kuhyx/WUT_Computer_Science
1
0
Fork 0
mirror of https://github.com/kuhyx/WUT_Computer_Science.git synced 2026-07-06 20:23:10 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

fix: remove duplicated l1, l2 and cityblock metricss

Browse Source
This commit is contained in:
Krzysztof Rudnicki 2023-06-11 23:26:13 +02:00
parent e05fef37a2
commit f7c66e552d
3 changed files with 20 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
final/code/main.py
View File

@ -299,6 +299,7 @@ def handle_arguments():
def auto_mode(data_limit=-1, seed=42, anime="RANDOM"): def auto_mode(data_limit=-1, seed=42, anime="RANDOM"):
print("Started auto mode") print("Started auto mode")
algorithm_spread = ['auto', 'brute'] algorithm_spread = ['auto', 'brute']
metric_spread = ['manhattan', 'euclidean', 'cosine']
neighbor_spread = [5, "sqrt", "half", "log", "n-1"] neighbor_spread = [5, "sqrt", "half", "log", "n-1"]
# No reason to access and waste computational power every time we run the simulation # No reason to access and waste computational power every time we run the simulation
starting_rating_data, starting_anime_contact_data, starting_rows_number = get_data( starting_rating_data, starting_anime_contact_data, starting_rows_number = get_data(
@ -310,7 +311,7 @@ def auto_mode(data_limit=-1, seed=42, anime="RANDOM"):
for algorithm in algorithm_spread: for algorithm in algorithm_spread:
possibleMetrics = [] possibleMetrics = []
if algorithm != 'auto': if algorithm != 'auto':
possibleMetrics = sorted(VALID_METRICS_SPARSE[algorithm]) possibleMetrics = metric_spread
print("testing for algorithm: ", algorithm, possibleMetrics) print("testing for algorithm: ", algorithm, possibleMetrics)
if possibleMetrics == []: if possibleMetrics == []:
possibleMetrics = [""] possibleMetrics = [""]

BIN
final/report/report.pdf
View File

Binary file not shown.

26
final/report/report.tex
View File

@ -20,23 +20,33 @@ We decided to use collaborative filtering to develop our model, It makes persona
We represent anime data-set as embedding vector \\ We represent anime data-set as embedding vector \\
We use K-nearest neighbors model and decided to test it out with different metrics, neighbors and algorithms \\ We use K-nearest neighbors model and decided to test it out with different metrics, neighbors and algorithms \\
\subsubsection{Algorithms} \subsubsection{Algorithms}
We decided to test our model with 3 algorithms: We decided to test our model with 2 algorithms:
\begin{enumerate} \begin{enumerate}
\item Ball Tree
\item KD Tree
\item Brute \item Brute
\item Auto
\end{enumerate} \end{enumerate}
Ball Tree and KD Tree do not work on sparse input (as is the case with our input) so we decided to omit them
\subsubsection{Neighbor number} \subsubsection{Neighbor number}
We decided to test our model with 5 different neighbor amount: We decided to test our model with 5 different neighbor amount:
\begin{enumerate} \begin{enumerate}
\item 5 \item 5 - Popular starting point for small-medium datasets
\item square root of available data \item square root of available data - Usually helps to balance between underfitting and overfitting
\item half of available data \item half of available data - Usually usefull for checking overall trend than specific nuances
\item logarithm of available data \item logarithm of available data - Used for very large datasets
\item n-1 neighbors \item n-1 neighbors - Usually leads to overgeneralization as we use all instances excepct one for prediciton
\end{enumerate} \end{enumerate}
\subsubsection{Metrics}
For brute algorithm we tested it will all possible metrics:
\begin{enumerate}
\item Cityblock
\item Cosine
\item Euclidean
\item l1
\item l2
\item Manhattan
\end{enumerate}
\section{Intermediate results} \section{Intermediate results}
\subsection{Results} \subsection{Results}