[Expand my own dataset?]

[ZuoxiYang]

[Expand my own dataset?]

[saghiles]

Hi, you can use Cornac's Reader class to load your data. For instance, to read preference data in the triplet format (User_id, Item_id, Rating) one can proceed as follows:

from cornac.data import Reader
reader = Reader()
data = reader.read(fpath, fmt='UIR', sep='\t')

More details in the documentation.

[ZuoxiYang]

Hi @ZuoxiYang,
Can you show all your code here so that we can reproduce the issue?

I guess you might need to change rating_threshold=1.0.
That’s what Cornac uses to identify feedback values as positive or negative ones.

Thanks for for your nice reply, it has worked well now. Thanks a lot.

[ZuoxiYang]

Hi, you can use Cornac's Reader class to load your data. For instance, to read preference data in the triplet format (User_id, Item_id, Rating) one can proceed as follows:

from cornac.data import Reader
reader = Reader()
data = reader.read(fpath, fmt='UIR', sep='\t')

More details in the documentation.

Thanks for your reply. And does the cornac only can work on the explicit data? When I make it play on the implicit data, it would console that:

Traceback (most recent call last):
File "baseline-cornac.py", line 29, in
exp.run()
File "/python3.6/site-packages/cornac/experiment/experiment.py", line 89, in run
user_based=self.user_based)
File "/python3.6/site-packages/cornac/eval_methods/base_method.py", line 477, in evaluate
verbose=self.verbose)
File "***/python3.6/site-packages/cornac/eval_methods/base_method.py", line 169, in ranking_eval
avg_results.append(sum(user_results[i].values()) / len(user_results[i]))
ZeroDivisionError: division by zero

How can it work well on the implicit data(the ratings only 1 and 0)

[tqtg]

Hi @ZuoxiYang,
Can you show all your code here so that we can reproduce the issue?

[ZuoxiYang]

Hi @ZuoxiYang,
Can you show all your code here so that we can reproduce the issue?

import cornac
from cornac.data import Reader
reader = Reader()
data = reader.read("./ratings.txt", fmt='UIR', sep='\t')

from cornac.eval_methods import RatioSplit

ratio_split = RatioSplit(data=data, test_size=0.2, rating_threshold=4.0, seed=123)
from cornac.models import MF, PMF, BPR

mf = MF(k=10, max_iter=25, learning_rate=0.01, lambda_reg=0.02, use_bias=True)
pmf = PMF(k=10, max_iter=100, learning_rate=0.001, lamda=0.001)
bpr = BPR(k=10, max_iter=200, learning_rate=0.001, lambda_reg=0.01)
mae = cornac.metrics.MAE()
rmse = cornac.metrics.RMSE()
rec_20 = cornac.metrics.Recall(k=20)
ndcg_20 = cornac.metrics.NDCG(k=20)
auc = cornac.metrics.AUC()

from cornac import Experiment

exp = Experiment(eval_method=ratio_split,
models=[mf, pmf, bpr],
metrics=[mae, rmse, rec_20, ndcg_20, auc],
user_based=True)
exp.run()

[tqtg]

I guess you might need to change rating_threshold=1.0.
That’s what Cornac uses to identify feedback values as positive or negative ones.

[ZuoxiYang]

Hi @ZuoxiYang,
Can you show all your code here so that we can reproduce the issue?

I guess you might need to change rating_threshold=1.0.
That’s what Cornac uses to identify feedback values as positive or negative ones.

Thanks for for your nice reply, it has worked well now. Thanks a lot.

I guess you might need to change rating_threshold=1.0.
That’s what Cornac uses to identify feedback values as positive or negative ones.

I guess you might need to change rating_threshold=1.0.
That’s what Cornac uses to identify feedback values as positive or negative ones.

And when I run the "CDR" code:
cdr = CDR(k=10,max_iter=20,learning_rate=0.001,autoencoder_structure=None),

It will console:

Traceback (most recent call last):
File "baseline-cornac-20.py", line 33, in
exp.run()
File "/python3.6/site-packages/cornac/experiment/experiment.py", line 89, in run
user_based=self.user_based)
File "/python3.6/site-packages/cornac/eval_methods/base_method.py", line 448, in evaluate
model.fit(self.train_set, self.val_set)
File "/python3.6/site-packages/cornac/models/cdr/recom_cdr.py", line 136, in fit
self._fit_cdr()
File "/python3.6/site-packages/cornac/models/cdr/recom_cdr.py", line 148, in _fit_cdr
text_feature = self.train_set.item_text.batch_bow(np.arange(n_items)) # bag of word feature
AttributeError: 'NoneType' object has no attribute 'batch_bow'

[ZuoxiYang]

I guess you might need to change rating_threshold=1.0.
That’s what Cornac uses to identify feedback values as positive or negative ones.

About the PCRL model, when I run the example code "
pcrl = PCRL(k=100, z_dims=[300], max_iter=300, batch_size=300, learning_rate=0.001, name='pcrl', trainable=True, verbose=False, w_determinist=True, init_params={'G_r': None, 'G_s': None, 'L_r': None, 'L_s': None})", It will console:

self.aux_data = self.train_set.item_graph.matrix[:self.train_set.num_items, :self.train_set.num_items] # item auxiliary information (items'context in the original paper)
AttributeError: 'NoneType' object has no attribute 'matrix'

[saghiles]

Hi @ZuoxiYang, can you share all your code to run either PCRL or CDR ? I suspect this is due to the fact your data doesn't contain any auxiliary information (in addition to the rating data) required for both models, e.g., context information for PCRL, see line 29 in PCRL's example.

[ZuoxiYang]

Hi @ZuoxiYang, can you share all your code to run either PCRL or CDR ? I suspect this is due to the fact your data doesn't contain any auxiliary information (in addition to the rating data) required for both models, e.g., context information for PCRL, see line 29 in PCRL's example.

Hi, thanks for your reply. My code is similar to the code I've above(I just add the PCRL code in the previous code). Actually, I do not contain the auxiliary data, is it necessary for the PCRL and CDR? and how to add the auxiliary data, just follow the example https://github.com/PreferredAI/cornac/blob/master/examples/pcrl_example.py? and need I provide the actual auxiliary data?

[saghiles]

PCRL and CDR are designed to work with ratings + some auxiliary information. Therefore if your own dataset doesn't come with the auxiliary information that PCRL and CDR are designed for, you should not consider using these algorithms. In case your data contains the desired auxiliary information then, yes you can just follow PCRL/CDR's examples to incorporate this information.

[ZuoxiYang]

PCRL and CDR are designed to work with ratings + some auxiliary information. Therefore if your own dataset doesn't come with the auxiliary information that PCRL and CDR are designed for, you should not consider using these algorithms. In case your data contains the desired auxiliary information then, yes you can just follow PCRL/CDR's examples to incorporate this information.

Thanks a lot. And are there any requirements for the auxiliary information?

[saghiles]

Not sure to understand the question. But if for example we have some dataset with ranting information: (uid, iid, rat) + item graph or network : (iid, iid, 1), then we can use PCRL for example. It really depends on which kind of information our dataset contains.

[ZuoxiYang]

Not sure to understand the question. But if for example we have some dataset with ranting information: (uid, iid, rat) + item graph or network : (iid, iid, 1), then we can use PCRL for example. It really depends on which kind of information our dataset contains.

Well, I just want to express: what's format of the auxiliary information? or when I wan to use the auxiliary information, what's format of the auxiliary information should I process for the model?

[saghiles]

That depends on specific models and implementations. The best resources are the models' papers, models' Cornac examples, models' Cornac documentation. For how to work with auxiliary data using Cornac please consider Cornac Modality classes.

[ZuoxiYang]

That depends on specific models and implementations. The best resources are the models' papers, models' Cornac examples, models' Cornac documentation. For how to work with auxiliary data using Cornac please consider Cornac Modality classes.

Ok, Thanks again.

[saghiles]

Welcome! It seems like the the original issue regarding how one can use his own dataset with Cornac has been solved. Hence, if there is no more question/comment on this specific topic, may I close this issue? Thanks!