# without initial config
>>> dora = Dora()
>>> dora.configure(output = 'A', data = 'path/to/data.csv')

# is the same as
>>> import pandas as pd
>>> dataframe = pd.read_csv('path/to/data.csv')
>>> dora = Dora(output = 'A', data = dataframe)

>>> dora.data
   A   B  C      D  useless_feature
0  1   2  0   left                1
1  4 NaN  1  right                1
2  7   8  2   left                1

# read data with missing and poorly scaled values
>>> import pandas as pd
>>> df = pd.DataFrame([
...   [1, 2, 100],
...   [2, None, 200],
...   [1, 6, None]
... ])
>>> dora = Dora(output = 0, data = df)
>>> dora.data
   0   1    2
0  1   2  100
1  2 NaN  200
2  1   6  NaN

# impute the missing values (using the average of each column)
>>> dora.impute_missing_values()
>>> dora.data
   0  1    2
0  1  2  100
1  2  4  200
2  1  6  150

# scale the values of the input variables (center to mean and scale to unit variance)
>>> dora.scale_input_values()
>>> dora.data
   0         1         2
0  1 -1.224745 -1.224745
1  2  0.000000  1.224745
2  1  1.224745  0.000000

# feature selection / removing a feature
>>> dora.data
   A   B  C      D  useless_feature
0  1   2  0   left                1
1  4 NaN  1  right                1
2  7   8  2   left                1

>>> dora.remove_feature('useless_feature')
>>> dora.data
   A   B  C      D
0  1   2  0   left
1  4 NaN  1  right
2  7   8  2   left

# extract an ordinal feature through one-hot encoding
>>> dora.extract_ordinal_feature('D')
>>> dora.data
   A   B  C  D=left  D=right
0  1   2  0       1        0
1  4 NaN  1       0        1
2  7   8  2       1        0

# extract a transformation of another feature
>>> dora.extract_feature('C', 'twoC', lambda x: x * 2)
>>> dora.data
   A   B  C  D=left  D=right  twoC
0  1   2  0       1        0     0
1  4 NaN  1       0        1     2
2  7   8  2       1        0     4

# plot a single feature against the output variable
dora.plot_feature('column-name')

# render plots of each feature against the output variable
dora.explore()

# create random partition of training / validation data (~ 80/20 split)
dora.set_training_and_validation()

# train a model on the data
X = dora.training_data[dora.input_columns()]
y = dora.training_data[dora.output]

some_model.fit(X, y)

# validate the model
X = dora.validation_data[dora.input_columns()]
y = dora.validation_data[dora.output]

some_model.score(X, y)

# save a version of your data
>>> dora.data
   A   B  C      D  useless_feature
0  1   2  0   left                1
1  4 NaN  1  right                1
2  7   8  2   left                1
>>> dora.snapshot('initial_data')

# keep track of changes to data
>>> dora.remove_feature('useless_feature')
>>> dora.extract_ordinal_feature('D')
>>> dora.impute_missing_values()
>>> dora.scale_input_values()
>>> dora.data
   A         B         C    D=left   D=right
0  1 -1.224745 -1.224745  0.707107 -0.707107
1  4  0.000000  0.000000 -1.414214  1.414214
2  7  1.224745  1.224745  0.707107 -0.707107

>>> dora.logs
["self.remove_feature('useless_feature')", "self.extract_ordinal_feature('D')", 'self.impute_missing_values()', 'self.scale_input_values()']

# use a previous version of the data
>>> dora.snapshot('transform1')
>>> dora.use_snapshot('initial_data')
>>> dora.data
   A   B  C      D  useless_feature
0  1   2  0   left                1
1  4 NaN  1  right                1
2  7   8  2   left                1
>>> dora.logs
[]

# switch back to your transformation
>>> dora.use_snapshot('transform1')
>>> dora.data
   A         B         C    D=left   D=right
0  1 -1.224745 -1.224745  0.707107 -0.707107
1  4  0.000000  0.000000 -1.414214  1.414214
2  7  1.224745  1.224745  0.707107 -0.707107
>>> dora.logs
["self.remove_feature('useless_feature')", "self.extract_ordinal_feature('D')", 'self.impute_missing_values()', 'self.scale_input_values()']