| --- |
| library_name: keras |
| pipeline_tag: text-classification |
| tags: |
| - security |
| - password |
| --- |
| |
| # Password security classifier |
| This is a keras model that gives a binary response showing how secure is a password. |
| I used this password list as a dataset + random password generation using the `random` library (I am aware of it being unsecure). |
| This model has a [huggingface space](https://huggingface.co/spaces/MoyAI/password-security). You can visit the link to try using the model online. |
| ## Model & Training |
| The model was trained on 4,2MiB (`200 000` lines) of .csv data for 2 epochs on Adam with learning rate 0.00001, batch size 4 and mse loss. |
| The model embeds every input character with the ord() builtin python function. The model has `128 969` dense layer parameters. |
| ## Evaluation |
| During training the model had: |
| **loss** - 0.0025 |
| **accuracy** - 0.9972 |
| The test metrics are: |
| **loss** - 0.0023 |
| **accuracy** - 0.9972 |
| ## Model usage |
| The `start.py` file has a `clf` function that inputs a string of a password and responds with a 0-1 float value. 1 means secure and 0 insecure. |
| To train the model, create a `dataset.csv` file. Here's an example: |
| ```csv |
| 0,qwerty |
| 0,123456 |
| 1,ISOdvsjs8r8 |
| 1,F(SEsDLxc__ |
| ``` |
| After the `dataset.csv` file is created, now you can adjust the settings in the `net.py` file and run it. |
