---
tags:
- recommender-system
- sasrec
datasets:
- ml-1m
---

# SASRec - ml-1m

## Model Description
SASRec (Self-Attentive Sequential Recommendation) is a sequential recommendation model that uses self-attention mechanisms to model user's historical behavior data.

## Training Results
### Best Valid Results
- **ndcg@10**: 0.1261
- **hit@10**: 0.2486
- **averagepopularity@10**: 756.7929
- **giniindex@10**: 0.7006
- **itemcoverage@10**: 0.759
- **shannonentropy@10**: 0.003
- **tailpercentage@10**: 0.0035

### Test Results
- **ndcg@10**: 0.1252
- **hit@10**: 0.2423
- **averagepopularity@10**: 764.8602
- **giniindex@10**: 0.7061
- **itemcoverage@10**: 0.7535
- **shannonentropy@10**: 0.003
- **tailpercentage@10**: 0.0034

## Configuration
The model was trained with the following configuration:
```yaml
ENTITY_ID_FIELD: entity_id
HEAD_ENTITY_ID_FIELD: head_id
ITEM_ID_FIELD: item_id
ITEM_LIST_LENGTH_FIELD: item_length
LABEL_FIELD: label
LIST_SUFFIX: _list
MAX_ITEM_LIST_LENGTH: 50
MODEL_INPUT_TYPE: InputType.POINTWISE
MODEL_TYPE: ModelType.SEQUENTIAL
NEG_PREFIX: neg_
POSITION_FIELD: position_id
RATING_FIELD: rating
RELATION_ID_FIELD: relation_id
TAIL_ENTITY_ID_FIELD: tail_id
TIME_FIELD: timestamp
USER_ID_FIELD: user_id
additional_feat_suffix: null
alias_of_entity_id: null
alias_of_item_id: null
alias_of_relation_id: null
alias_of_user_id: null
attn_dropout_prob: 0.5
benchmark_filename: null
checkpoint_dir: saved
clip_grad_norm: null
dataloaders_save_path: null
dataset: ml-1m
dataset_save_path: null
device: cuda
discretization: null
embedding_size: 64
enable_amp: false
enable_scaler: false
entity_kg_num_interval: '[0,inf)'
epochs: 500
eval_args:
  group_by: user
  mode:
    test: full
    valid: full
  order: TO
  split:
    LS: valid_and_test
eval_batch_size: 12288
eval_step: 1
eval_type: EvaluatorType.RANKING
field_separator: "\t"
filter_inter_by_user_or_item: true
gpu_id: '0'
hidden_act: gelu
hidden_dropout_prob: 0.5
hidden_size: 64
initializer_range: 0.02
inner_size: 256
item_inter_num_interval: '[10,inf)'
kg_reverse_r: false
layer_norm_eps: 1.0e-12
learner: adam
learning_rate: 0.0005
load_col:
  inter:
  - user_id
  - item_id
  - rating
  - timestamp
local_rank: 0
log_wandb: false
loss_decimal_place: 4
loss_type: CE
metric_decimal_place: 4
metrics:
- NDCG
- Hit
- AveragePopularity
- GiniIndex
- ItemCoverage
- ShannonEntropy
- TailPercentage
model: SASRec
n_heads: 2
n_layers: 2
normalize_all: null
normalize_field: null
numerical_features: []
preload_weight: null
relation_kg_num_interval: '[0,inf)'
repeatable: true
reproducibility: true
require_pow: false
rm_dup_inter: null
save_dataloaders: false
save_dataset: false
seed: 2020
seq_len: null
seq_separator: ' '
show_progress: false
shuffle: true
single_spec: true
state: INFO
stopping_step: 10
test_neg_sample_args:
  distribution: uniform
  sample_num: none
threshold: null
topk:
- 10
train_batch_size: 12288
train_neg_sample_args:
  alpha: none
  candidate_num: 0
  distribution: none
  dynamic: false
  sample_num: none
transform: null
unload_col: null
unused_col:
  inter:
  - rating
use_gpu: true
user_inter_num_interval: '[10,inf)'
val_interval:
  rating: '[3,inf)'
valid_metric: NDCG@10
valid_metric_bigger: true
valid_neg_sample_args:
  distribution: uniform
  sample_num: none
wandb_project: recbole
weight_decay: 0.0
worker: 0

```

## Usage