Ideas for a rule based classifier after discussion with

@koaning: The hope with that idea is that you can define case_when like statements that can be used as a rule based system.

This has a few benefits.

1. It's simple to create for a domain person.
2. It's possible to create a ui/webapp for it.
3. You might even be able to generate SQL so that the ML system can also "be deployed" in a database.

This classifier would not have the full power of Python, but is rather a collection of rules entered by domain experts who are not necessarily technical people.

Rules

Rules have no structure and are always interpreted as disjunctions (or) and can be composed of conjunctions (and). To resolve conflict they can have a simple priority field.

Format of the rules could be

term:
   feature_name op value

op: '=', '<>', '<', '>', '<=', '>='

expr: term 
       | term 'and' term

rule : term '=>' prediction (prio)?

Examples

• age < 60 => low
• sex = 'f' and fare <> => high 10

Rules need not be expressed as plain text, but also a structured format of nested lists/arrays. A parser for a text format like this would be possible with a very simple recursive descent parser.

API

class ClassifierBase:
    def predict(self, X):
        return np.array([ self.predict_single(x) for x in X])
    def predict_proba(self, X):
        return np.array([probas[xi] for xi in self.predict(X)])
    def score(self, X, y):
        n = len(y)
        correct = 0
        predictions = self.predict(X)
        for prediction, ground_truth in zip(predictions, y):
            if prediction == ground_truth:
                correct = correct + 1
        return correct / n

class CaseWhenClassifier(ClassifierBase):
    def predict_single(self, x):
       ...

    def .from_sklearn_tree(self, tree):
       ...

    def .to_sklearn_tree(self):
       ...

    def to_python_code(self, code_style):
      ...

    def parse(self, rules_as_text):
      ...

rules = ...
rule_clf = CaseWhenClassifier(features, categories, rules)

Debugging support for plotting pairwise decision boundaries would be helpful.

Hi, I'm trying to draw model on jupyter by referring to this link but it doesn't aprear anything.

jupyter was run on ubuntu machine and accessed from another remote computer in the same subnet.

bokeh==2.4.3
human-learn==0.3.1
ipywidgets==7.7.1
jupyter==1.0.0
jupyter-client==7.3.4
jupyter-console==6.4.4
jupyter-core==4.11.1
jupyter-server==1.18.1
jupyterlab==3.4.4
jupyterlab-pygments==0.2.2
jupyterlab-server==2.15.0
jupyterlab-widgets==1.1.1

Hey there! Human learn has been super helpful so far. One thing I am a bit missing is the ability to see some of the underlying data about each data point. It would be very helpful to have a tooltip and having the option to pick a list of columns from the data frame to see in the tooltip.

Right now, I am using Plotly separately to do that which allows me to more easily explore clusters. Then I try to find this cluster and draw on it.

What do you think? Cheers, Nicolas

Hi there! I noticed that when the column used for the labels or the color in an InteractiveCharts contains more than 5 unique values, adding a chart throws an error because the number of available colors in _colors is too low.

# group_kind contains 7 unique values
clf = InteractiveCharts(dfs, labels=["spam", "not_spam"], color="group_kind")
clf.add_chart(x="umap_1", y="umap_2")

It throws the error:

KeyError                                  Traceback (most recent call last)
<ipython-input-108-2daa1de2581a> in <module>
----> 1 clf.add_chart(x="umap_1", y="umap_2")

~/anaconda3/envs/nlp_fb_posts_topics/lib/python3.8/site-packages/hulearn/experimental/interactive.py in add_chart(self, x, y, size, alpha, width, height, legend)
     84         ```
     85         """
---> 86         chart = SingleInteractiveChart(
     87             dataf=self.dataf.copy(),
     88             labels=self.labels,

~/anaconda3/envs/nlp_fb_posts_topics/lib/python3.8/site-packages/hulearn/experimental/interactive.py in __init__(self, dataf, labels, x, y, size, alpha, width, height, color, legend)
    160                 color_labels = list(dataf[self.color_column].unique())
    161                 d = {k: col for k, col in zip(color_labels, self._colors)}
--> 162                 dataf = dataf.assign(color=[d[lab] for lab in dataf[self.color_column]])
    163             self.source = ColumnDataSource(data=dataf)
    164             self.labels = labels

~/anaconda3/envs/nlp_fb_posts_topics/lib/python3.8/site-packages/hulearn/experimental/interactive.py in <listcomp>(.0)
    160                 color_labels = list(dataf[self.color_column].unique())
    161                 d = {k: col for k, col in zip(color_labels, self._colors)}
--> 162                 dataf = dataf.assign(color=[d[lab] for lab in dataf[self.color_column]])
    163             self.source = ColumnDataSource(data=dataf)
    164             self.labels = labels

KeyError: 'bulletin_board'

Maybe using a colormap instead of a fixed set of colors would fix the issue?

Hi, I'm trying the library just like I've seen on https://calmcode.io/human-learn/draw.html, but with my own data. This is what I got:

from hulearn.experimental.interactive import InteractiveCharts
clf = InteractiveCharts(df_labeled, labels="cluster")

BokehJS 2.2.1 successfully loaded

clf.add_chart(x='dst_ip',y='avg_duration')

The graph appears, data is colored as expected and I can interact with it (zoom and so), but I can't draw the areas.

I'm using Python 3.7.3, IPython 7.14.0 and Jupyter 5.7.8

I am trying basically to reproduce the PyData Berlin environment using human-learn with sentence embeddings and UMAP so that I can draw boundaries, explore, and quickly label text data.

The problem I am having is that the human-learn charts are not rendering in the VSC notebook. VSC is using Jupyter for the notebook and I am on Windows. I can render Pyplot, Seaborn, even Bokeh into the notebooks but the human-learn charts do not display:

Is anyone else having this issue? Is there some Jupyter extension I need or some Jupyter command I need to run? Bokeh is 2.3.2, human-learn is 0.3.1

I am trying to run the interactive classifier notebook downloaded from the link at the bottom of this page - https://koaning.github.io/human-learn/guide/drawing-classifier/drawing.html.

This is being run on a Windows x86-64 laptop, with the latest minconda3, python3.8 and jupyter-lab. I run into a traceback error on cell 3 from hulearn.experimental.interactive import InteractiveCharts, InteractiveChart

ImportError                               Traceback (most recent call last)
<ipython-input-3-9933ce75800d> in <module>()
----> 1 from hulearn.experimental.interactive import InteractiveCharts, InteractiveChart

ImportError: cannot import name 'InteractiveChart' from 'hulearn.experimental.interactive' (C:\<mypath>\miniconda3\envs\myenv\lib\site-packages\hulearn\experimental\interactive.py)

Not able to figure out what's up; issue reproduces on a unix environment (on Mac) as well.

Sometimes I hit this error:

ERROR:bokeh.server.views.ws:Refusing websocket connection from Origin 'http://localhost:8889';                       use --allow-websocket-origin=localhost:8889 or set BOKEH_ALLOW_WS_ORIGIN=localhost:8889 to permit this; currently we allow origins {'localhost:8888'}
WARNING:tornado.access:403 GET /ws (::1) 1.65ms

Would be nice to get an automated fix for this.

from hulearn.preprocessing import InteractivePreprocessor
tfm = InteractivePreprocessor(json_desc=charts.data())

df.pipe(tfm.pandas_pipe).loc[lambda d: d['group'] != 0].sample(10)

gives error :

---------------------------------------------------------------------------
FileNotFoundError                         Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_28956/1501149949.py in <module>
----> 1 from hulearn.preprocessing import InteractivePreprocessor
      2 tfm = InteractivePreprocessor(json_desc=charts.data())
      3 
      4 df.pipe(tfm.pandas_pipe).loc[lambda d: d['group'] != 0].sample(10)

~\AppData\Roaming\Python\Python39\site-packages\hulearn\preprocessing\__init__.py in <module>
      1 from hulearn.preprocessing.pipetransformer import PipeTransformer
----> 2 from hulearn.preprocessing.interactivepreprocessor import InteractivePreprocessor
      3 
      4 __all__ = ["PipeTransformer", "InteractivePreprocessor"]

~\AppData\Roaming\Python\Python39\site-packages\hulearn\preprocessing\interactivepreprocessor.py in <module>
      4 import numpy as np
      5 import pandas as pd
----> 6 from shapely.geometry import Point
      7 from shapely.geometry.polygon import Polygon
      8 

~\AppData\Roaming\Python\Python39\site-packages\shapely\geometry\__init__.py in <module>
      2 """
      3 
----> 4 from .base import CAP_STYLE, JOIN_STYLE
      5 from .geo import box, shape, asShape, mapping
      6 from .point import Point, asPoint

~\AppData\Roaming\Python\Python39\site-packages\shapely\geometry\base.py in <module>
     17 
     18 from shapely.affinity import affine_transform
---> 19 from shapely.coords import CoordinateSequence
     20 from shapely.errors import WKBReadingError, WKTReadingError
     21 from shapely.geos import WKBWriter, WKTWriter

~\AppData\Roaming\Python\Python39\site-packages\shapely\coords.py in <module>
      6 from ctypes import byref, c_double, c_uint
      7 
----> 8 from shapely.geos import lgeos
      9 from shapely.topology import Validating
     10 

~\AppData\Roaming\Python\Python39\site-packages\shapely\geos.py in <module>
    147     if os.getenv('CONDA_PREFIX', ''):
    148         # conda package.
--> 149         _lgeos = CDLL(os.path.join(sys.prefix, 'Library', 'bin', 'geos_c.dll'))
    150     else:
    151         try:

~\Anaconda3\envs\human-learn\lib\ctypes\__init__.py in __init__(self, name, mode, handle, use_errno, use_last_error, winmode)
    380 
    381         if handle is None:
--> 382             self._handle = _dlopen(self._name, mode)
    383         else:
    384             self._handle = handle

FileNotFoundError: Could not find module 'C:\Users\BORG7803\Anaconda3\envs\human-learn\Library\bin\geos_c.dll' (or one of its dependencies). Try using the full path with constructor syntax.

Hi Vincent,

I was particularly impressed by how we could classify the data by just drawing. Kudos to you.

However, I have been trying to implement the same in a different dataset but it's repeatedly throwing the below error .

I am also linking my notebook just in case : https://www.kaggle.com/nishantrock/notebook8935105440

Do suggest why this error is happening. I've tried it multiple times but it throws the same error.

AttributeError Traceback (most recent call last) in ----> 1 clf.add_chart(x = 'Health Indicator', y = 'Reco_Policy_Premium')

/opt/conda/lib/python3.7/site-packages/hulearn/experimental/interactive.py in add_chart(self, x, y, size, alpha, width, height, legend) 97 ) 98 self.charts.append(chart) ---> 99 chart.show() 100 101 def data(self):

/opt/conda/lib/python3.7/site-packages/hulearn/experimental/interactive.py in show(self) 199 200 def show(self): --> 201 show(self.app) 202 203 def _replace_xy(self, data):

/opt/conda/lib/python3.7/site-packages/bokeh/io/showing.py in show(obj, browser, new, notebook_handle, notebook_url, **kw) 135 # in Tornado) just in order to show a non-server object 136 if is_application or callable(obj): --> 137 return run_notebook_hook(state.notebook_type, 'app', obj, state, notebook_url, **kw) 138 139 return _show_with_state(obj, state, browser, new, notebook_handle=notebook_handle)

/opt/conda/lib/python3.7/site-packages/bokeh/io/notebook.py in run_notebook_hook(notebook_type, action, *args, **kw) 296 if _HOOKS[notebook_type][action] is None: 297 raise RuntimeError("notebook hook for %r did not install %r action" % notebook_type, action) --> 298 return _HOOKS[notebook_type][action](*args, **kw) 299 300 #-----------------------------------------------------------------------------

/opt/conda/lib/python3.7/site-packages/bokeh/io/notebook.py in show_app(app, state, notebook_url, port, **kw) 463 464 from tornado.ioloop import IOLoop --> 465 from ..server.server import Server 466 467 loop = IOLoop.current()

/opt/conda/lib/python3.7/site-packages/bokeh/server/server.py in 39 # External imports 40 from tornado import version as tornado_version ---> 41 from tornado.httpserver import HTTPServer 42 from tornado.ioloop import IOLoop 43

/opt/conda/lib/python3.7/site-packages/tornado/httpserver.py in 30 31 from tornado.escape import native_str ---> 32 from tornado.http1connection import HTTP1ServerConnection, HTTP1ConnectionParameters 33 from tornado import httputil 34 from tornado import iostream

/opt/conda/lib/python3.7/site-packages/tornado/http1connection.py in 32 from tornado import gen 33 from tornado import httputil ---> 34 from tornado import iostream 35 from tornado.log import gen_log, app_log 36 from tornado.util import GzipDecompressor

/opt/conda/lib/python3.7/site-packages/tornado/iostream.py in 208 209 --> 210 class BaseIOStream(object): 211 """A utility class to write to and read from a non-blocking file or socket. 212

/opt/conda/lib/python3.7/site-packages/tornado/iostream.py in BaseIOStream() 284 self._closed = False 285 --> 286 def fileno(self) -> Union[int, ioloop._Selectable]: 287 """Returns the file descriptor for this stream.""" 288 raise NotImplementedError()

AttributeError: module 'tornado.ioloop' has no attribute '_Selectable'

Currently, the library does not support adding custom title rather the x and y labels passed to the Interactive chart becomes the title

self.plot = figure(width=width, height=height, title=f"{x} vs. {y}")

as shown above we can add common accessors to deal with this?

Hi Vincent. I'm super into this framework. As a domain expert, I see some helpful ise cases with this tool involving regression. However, I'm not confident to apply regression as no example are provided.

I noticed on reviewing this PR that SingleInteractiveChart does not check if the inputs make sense with regards to the dataframe that is passed in. We don't want to create an extra SingleInteractiveChart under the InteractiveCharts object because this causes side effects (unneeded json data).

Let's add some ValueErrors there.