https://github.com/comet-ml/hn-predictor
Example Project for the Working Sessions for Standardizing the Experiment Series
https://github.com/comet-ml/hn-predictor
Last synced: about 1 year ago
JSON representation
Example Project for the Working Sessions for Standardizing the Experiment Series
- Host: GitHub
- URL: https://github.com/comet-ml/hn-predictor
- Owner: comet-ml
- Created: 2022-01-11T20:10:30.000Z (over 4 years ago)
- Default Branch: main
- Last Pushed: 2022-02-17T17:38:36.000Z (over 4 years ago)
- Last Synced: 2025-04-03T02:54:46.614Z (about 1 year ago)
- Language: Python
- Homepage:
- Size: 36.1 KB
- Stars: 0
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# hn-predictor
## Objective
The business objective that we are trying to achieve is to maximize the traffic we receive from our company’s post on Hacker News.
Our requirement is to create a model to predict the performance of a post on Hacker News. Our model would help us optimize our post to maximize the likelihood that it will trend.
## Technical Requirements
Here we define the technical assumptions we are making for this project. These assumptions affect how code for running experiments should be written.
### Artifacts Requirements
#### Dataset Type Artifacts
For this project, all dataset type Artifacts should contain a training and validation dataset saved as pickle files. Artifact Metadata should follow this format.
```
{
"filenames": {
"train": "",
"valid": "",
},
"columns": {"": "", ...},
}
```
The columns field should be a dictionary containing the name of the feature column as the key, and a description of the column as its value.
Our Dataset Artifact fetching functions will assume this schema and will not work otherwise.
### Experimentation Requirements
1. When running an experiment, pass in a message (similar to a git commit message) using the command line args in order to keep the project organized
2. When logging a prediction experiment, save the predictions as a `predictions.csv` and log them as an experiment asset
3. Log trained model assets using `experiment.log_model()` so that they can be added to the [Model Registry](https://www.comet.ml/site/using-comet-model-registry/).