This project involves using multi-layer LSTMs to predict the sales problem.
Column | Type | Meaning |
---|---|---|
日期 | date | time of data recording |
浏览量 | int | the number of times users view the page on the e-commerce platform |
访客数 | int | the number of users to e-commerce platform pages |
人均浏览量 | float | the average number of times every user views a page on an e-commerce platform in a day |
平均停留时间 | float | the average time spent by users on the page |
跳失率 | float | the proportion of visits where users enter through the corresponding portal and leave after visiting only one page to the total number of visits to that page |
成交客户数 | int | the number of customers who successfully paid |
成交单量 | int | the number of orders successfully paid |
成交金额 | int | the total amount of successful payments |
客单价 | float | the average amount of goods purchased per user |
成交商品件数 | int | the number of goods successfully paid for |
下单客户数 | int | the number of customers who have placed orders |
下单单量 | int | the number of orders placed |
下单金额 | int | the total amount of orders placed |
下单商品件数 | int | the number of goods ordered |
- feature_columns :
columns used as features in the csv dataset, with columns numbered 0, 1, 2,···
- label_columns :
columns used as labels in the csv dataset, with columns numbered 0, 1, 2,···
- predict_day :
predict how many days in the future
- input_size :
the size of input layer, that is, the number of columns used as features
- output_size :
the size of output layer, that is, the number of columns used as labels
- hidden_size :
the size of hidden layer
- lstm_layers :
the number of layers of lstm
- dropout_rate :
dropout probability
- time_step :
how many days before to predict the next day
- do_train :
whether to train the model
- do_predict :
whether the model is used for prediction
- add_train :
whether to continue training on the trained weights
- shuffle_train_data :
whether to randomly disrupt the training data
- use_cuda :
whether to use GPU training
- train_data_rate :
the ratio of training data to total data
- valid_data_rate :
the ratio of validation data to training_data
- batch_size :
the number of samples passed to the model for training in a epoch
- learning_rate :
learning rate
- epoch :
the number of times the model is trained
- patience :
how many epochs to train and stop if the validation set does not improve
- random_seed :
random seed, guaranteed reproducible
- do_continue_train :
take the final state of the previous training as the next init state for each training
- debug_mode :
In debugging mode, it is to run through the code and pursue speed
- debug_num :
debugging with only debug_num pieces of data
- train_data_path :
dataset save path
- model_save_path :
model weights save path
- figure_save_path :
prediction result save path
- log_save_path :
training log save path
- do_log_print_to_screen :
whether to display the log and training process on the screen
- do_log_save_to_file :
whether to record the config and training process
- do_figure_save :
whether to save the prediction result image
- do_train_visualized :
training loss visualization