chuangcaleb / music-theme-recognition Goto Github PK

—

Created via Raycast

• Export results in pickle
• Export results to json too?
• Primitive Prediction
• Script: convert to latex

Export results in csv

Let another script handle analysis

parallel_dimensions value is THERE

—

Created via Raycast

—

Created via Raycast

Write console output to local markdown file, tables formatted with tabulate

—

Created via Raycast

—

Created via Raycast

https://scikit-learn.org/stable/modules/generated/sklearn.inspection.permutation_importance.html

https://scikit-learn.org/stable/auto_examples/ensemble/plot_forest_importances.html

https://scikit-learn.org/stable/auto_examples/inspection/plot_permutation_importance.html#sphx-glr-auto-examples-inspection-plot-permutation-importance-py

—

Created via Raycast

—

Created via Raycast

Add progress indicators (which label out of total, which seed, etc.)

—

Created via Raycast

—

Created via Raycast

Theme Label Values

1. There will be enough samples of each value. Problem is only if there is meaningless overlap i.e. maturity overlaps with identity.
2. Labels are assigned to what the song represents/means/symbolises. Meaning: if the song talks a lot about the status quo/norm BUT it is about how tradition is a bad thing... then, the song's theme really represents change.
3. Each of these labels already intuitively have some musical characteristics tied to them.

Revision 3: Table format for Interim

—

Created via Raycast

Centralize all data, raw and processed, into one subdirectory space.

Also, need to correctly update scripts to follow the new file paths.

—

Created via Raycast

print missing file in the console

—

Created via Raycast

Feature Explanation

• Dissertation > Methodology > Machine Learning > Feature Selection

Main link: scikit-multilearn: Multi-Label Classification in Python

Model selection: scikit-multilearn: Model Selection

• Data Loading
• Data Preparation
• Sampling
• Building Model
• Prediction
• Evaluation
• Cross-Evaluation

—

Created via Raycast

Go through all the data and get usable samples

• Recognizable
• (has) Thematic (value)
• Safe lyrics
• Non-Duplicates

• Management in a large project - Overleaf, Online LaTeX Editor
• https://web.mit.edu/rsi/www/pdfs/new-latex.pdf

—

Created via Raycast

Get the best estimator score — and average?

—

Created via Raycast

[0 : 1] Normalization, because distribution is unknown

—

Created via Raycast

Instead of complete overwriting

—

Created via Raycast

Using some editable data file representation to store all the features

Perhaps in a dictionary json-like structure, so that we can easily pick out the list of features according to the category.

—

Created via Raycast

Refactor to keep the data type consistent between features data and label data

Just put the whole thing through jSymbolic—probably takes a few hours of processing.

—

Created via Raycast

• Import citations

FINAL YEAR (3RD+YEAR) PROJECT HANDBOOK - COMPUTER SCIENCE INTRANET - WORKSPACE

e.g. A Whole New World — trailing silence of ≈1 minute

Go through jSymbolic's features and find outliers and anomalies...

ROC AUC: https://towardsdatascience.com/feature-selection-in-scikit-learn-dc005dcf38b7

RFECV: https://medium.com/analytics-vidhya/feature-selection-using-scikit-learn-5b4362e0c19b

https://stackoverflow.com/questions/32210569/using-gridsearchcv-with-adaboost-and-decisiontreeclassifier

Multi-class AdaBoosted Decision Trees — scikit-learn 1.0.2 documentation
—

Created via Raycast

—

Created via Raycast

—

Created via Raycast

7th February 2022

• justifications for feature selection
• more justifications: 5.2: choosing MIDI
• kNN justification
• how to show commonly used?
  citations for the choice of algorithm (can be for other topics)
  most commonly used -> commonly used (just show a few appropriate examples)
• Single/double quote

Originally posted by @chuangcaleb in #7 (comment)

Getting familiar with jSymbolic

Also update the README accordingly

• Standardize em-dash spacing
• Standardize n-person language usage (we, our, us)
• Everything's defined first before usage (e.g. MIR, MGR, MER, etc.)
• Spell check
• Page break/orphans/widows
• Readme's requirement.txt with versioning
• Past tense
• Single/double quote

—

Created via Raycast

Count

• total recognised samples
• Percentage recognised / total
• total tentative samples
• total positives for a label
• percentage positives for a label

—

Created via Raycast

—

Created via Raycast

—

Created via Raycast