric mae
- computing metric rmse - computing metric rmse
Training loo predictions Training loo predictions
Training full predictions Training full predictions
Handling model baseline_4 Handling model baseline_4
Training split predictions Training split predictions
- computing metric mae - computing metric mae
- computing metric rmse - computing metric rmse
Training loo predictions Training loo predictions
Training full predictions Training full predictions
Handling model ContentBased_sample Handling model ContentBased_sample
Training split predictions Training split predictions
- computing metric mae - computing metric mae
- computing metric rmse - computing metric rmse
Training loo predictions Training loo predictions
Training full predictions Training full predictions
Handling model ContentBased_score Handling model ContentBased_score
Training split predictions Training split predictions
- computing metric mae - computing metric mae
- computing metric rmse - computing metric rmse
Training loo predictions Training loo predictions
Training full predictions Training full predictions
Handling model ContentBased_Lr Handling model ContentBased_Lr
Training split predictions Training split predictions
- computing metric mae
- computing metric rmse
Training loo predictions
Training full predictions
The data has been exported to the evaluation report
--------------------------------------------------------------------------- mae rmse hit_rate novelty
KeyError Traceback (most recent call last) baseline_1 1.561178 1.792482 0.074766 99.405607
File /Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/surprise/trainset.py:180, in Trainset.to_raw_iid(self, iiid) baseline_2 1.471412 1.819364 0.000000 429.942991
179 try: baseline_3 0.878270 1.085591 0.074766 99.405607
--> 180 return self._inner2raw_id_items[iiid] baseline_4 0.705673 0.912313 0.130841 60.202804
181 except KeyError: ContentBased_sample 1.013747 1.350417 0.084112 178.048598
KeyError: 1210 ContentBased_score 1.461846 1.803067 0.018692 437.222430
ContentBased_Lr 1.202626 1.460273 0.084112 278.046729
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
Cell In[64], line 16
14 sp_ratings = load_ratings(surprise_format=True)
15 precomputed_dict = precomputed_information(pd.read_csv("data/tiny/evidence/ratings.csv"))
---> 16 evaluation_report = create_evaluation_report(EvalConfig, sp_ratings, precomputed_dict, AVAILABLE_METRICS)
17 export_evaluation_report(evaluation_report)
Cell In[62], line 95, in create_evaluation_report(eval_config, sp_ratings, precomputed_dict, available_metrics)
93 if len(eval_config.split_metrics) > 0:
94 print('Training split predictions')
---> 95 predictions = generate_split_predictions(algo, sp_ratings, eval_config)
96 for metric in eval_config.split_metrics:
97 print(f'- computing metric {metric}')
Cell In[62], line 9, in generate_split_predictions(algo, ratings_dataset, eval_config)
6 trainset, testset = train_test_split(ratings_dataset, test_size=eval_config.test_size)
8 # Training the algorithm on the train data set
----> 9 algo.fit(trainset)
11 # Predict ratings for the testset
12 predictions = algo.test(testset)
File ~/vscodeworkspace/recomsys/models.py:130, in ContentBased.fit(self, trainset)
126 user_ratings = [(self.trainset.to_raw_iid(i), rating) for (i, rating) in self.trainset.ur[u]]
128 df_user = pd.DataFrame(user_ratings, columns = ["item_id", "user_ratings"])
--> 130 df_user["item_id"] = df_user["item_id"].map(trainset.to_raw_iid)
132 df_user = df_user.merge(self.content_features, left_on = "item_id", right_index = True, how = 'left')
134 X = df_user['n_character_title'].values.reshape(-1,1)
File /Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/pandas/core/series.py:4544, in Series.map(self, arg, na_action)
4464 def map(
4465 self,
4466 arg: Callable | Mapping | Series,
4467 na_action: Literal["ignore"] | None = None,
4468 ) -> Series:
4469 """
4470 Map values of Series according to an input mapping or function.
4471
(...)
4542 dtype: object
4543 """
-> 4544 new_values = self._map_values(arg, na_action=na_action)
4545 return self._constructor(new_values, index=self.index, copy=False).__finalize__(
4546 self, method="map"
4547 )
File /Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/pandas/core/base.py:921, in IndexOpsMixin._map_values(self, mapper, na_action, convert)
918 if isinstance(arr, ExtensionArray):
919 return arr.map(mapper, na_action=na_action)
--> 921 return algorithms.map_array(arr, mapper, na_action=na_action, convert=convert)
File /Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/pandas/core/algorithms.py:1814, in map_array(arr, mapper, na_action, convert)
1812 values = arr.astype(object, copy=False)
1813 if na_action is None:
-> 1814 return lib.map_infer(values, mapper, convert=convert)
1815 else:
1816 return lib.map_infer_mask(
1817 values, mapper, mask=isna(values).view(np.uint8), convert=convert
1818 )
File lib.pyx:2926, in pandas._libs.lib.map_infer()
File /Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/surprise/trainset.py:182, in Trainset.to_raw_iid(self, iiid)
180 return self._inner2raw_id_items[iiid]
181 except KeyError:
--> 182 raise ValueError(str(iiid) + " is not a valid inner id.")
ValueError: 1210 is not a valid inner id.
%% Cell type:markdown id:6f8b6d19 tags: %% Cell type:markdown id:6f8b6d19 tags:
dire quel modèle est meilleur ? dire quel modèle est meilleur ?
......
0% ou .
You are about to add 0 people to the discussion. Proceed with caution.
Terminez d'abord l'édition de ce message.
Veuillez vous inscrire ou vous pour commenter