Created by: adrianhust

data layer 报错；

def cluster_reader_cluster(base_dir, node_id):
    def data_reader():
        paths = os.listdir(base_dir)
        for file_dir in paths:
            files = os.listdir(base_dir + '/' + file_dir)
            for fi in files:
                with open(base_dir + '/' + file_dir + '/' + fi, "r") as f:
                #with open(base_dir + '/' + file_dir + '/' + fi, "r") as f:
                    for line in f:
                        items = line.strip().split('\t')
                        if len(items) < 138:
                            continue
                        clk = int(items[2])
                        pv = int(items[1])
                        fea = []
                        idx_fea = []
                        url_fea = []
                        ideaid_fea = []
                        for slot_val in items[15:]:
                            slot, value = slot_val.split(':')
                            try:
                                value = float(value)
                                slot = int(slot)
                            except:
                                value = 0
                            if slot < 55314:
                                fea.append((slot, value))
                            elif slot >= 55314 and slot < 55317: #dim = 3
                                value = int(value)
                                idx_fea.append((slot, value))
                                #idx_fea.append([value])
                            elif slot >= 55317 and slot < 55767: #dim = 450
                                value = int(value)
                                url_fea.append((slot, value))
                                #url_fea.append([value])
                            elif slot >= 55767 and slot < 55877: #dim = 110
                                value = int(value)
                                ideaid_fea.append((slot, value))
                                #ideaid_fea.append([value])
                        #pdb.set_trace()
                        yield fea, idx_fea, url_fea, ideaid_fea, clk

    return data_reader

data reader 返回多个字段数据，带slot 的格式是：[(slot, value)]， 不带就是[(value)]

定义的input

   img_input = paddle.layer.data(
        name="img_input", type=paddle.data_type.sparse_vector(img_fea_dim))

    idx_input = paddle.layer.data(
        name="idx_input", type=paddle.data_type.integer_value_sequence(idx_fea_dim))

    url_input = paddle.layer.data(
        name="url_input", type=paddle.data_type.integer_value_sequence(url_fea_dim))

    ideaid_input = paddle.layer.data(
        name="ideaid_input", type=paddle.data_type.integer_value_sequence(ideaid_fea_dim))

    label = paddle.layer.data("label", paddle.data_type.integer_value(class_dim))

feeding

feeding = {'img_input': 0, 'idx_input':1, 'url_input':2, 'ideaid_input': 3, 'label': 4}

debug 跟了一下，感觉是IndexScanner 部门报错，其他类型的都可以正常过

感觉是输入数据跟到 input 的data layer 部分出了问题

感谢！