Giter VIP home page Giter VIP logo

recsys2019's People

Contributors

milakyr avatar pjankiewicz avatar sienkowski avatar wrosinski avatar

Stargazers

 avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar  avatar

Watchers

 avatar  avatar  avatar  avatar

recsys2019's Issues

Ideas

  • vowpal wabbit model with interactions
  • improve item similarity beyond a set intersection - maybe weighted properties by 1/freq or model based
  • graph-based similarity of users: user->item->user->item or user->user->item
  • proper ensemble
  • factorization machines (including field aware)
  • sequence models

Issue of running 'Current process (full)' - no comp_v0_selected.csv

I have followed exactly of your instructions for 'Current process (full)', but when I ran python generate_data_parallel_all.py, I found that it requires comp_v0_selected.csv (not sure what it is) but this file has not been generated in the previous steps, so could you give me some hints? Thanks.

Przewidywanie klikniętych indeksów

Do przewidzenia jest kliknięty indeks itemu.

Każda sesja składa się ze słownika {'length': 25, 'ind': [0, 0, 0, 0]} gdzie length to liczba wyników wyszukiwania a ind to indeksy klikniętych wyników. Jeden użytkownik może mieć wiele takich sesji.

BU7YM5MD2HZB	[{'length': 25, 'ind': [0, 0, 0, 0]}]
SU92UN3VX8S2	[{'length': 25, 'ind': [5, 4, 11]}, {'length': 25, 'ind': [0]}, {'length': 25, 'ind': [0, 6, 11]}, {'length': 25, 'ind': [4]}, {'length': 25, 'ind': [0, 6, 12]}, {'length': 25, 'ind': [1, 2]}, {'length': 25, 'ind': [0, 8]}, {'length': 25, 'ind': [0, 0, 1, 6]}, {'length': 25, 'ind': [6]}, {'length': 17, 'ind': [8]}, {'length': 23, 'ind': [0, 5]}]
ZYOFGZBSOCNO	[{'length': 25, 'ind': [0]}]
H25VWNRELOG0	[{'length': 25, 'ind': [0]}]
318914ILC032	[{'length': 25, 'ind': [7, 0]}]
E0ZABCG33DUJ	[{'length': 25, 'ind': [5, 2]}]
M0C2Y3U1MM89	[{'length': 25, 'ind': [21, 10]}]
YZVDXEGPHJIU	[{'length': 25, 'ind': [0]}, {'length': 25, 'ind': [2, 3]}]
EX651IQMN4M3	[{'length': 25, 'ind': [0, 1]}]
JZA19MUU8GE1	[{'length': 24, 'ind': [2, 16]}, {'length': 25, 'ind': [2]}]
113S11DCJTUJ	[{'length': 25, 'ind': [4]}]
DVRKLT2AA3N9	[{'length': 25, 'ind': [12]}]
KTBKO02O4E3E	[{'length': 25, 'ind': [8]}]
MFJEX60PPKG7	[{'length': 25, 'ind': [6]}]
PMNSJLDEYUK1	[{'length': 25, 'ind': [0]}]
XVU8RFF35RPL	[{'length': 25, 'ind': [0, 0]}, {'length': 25, 'ind': [0]}, {'length': 25, 'ind': [0]}]
9TLMAVX1VJZA	[{'length': 25, 'ind': [0]}]
9YGB7B1X3WH6	[{'length': 25, 'ind': [1]}]
ESIJ4NIOOZWI	[{'length': 25, 'ind': [18]}]
UAL8B5S1F9XD	[{'length': 25, 'ind': [7]}]

Zbiór danych: https://storage.googleapis.com/logicai-recsys2019/problems/clickind.csv

Potrzebny jest model/formuła, która pod warunkiem poprzednich, obecnej sesji i rankingu rozpatrywanego elementu zwróci prawdopodobieństwo jego kliknięcia.

Zwykle kliknięcia następują w porządku, w tym momencie klasyfikacja może być np. taka:

def classify_sequence(seq):
    if len(seq) == 0:
        return "empty"
    elif len(set(seq)) == 1:
        return "constant"
    elif len(set(seq)) == len(seq) and min(seq) == seq[0] and max(seq) == seq[-1]:
        return "ideal sequence"
    elif seq == sorted(seq):
        return "non ideal sequence"
    elif len(set(seq)) == len(seq) and min(seq) == seq[-1] and max(seq) == seq[0]:
        return "ideal sequence rev"
    elif seq == sorted(seq, reverse=True):
        return "non ideal sequence rev"
    else:
        return "other"  

Przetestowanie Vowpal Wabbit

Przetestowanie modeli VW

  • konwersja zmiennych na format VW
  • test modeli liniowych z interakcjami (user x properties, platform x properties, device x platform x properties itp)

Feature selection

Teraz feature'y są wrzucone trochę nadmiarowo. Trzeba jakies wyrzucić ponieważ brakuje pamięci na nowe (być może lepsze).

Write ACM Paper

Things to write about

  • [Magda] Introduction
  • [Magda] Problem formulation
  • [Magda] Dataset description
  • [Mila] metric NDCG vs MRR in the context of RecSys (can be shown that NDCG is almost equivalent to MRR under the circumstances that only 1 hotel is clicked)
  • [PJ] methodology: transposition of data (item per row), quick validation process, iteratively check the idea
  • [PJ] features: numerical features, ranking transformations within the clickout group, lag features - compare items to next/prev items, accumulators (prevents leaks)
  • additional features extracted from properties (all the data_prep scripts)
  • feature importance - both meta features and particular features (numeric)
  • too revealing data - what would happen if there was no time and/or rank
  • [Mila] model (LightGBM) - parameter tuning, C++ custom objective
  • blend - list of models used (with parameters)
  • possible improvements - don't include the time of the click in the dataset

Jakiej wersji interpretera używamy

  1. Sugerowane jest odpalanie części kodu w pypy, które (stabilne) wspiera 3.5 co najwyżej
  2. Zobaczyłem tu i tam f-stringi (f"{foo}") które jest nowsze.

Nowe feature'y

Taka sytuacja jest dziwna - item na 15 pozycji został kliknięty i ma znacząco różne item_id

image

Runtime error for quick validation

Hi, I have tried to run quick validation and I followed exactly of you instructions. After fixed several bugs in the code I finally reached the last step, but when I ran python quick_validate.py, I encountered a Key Error of 'last_event_ts', I deleted code related to last_event_ts but more Key Errors raised (e.g. last_item_clickout), so could you please provide me some tips of fixing it? Thanks a lot.

Rewrite data generation to Scala

  • clickout_item_clicks
  • clickout_item_impressions
  • clickout_item_platform_clicks
  • clickout_user_item_clicks
  • clickout_user_item_impressions
  • identical_impressions_item_clicks
  • interaction_deal_freq
  • interaction_img_diff_ts
  • interaction_img_freq
  • interaction_info_freq
  • is_impression_the_same
  • last_10_actions
  • last_event_ts
  • last_filter
  • last_filter_selection
  • last_item_clickout
  • last_item_index
  • last_poi
  • last_poi_item_clicks
  • last_poi_item_impressions
  • last_sort_order
  • was_interaction_deal
  • was_interaction_img
  • was_interaction_info
  • was_item_searched

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.