def test_model_nrms(mind_resource_path): train_news_file = os.path.join(mind_resource_path, "train", r"news.tsv") train_behaviors_file = os.path.join(mind_resource_path, "train", r"behaviors.tsv") valid_news_file = os.path.join(mind_resource_path, "valid", r"news.tsv") valid_behaviors_file = os.path.join(mind_resource_path, "valid", r"behaviors.tsv") wordEmb_file = os.path.join(mind_resource_path, "utils", "embedding.npy") userDict_file = os.path.join(mind_resource_path, "utils", "uid2index.pkl") wordDict_file = os.path.join(mind_resource_path, "utils", "word_dict.pkl") yaml_file = os.path.join(mind_resource_path, "utils", r"nrms.yaml") if not os.path.exists(train_news_file): download_deeprec_resources( r"https://recodatasets.z20.web.core.windows.net/newsrec/", os.path.join(mind_resource_path, "train"), "MINDdemo_train.zip", ) if not os.path.exists(valid_news_file): download_deeprec_resources( r"https://recodatasets.z20.web.core.windows.net/newsrec/", os.path.join(mind_resource_path, "valid"), "MINDdemo_dev.zip", ) if not os.path.exists(yaml_file): download_deeprec_resources( r"https://recodatasets.z20.web.core.windows.net/newsrec/", os.path.join(mind_resource_path, "utils"), "MINDdemo_utils.zip", ) hparams = prepare_hparams( yaml_file, wordEmb_file=wordEmb_file, wordDict_file=wordDict_file, userDict_file=userDict_file, epochs=1, ) assert hparams is not None iterator = MINDIterator model = NRMSModel(hparams, iterator) assert model.run_eval(valid_news_file, valid_behaviors_file) is not None assert isinstance( model.fit(train_news_file, train_behaviors_file, valid_news_file, valid_behaviors_file), BaseModel, )
def test_nrms_component_definition(mind_resource_path): wordEmb_file = os.path.join(mind_resource_path, "utils", "embedding.npy") userDict_file = os.path.join(mind_resource_path, "utils", "uid2index.pkl") wordDict_file = os.path.join(mind_resource_path, "utils", "word_dict.pkl") yaml_file = os.path.join(mind_resource_path, "utils", r"nrms.yaml") if not os.path.exists(yaml_file): download_deeprec_resources( r"https://recodatasets.z20.web.core.windows.net/newsrec/", os.path.join(mind_resource_path, "utils"), "MINDdemo_utils.zip", ) hparams = prepare_hparams( yaml_file, wordEmb_file=wordEmb_file, wordDict_file=wordDict_file, userDict_file=userDict_file, epochs=1, ) iterator = MINDIterator model = NRMSModel(hparams, iterator) assert model.model is not None assert model.scorer is not None assert model.loss is not None assert model.train_optimizer is not None
def test_model_nrms(tmp): yaml_file = os.path.join(tmp, "nrms.yaml") train_file = os.path.join(tmp, "train.txt") valid_file = os.path.join(tmp, "test.txt") wordEmb_file = os.path.join(tmp, "embedding.npy") if not os.path.exists(yaml_file): download_deeprec_resources( "https://recodatasets.blob.core.windows.net/newsrec/", tmp, "nrms.zip") hparams = prepare_hparams(yaml_file, wordEmb_file=wordEmb_file, epochs=1) assert hparams is not None iterator = NewsIterator model = NRMSModel(hparams, iterator) assert model.run_eval(valid_file) is not None assert isinstance(model.fit(train_file, valid_file), BaseModel)
def dist_eval(args): iterator = MINDIterator model = NRMSModel(hparams, iterator, seed=seed) model.model.load_weights( os.path.join(model_dir, "ckpt_ep{}".format(args.ep))) test_news_file = os.path.join(data_path, "valid", 'news.tsv') test_behaviors_file = os.path.join(data_path, "valid", 'behaviors.{}.tsv'.format(args.fsplit)) group_impr_indexes, group_labels, group_preds = model.run_slow_eval( test_news_file, test_behaviors_file) with open( os.path.join( data_path, 'results/nrms-valid-prediction.{}.txt'.format(args.fsplit)), 'w') as f: for labels, preds in tqdm(zip(group_labels, group_preds)): label_str = ",".join([str(x) for x in labels]) pred_str = ",".join([str(x) for x in preds]) f.write("{}\t{}\n".format(label_str, pred_str))
def test(args): iterator = MINDIterator model = NRMSModel(hparams, iterator, seed=seed, test_mode=True) model.model.load_weights( os.path.join(model_dir, "ckpt_ep{}".format(args.ep))) test_news_file = os.path.join(data_path, "test", 'news.tsv') test_behaviors_file = os.path.join(data_path, "test", 'behaviors.{}.tsv'.format(args.fsplit)) group_impr_indexes, group_labels, group_preds = model.run_slow_eval( test_news_file, test_behaviors_file) with open( os.path.join( data_path, 'results/nrms-test-prediction.{}.txt'.format(args.fsplit)), 'w') as f: for impr_index, preds in tqdm(zip(group_impr_indexes, group_preds)): impr_index += 1 pred_rank = (np.argsort(np.argsort(preds)[::-1]) + 1).tolist() pred_rank = '[' + ','.join([str(i) for i in pred_rank]) + ']' f.write(' '.join([str(impr_index), pred_rank]) + '\n')
def test_nrms_component_definition(tmp): yaml_file = os.path.join(tmp, "nrms.yaml") wordEmb_file = os.path.join(tmp, "embedding.npy") if not os.path.exists(yaml_file): download_deeprec_resources( "https://recodatasets.blob.core.windows.net/newsrec/", tmp, "nrms.zip") hparams = prepare_hparams(yaml_file, wordEmb_file=wordEmb_file, epochs=1) iterator = NewsIterator model = NRMSModel(hparams, iterator) assert model.model is not None assert model.scorer is not None assert model.loss is not None assert model.train_optimizer is not None
userDict_file = os.path.join(data_path, "utils", "uid2index.pkl") wordDict_file = os.path.join(data_path, "utils", "word_dict_all.pkl") yaml_file = os.path.join(data_path, "utils", r'nrms.yaml') entityDict_file = os.path.join(data_path, "utils", "entity_dict_all.pkl") entity_embedding_file = os.path.join(data_path, "utils", "entity_embeddings_5w_100_all.npy") context_embedding_file = os.path.join(data_path, "utils", "context_embeddings_5w_100_all.npy") hparams = prepare_hparams(yaml_file, wordEmb_file=wordEmb_file, \ wordDict_file=wordDict_file, userDict_file=userDict_file, \ epochs=epochs, entityEmb_file=entity_embedding_file, contextEmb_file=context_embedding_file, \ entityDict_file=entityDict_file, show_step=10) print(hparams) iterator = MINDIterator model = NRMSModel(hparams, iterator, seed=seed) model_path = os.path.join(data_path, "model") model.model.load_weights(os.path.join(model_path, "my_nrms_ckpt")) f = open(test_behaviors_file) print('total test samples:', len(f.readlines())) f.close() group_impr_indexes, group_labels, group_preds = model.run_fast_eval(test_news_file, test_behaviors_file, test=1) import numpy as np from tqdm import tqdm with open(os.path.join(data_path, 'my_prediction.txt'), 'w') as f: for impr_index, preds in tqdm(zip(group_impr_indexes, group_preds)): impr_index += 1 pred_rank = (np.argsort(np.argsort(preds)[::-1]) + 1).tolist()
download_deeprec_resources(mind_url, os.path.join(data_path, 'train'), mind_train_dataset) if not os.path.exists(valid_news_file): download_deeprec_resources(mind_url, \ os.path.join(data_path, 'valid'), mind_dev_dataset) if not os.path.exists(yaml_file): download_deeprec_resources(r'https://recodatasets.blob.core.windows.net/newsrec/', \ os.path.join(data_path, 'utils'), mind_utils) hparams = prepare_hparams(yaml_file, wordEmb_file=wordEmb_file, \ wordDict_file=wordDict_file, userDict_file=userDict_file, \ epochs=epochs, show_step=10) print("[NRMS] Config,", hparams) iterator = MINDIterator model = NRMSModel(hparams, iterator, seed=seed) print("[NRMS] First run:", model.run_eval(valid_news_file, fast_valid_behaviors_file)) model.fit(train_news_file, train_behaviors_file, valid_news_file, fast_valid_behaviors_file, model_save_path=model_dir) # res_syn = model.run_eval(valid_news_file, valid_behaviors_file) # print(res_syn)
os.path.join(data_path, 'test'), mind_test_dataset) if not os.path.exists(yaml_file): download_deeprec_resources(r'https://recodatasets.blob.core.windows.net/newsrec/', \ os.path.join(data_path, 'utils'), mind_utils) hparams = prepare_hparams(yaml_file, wordEmb_file=wordEmb_file, wordDict_file=wordDict_file, userDict_file=userDict_file, batch_size=batch_size, epochs=epochs, show_step=10) print(hparams) iterator = MINDIterator model = NRMSModel(hparams, iterator, seed=seed) # print(model.run_eval(valid_news_file, valid_behaviors_file)) model.fit(train_news_file, train_behaviors_file, valid_news_file, valid_behaviors_file) # res_syn = model.run_eval(valid_news_file, valid_behaviors_file) # print(res_syn) # pm.record("res_syn", res_syn) model_path = os.path.join(model_path, "model") os.makedirs(model_path, exist_ok=True) model.model.save_weights(os.path.join(model_path, "nrms_ckpt")) group_impr_indexes, group_labels, group_preds = model.run_slow_eval(test_news_file, test_behaviors_file)
wordDict_file=wordDict_file, userDict_file=userDict_file, vertDict_file=vertDict_file, subvertDict_file=subvertDict_file, batch_size=batch_size, epochs=epochs, show_step=10) logging.info(hparams) # ## Train the NRMS model if model_type == 'nrms': iterator = MINDIterator model = NRMSModel(hparams, iterator, seed=seed) elif model_type == 'naml': iterator = MINDAllIterator model = NAMLModel(hparams, iterator, seed=seed) elif model_type == 'npa': iterator = MINDIterator model = NPAModel(hparams, iterator, seed=seed) elif model_type == 'nrmma': iterator = MINDAllIterator model = NRMMAModel(hparams, iterator, seed=seed) else: raise NotImplementedError(f"{exp_name} is not implemented") # In[8]: model_path = os.path.join(exp_path, model_type)
download_deeprec_resources( r'https://recodatasets.blob.core.windows.net/newsrec/', os.path.join(data_root, 'utils'), mind_utils) hparams = prepare_hparams(yaml_file, wordEmb_file=wordEmb_file, wordDict_file=wordDict_file, userDict_file=userDict_file, batch_size=batch_size, epochs=epochs, show_step=10) logger.debug(f"hparams: {hparams}") iterator = MINDIterator model = NRMSModel(hparams, iterator, seed=seed) logger.info(model.run_eval(valid_news_file, valid_behaviors_file)) model.fit(train_news_file, train_behaviors_file, valid_news_file, valid_behaviors_file) res_syn = model.run_eval(valid_news_file, valid_behaviors_file) logger.debug(f"res_syn: {res_syn}") sb.glue("res_syn", res_syn) model_path = os.path.join(BASE_DIR, "ckpt") os.makedirs(model_path, exist_ok=True) model.model.save_weights(os.path.join(model_path, "nrms_ckpt"))
# subvertDict_file = os.path.join(data_path, "utils", "subvert_dict.pkl") # vertDict_file = os.path.join(data_path, "utils", "vert_dict.pkl") yaml_file = os.path.join(data_path, "utils", '{}.yaml'.format(opt.model_name)) hparams = prepare_hparams(yaml_file, wordEmb_file=wordEmb_file, wordDict_file=wordDict_file, userDict_file=userDict_file, batch_size=batch_size, epochs=epochs, show_step=10) print(hparams) iterator = MINDIterator if opt.model_name == 'nrms': model = NRMSModel(hparams, iterator, seed=seed) elif opt.model_name == 'npa': # model can not save model = NPAModel(hparams, iterator, seed=seed) elif opt.model_name == 'lstur': model = LSTURModel(hparams, iterator, seed=seed) elif opt.model_name == 'naml': # problematic model = NAMLModel(hparams, iterator, seed=seed) # print(model.run_slow_eval(news_file, valid_behaviors_file)) model.fit(news_file, train_behaviors_file, news_file, valid_behaviors_file) # model_path = os.path.join(model_path, "model") # os.makedirs(model_path, exist_ok=True) # model.model.save_weights(os.path.join(model_path, "nrms_ckpt"))