https://github.com/ghazaleze/recommendersystem_lightfm
with Kaggle datasets
https://github.com/ghazaleze/recommendersystem_lightfm
dataset lightfm pandas python recommender-system
Last synced: 4 days ago
JSON representation
with Kaggle datasets
- Host: GitHub
- URL: https://github.com/ghazaleze/recommendersystem_lightfm
- Owner: GhazaleZe
- License: mit
- Created: 2020-07-15T06:31:25.000Z (almost 5 years ago)
- Default Branch: master
- Last Pushed: 2020-08-23T18:58:13.000Z (almost 5 years ago)
- Last Synced: 2025-04-01T12:49:17.492Z (3 months ago)
- Topics: dataset, lightfm, pandas, python, recommender-system
- Language: Python
- Homepage:
- Size: 5.44 MB
- Stars: 6
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# RecommenderSystem_lightFM :blush:
## Overview:
Recommender systems are a very popular topic in e-commerce. They are frameworks and engines which help us to implement a recommender system easier. One of these engines is [LightFM](https://making.lyst.com/lightfm/docs/home.html) which is a satisfiable python framework.
This project is part of a 4-person team project, each person has worked on a special package my package has been **LightFM**.
You can see our final report in **Final_Report_Group8.pdf**.
## Goal:
The goal is to implement a recommender system with LightFM with 3 of [Kaggle](https://www.kaggle.com/) datasets.
3 data set and their related files are:
### 1.[Restaurant](https://www.kaggle.com/uciml/restaurant-data-with-consumer-ratings/metadata)
files:
- collabrative_restaurant.py (main file)
- geoplaces2.json
- rating_final_U1011.json
- userprofile.json
- resturant_json.py
### 2.Book
files:
- book.py (main file)
- test_train.py
- book_wrap_vs_BPR.py
- book_eval_K_OS.py
- mainbookup_lim1.json (10000 rows of main data in randomly picked)
### 3.[Size for clothes](https://www.kaggle.com/rmisra/clothing-fit-dataset-for-size-recommendation/metadata)
files:
- size_test&trainWRAPvsBPR.py
- size_test&trainWRAP.py
- random_pick.py
- renttherunway_lim.json (10000 rows of main data in randomly picked)
- sizeRs.py (main file)
## Installation:
I used **Conda** in **Pycharm** and install **LightFM** with:
```
conda install -c conda-forge lightfm
conda install -c conda-forge/label/gcc7 lightfm
conda install -c conda-forge/label/cf201901 lightfm
conda install -c conda-forge/label/cf202003 lightfm
```
## Data and LightFM:
One of the most important challenges is how to give the data to the package. First, read the Json file and create a dataset of lightfm :
```python
f = open('rating_final_U1011.json', )
ff = open('userprofile.json', )
df = open(r'geoplaces2.json')
data_User = json.load(ff)
data_item = json.load(df)
data = json.load(f)
dataset = Dataset()
```
Then fit the dataset with your data:
```python
dataset.fit((x['userID'] for x in data),
(x['placeID'] for x in data), (x['budget'] for x in data_User),(x['price'] for x in data_item))
```
Now it's possible to create the matrixes:
```python
(interactions, weights) = dataset.build_interactions(((x['userID'], x['placeID']) for x in data))
print(repr(interactions))
user_interactions = dataset.build_user_features((x['userID'], [x['budget']]) for x in data_User)
print(repr(user_interactions))
item_interactions = dataset.build_item_features((x['placeID'], [x['price']]) for x in data_item)
print(repr(item_interactions))
```
This package is a model base package define and fit package:
```python
alpha = 1e-05
epochs = 70
num_components = 32
model = LightFM(no_components=num_components,
loss='warp',
learning_schedule='adadelta',
user_alpha=alpha,
item_alpha=alpha)
```
For testing and validating the model you need to split data to test and train like in test_train.py.
Testing **learning_schedule** adadelta vs adagrad for cloth dataset:
Testing **loss** WARP vs BPR for cloth dataset:
## Resourses:
http://www2.informatik.uni-freiburg.de/~cziegler/BX/
https://making.lyst.com/lightfm/docs/home.html
https://github.com/lyst/lightfm
## Support:
Reach out to me at [email protected].
Thanks [@alirezaomidi](https://github.com/alirezaomidi) :sweat_smile: