mirror of
https://github.com/maxkratz/community-tutorials.git
synced 2024-09-16 11:28:12 +00:00
finalize mlflow tutorial
This commit is contained in:
parent
5e2eb5194d
commit
3625337a28
1 changed files with 10 additions and 3 deletions
|
@ -250,8 +250,10 @@ $ source ai_playground/bin/activate
|
|||
```
|
||||
|
||||
We are going to utilize `scikit-learn` to train a [random forest classifiers](https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html) of different sizes (i.e. nunbers of trees) on the [iris dataset](https://scikit-learn.org/stable/datasets/toy_dataset.html#iris-dataset) (a toy example for classification on flowers) to compare their accuracy to each other.
|
||||
We create an experiment called `RF Tree Nunbers`, iterate over sizes from 1 to 20 trees and log the numbers of trees and the accuracy of the random forest in a so called run (a substructure of an MLFlow experiment).
|
||||
Finally, the trained random forest will be also saved along with the parameter and accuracy, so you can download the model afterwards.
|
||||
|
||||
We are going to use the following script to do our first experiement (it is well commented):
|
||||
We are going to use the following script to do our first experiment (it is well commented):
|
||||
|
||||
```python
|
||||
from sklearn.datasets import load_iris
|
||||
|
@ -289,6 +291,10 @@ for n_est in range(1, 21): # Check forests from 1 to 20 trees
|
|||
mlflow.end_run()
|
||||
```
|
||||
|
||||
A detailed explanation on how random forests work is out of scope of this tutorial. In short words: a random forest is a bunch of decision trees.
|
||||
Each tree decides on its own, which class a sample belongs to.
|
||||
In the end there is a vote and the class voted the most will be chosen as result. More details can be read [here](https://en.m.wikipedia.org/wiki/Random_forest).
|
||||
|
||||
I saved it under `example.py` and executed it by calling `python example.py` within my local virtual environment.
|
||||
|
||||
After the script has finish its work, the webui looks a bit different:
|
||||
|
@ -299,9 +305,10 @@ Each run of an experiment is listed there und can be explored by clicking on the
|
|||
Additionally, you can also download the trained models and do comparisons.
|
||||
|
||||
# Conclusion
|
||||
SSH proxy can now be used. It is recommended to perform tests before productive use.
|
||||
You set up a central MLFlow instance to track your AI experiments for sutstainable and reproducable data science.
|
||||
Additionaly, you tracked your first machine learning experiments.
|
||||
|
||||
![images/The SOCKS proxy can handle multiple ports simultaneously](community-tutorials/setup-and-use-sshproxy/images/socks.png)
|
||||
Further steps would be to secure the access via TLS and improve scalability in the long run.
|
||||
|
||||
# Licence
|
||||
|
||||
|
|
Loading…
Reference in a new issue