def predict_main(list_posts, data_folder_path, saved_model_path,
temp_save_path):
model_type = 'hier_fuse'
word_feats_raw = [{
'emb': 'elmo',
's_enc': 'rnn',
'm_id': '11'
}, {
'emb': 'glove',
's_enc': 'rnn',
'm_id': '21'
}]
sent_enc_feats_raw = [{'emb': 'bert_pre', 'm_id': '1'}]
conf_dict_com = {
'poss_sent_enc_feats_emb_dict': {
'bert_pre': 768
},
'poss_word_feats_emb_dict': {
'glove': 300,
'ling': 33,
'elmo': 3072
}
}
classi_probs_label_info = [[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13],
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13]]
data_dict = load_data(list_posts, 16, 35, 198,
data_folder_path + 'esp_class_maps.txt')
# for cl_post in data_dict['text_sen']:
# print(cl_post)
word_feats, word_feat_str = word_featurize(
word_feats_raw, model_type, data_dict,
conf_dict_com['poss_word_feats_emb_dict'], False, False,
data_folder_path, temp_save_path, True)
sent_enc_feats, sent_enc_feat_str = sent_enc_featurize(
sent_enc_feats_raw, model_type, data_dict,
conf_dict_com['poss_sent_enc_feats_emb_dict'], False, False,
data_folder_path, temp_save_path, True)
test_generator = TestGenerator(
np.arange(data_dict['test_st_ind'], data_dict['test_en_ind']),
word_feats, sent_enc_feats, data_dict, 64)
mod_op_list = gen_raw_output(len(classi_probs_label_info),
saved_model_path, test_generator,
data_dict['NUM_CLASSES'])
pred_vals = evaluate_model(mod_op_list, data_dict, classi_probs_label_info)
pred_categories = []
for pred_val_set in pred_vals:
pred_categories.append(
[data_dict['FOR_LMAP'][x] for x in pred_val_set])
return pred_categories
conf_dict_com["save_folder_name"],
conf_dict_com['use_conf_scores'],
conf_dict_com['multi_task_tl'],
conf_dict_com['classi_loss_weight'], prime_filename,
aux_filename_str, conf_dict_com['augment_data'],
conf_dict_com['single_inp_tasks_list'],
conf_dict_com['sep_inp_tasks_list'],
conf_dict_com['uncorr_c_pairs_filename'],
conf_dict_com['beta'], conf_dict_com['label_corr_setting'],
conf_dict_com['train_ratio'])
for model_type in conf_dict["model_types"]:
for word_feats_raw in conf_dict["word_feats_l"]:
word_feats, word_feat_str = word_featurize(
word_feats_raw, model_type, data_dict,
conf_dict_com['poss_word_feats_emb_dict'],
conf_dict_com['use_saved_word_feats'],
conf_dict_com['save_word_feats'],
conf_dict_com["data_folder_name"],
conf_dict_com["save_folder_name"],
conf_dict_com["test_mode"], prime_filename)
for sep_task_tup in conf_dict_com['sep_inp_tasks_list']:
sep_task_name, sep_task_dict = sep_task_tup
sep_task_dict['word_feats'], sep_task_dict[
'word_feat_str'] = word_featurize(
word_feats_raw, model_type, sep_task_dict,
conf_dict_com['poss_word_feats_emb_dict'],
conf_dict_com['use_saved_word_feats'],
conf_dict_com['save_word_feats'],
conf_dict_com["data_folder_name"],
conf_dict_com["save_folder_name"],
conf_dict_com["test_mode"],
get_filename(sep_task_dict['filename']))
f_tsv.write("model\tword feats\tsent feats\ttrans\tclass imb\tcnn fils\tcnn kernerls\tthresh\trnn dim\tatt dim\tpool k\tstack RNN\tf-We\tf-Ma\tf-Mi\tacc\tp-We\tp_Ma\tp_Mi\tr_We\tr_Ma\tr_Mi\tstd_f_we\trnn type\tl rate\tb size\tdr1\tdr2\ttest mode\tclassi probs labels\n")
elif data_dict['prob_type'] == 'binary':
f_tsv.write("model\tword feats\tsent feats\ttrans\tclass imb\tcnn fils\tcnn kernerls\tthresh\trnn dim\tatt dim\tpool k\tstack RNN\tf\tp\tr\tacc\tstd_f\trnn type\tl rate\tb size\tdr1\tdr2\ttest mode\tclassi probs labels\n")
print("max # sentences: %d, max # words per sentence: %d, max # words per post: %d" % (data_dict['max_num_sent'], data_dict['max_words_sent'], data_dict['max_post_length']))
metr_dict = init_metr_dict(data_dict['prob_type'])
classi_probs_label_str = str(conf_dict_com['classi_probs_label_info'])[2:-2].replace('], [', '+').replace(', ','_')
for conf_dict in conf_dict_list:
for prob_trans_type in conf_dict["prob_trans_types"]:
trainY_list, trainY_noncat_list, num_classes_list, bac_map = transform_labels(data_dict['lab'][:data_dict['train_en_ind']], prob_trans_type, conf_dict_com['test_mode'], conf_dict_com["save_folder_name"], data_dict['NUM_CLASSES'], data_dict['prob_type'], conf_dict_com['classi_probs_label_info'], classi_probs_label_str, conf_dict_com['use_saved_data_stuff'], conf_dict_com['save_data_stuff'])
for class_imb_flag in conf_dict["class_imb_flags"]:
loss_func_list, nonlin, out_vec_size_list, cw_list = class_imb_loss_nonlin(trainY_noncat_list, class_imb_flag, num_classes_list, prob_trans_type, conf_dict_com['test_mode'], conf_dict_com["save_folder_name"], conf_dict_com['use_saved_data_stuff'], conf_dict_com['save_data_stuff'], classi_probs_label_str)
for model_type in conf_dict["model_types"]:
for word_feats_raw in conf_dict["word_feats_l"]:
word_feats, word_feat_str = word_featurize(word_feats_raw, model_type, data_dict, conf_dict_com['poss_word_feats_emb_dict'], conf_dict_com['use_saved_word_feats'], conf_dict_com['save_word_feats'], conf_dict_com["data_folder_name"], conf_dict_com["save_folder_name"], conf_dict_com["test_mode"])
for sent_enc_feats_raw in conf_dict["sent_enc_feats_l"]:
sent_enc_feats, sent_enc_feat_str = sent_enc_featurize(sent_enc_feats_raw, model_type, data_dict, conf_dict_com['poss_sent_enc_feats_emb_dict'], conf_dict_com['use_saved_sent_enc_feats'], conf_dict_com['save_sent_enc_feats'], conf_dict_com["data_folder_name"], conf_dict_com["save_folder_name"], conf_dict_com["test_mode"])
for num_cnn_filters in conf_dict["num_cnn_filters"]:
for max_pool_k_val in conf_dict["max_pool_k_vals"]:
for cnn_kernel_set in conf_dict["cnn_kernel_sets"]:
cnn_kernel_set_str = str(cnn_kernel_set)[1:-1].replace(', ','_')
for rnn_type in conf_dict["rnn_types"]:
for rnn_dim in conf_dict["rnn_dims"]:
for att_dim in conf_dict["att_dims"]:
for stack_rnn_flag in conf_dict["stack_rnn_flags"]:
mod_op_list_save_list = []
for thresh in conf_dict["threshes"]:
startTime = time.time()
info_str = "model: %s, word_feats = %s, sent_enc_feats = %s, classi_probs_label_info = %s, prob_trans_type = %s, class_imb_flag = %s, num_cnn_filters = %s, cnn_kernel_set = %s, rnn_type = %s, rnn_dim = %s, att_dim = %s, max_pool_k_val = %s, stack_rnn_flag = %s, thresh = %s, test mode = %s" % (model_type,word_feat_str,sent_enc_feat_str,classi_probs_label_str,prob_trans_type,class_imb_flag,num_cnn_filters,cnn_kernel_set,rnn_type,rnn_dim,att_dim,max_pool_k_val,stack_rnn_flag, thresh, conf_dict_com["test_mode"])
fname_part = ("%s~%s~%s~%s~%s~%s~%s~%s~%s~%s~%s~%s~%s~%s" % (model_type,word_feat_str,sent_enc_feat_str,classi_probs_label_str,prob_trans_type,class_imb_flag,num_cnn_filters,cnn_kernel_set_str,rnn_type,rnn_dim,att_dim,max_pool_k_val,stack_rnn_flag, conf_dict_com["test_mode"]))