bump libs (9f033f66) · Commits · binderhub / beginning-supervised-machine-learning-with-python

class/Supervised ML.ipynb

+121 −1761

File changed.

Preview size limit exceeded, changes collapsed.

requirements.txt

+11 −11

Changes for requirements.txt: 11 added lines, 11 removed lines.
Original line number	Diff line number	Diff line
		shap==0.35.0
		matplotlib==3.2.1
		notebook==6.0.3
		numpy==1.18.4
		pandas==1.0.3
		matplotlib==3.3.3
		notebook==6.1.5
		numpy==1.19.4
		pandas==1.1.4
		scikit-learn==0.23.2
		scipy==1.5.4
		seaborn==0.11.0
		shap==0.37.0
		xgbfir==0.3.1
		xgboost==1.0.2
		xgboost==1.2.1
		xlrd==1.2.0
		XlsxWriter==1.2.8
		yellowbrick==1.1
		scikit-learn==0.22.2.post1
		scipy==1.4.1
		seaborn==0.10.1
		XlsxWriter==1.3.7
		yellowbrick==1.2

solutions/Supervised ML.ipynb

+321 −869

File changed.

Preview size limit exceeded, changes collapsed.

solutions/Supervised ML.py

+39 −27

Changes for solutions/Supervised ML.py: 39 added lines, 27 removed lines.
Original line number	Diff line number	Diff line
		@@ -6,7 +6,7 @@
		# extension: .py
		# format_name: light
		# format_version: '1.5'
		# jupytext_version: 1.4.2
		# jupytext_version: 1.7.1
		# kernelspec:
		# display_name: Python 3
		# language: python
		@@ -37,22 +37,25 @@ from yellowbrick import regressor

		%%time
		# Resampling data from minute interval to day
		bit_df = pd.read_csv('../data/coinbaseUSD_1-min_data_2014-12-01_to_2018-01-08.csv')
		bit_df = (pd.read_csv('../data/coinbaseUSD_1-min_data_2014-12-01_to_2018-01-08.csv')
		# Convert unix time to datetime
		bit_df['date'] = pd.to_datetime(bit_df.Timestamp, unit='s')
		.assign(date=lambda df_:pd.to_datetime(df_.Timestamp, unit='s'))
		# Reset index
		bit_df = bit_df.set_index('date')
		.set_index('date')
		# Rename columns so easier to code
		bit_df = bit_df.rename(columns={'Open':'open', 'High': 'hi', 'Low': 'lo',
		.rename(columns={'Open':'open', 'High': 'hi', 'Low': 'lo',
		'Close': 'close', 'Volume_(BTC)': 'vol_btc',
		'Volume_(Currency)': 'vol_cur',
		'Weighted_Price': 'wp', 'Timestamp': 'ts'})
		# Resample and only use recent samples that aren't missing
		bit_df = bit_df.resample('d').agg({'open': 'first', 'hi': 'max',
		.resample('d')
		.agg({'open': 'first', 'hi': 'max',
		'lo': 'min', 'close': 'last', 'vol_btc': 'sum',
		'vol_cur': 'sum', 'wp': 'mean', 'ts': 'min'}).iloc[-1000:]
		'vol_cur': 'sum', 'wp': 'mean', 'ts': 'min'})
		# drop last row as it is not complete
		bit_df = bit_df.iloc[:-1]
		.iloc[-1000:-1]
		)


		bit_df

		@@ -99,10 +102,9 @@ ax.plot(X, X*lr_model.coef_ + lr_model.intercept_+ 100, linestyle='--', color='g
		# regression tries to minimize the square of the distance.
		metrics.mean_squared_error(y, pred)

		# Vertical distance between line and point is the error. Ordinary Least Squares
		# regression tries to minimize the square of the distance.
		metrics.mean_squared_error(y, pred)
		metrics.mean_absolute_error(y, pred)

		lr_model.score(X, y)


		# ### Lab Data
		@@ -258,25 +260,28 @@ from sklearn import ensemble, model_selection, preprocessing, tree
		from yellowbrick import classifier


		# %%time
		%%time
		# Resampling data from minute interval to day
		bit_df = pd.read_csv('../data/coinbaseUSD_1-min_data_2014-12-01_to_2018-01-08.csv')
		bit_df = (pd.read_csv('../data/coinbaseUSD_1-min_data_2014-12-01_to_2018-01-08.csv')
		# Convert unix time to datetime
		bit_df['date'] = pd.to_datetime(bit_df.Timestamp, unit='s')
		.assign(date=lambda df_:pd.to_datetime(df_.Timestamp, unit='s'))
		# Reset index
		bit_df = bit_df.set_index('date')
		.set_index('date')
		# Rename columns so easier to code
		bit_df = bit_df.rename(columns={'Open':'open', 'High': 'hi', 'Low': 'lo',
		.rename(columns={'Open':'open', 'High': 'hi', 'Low': 'lo',
		'Close': 'close', 'Volume_(BTC)': 'vol_btc',
		'Volume_(Currency)': 'vol_cur',
		'Weighted_Price': 'wp', 'Timestamp': 'ts'})
		# Resample and only use recent samples that aren't missing
		bit_df = bit_df.resample('d').agg({'open': 'first', 'hi': 'max',
		.resample('d')
		.agg({'open': 'first', 'hi': 'max',
		'lo': 'min', 'close': 'last', 'vol_btc': 'sum',
		'vol_cur': 'sum', 'wp': 'mean', 'ts': 'min'}).iloc[-1000:]
		bit_df['buy'] = (bit_df.close.shift(-1) > bit_df.close).astype(int)
		'vol_cur': 'sum', 'wp': 'mean', 'ts': 'min'})
		# drop last row as it is not complete
		bit_df = bit_df.iloc[:-1]
		.assign(buy=lambda df_:(df_.close.shift(-1) > df_.close).astype(int))
		.iloc[-1000:-1]
		)


		bit_df

		@@ -292,6 +297,9 @@ bit_df.describe()
		# * To predict values of y (y hat), call `.predict(X)`
		# * To get the accuracy call `.score(X, y)`

		# Calculating is NP hard problem (hence randomness)
		# Uses GINI Impurity by default
		# https://en.wikipedia.org/wiki/Decision_tree_learning#Gini_impurity
		ignore = {'buy'}
		cols = [c for c in bit_df.columns if c not in ignore]
		X = bit_df[cols]
		@@ -317,9 +325,10 @@ tree.export_graphviz(dt_model, out_file='/tmp/tree1.dot',
		filled=True
		)

		%%bash
		# This doesn't run on Windows. Also requires that you have graphviz installed (not a Python module)
		dot -Tpng -otree1.png /tmp/tree1.dot
		# + language="bash"
		# # This doesn't run on Windows. Also requires that you have graphviz installed (not a Python module)
		# dot -Tpng -otree1.png /tmp/tree1.dot
		# -

		# <img src="tree1.png">

		@@ -401,6 +410,13 @@ classifier.confusion_matrix(dt_model, X_train, y_train, X_test, y_test,
		classifier.roc_auc(dt_model, X_train, y_train, X_test, y_test,
		classes=['sell', 'buy'], micro=False, macro=False)

		# Using more data to create a more normal plot
		# aspect fixing
		fig, ax = plt.subplots(figsize=(4,4))
		classifier.roc_auc(dt_model, X_train, y_train, X, y,
		classes=['sell', 'buy'], micro=False, macro=False,
		ax=ax)

		classifier.precision_recall_curve(dt_model, X_train, y_train, X_test, y_test,
		classes=['sell', 'buy'], micro=False, macro=False)

		@@ -430,7 +446,3 @@ classifier.roc_auc(mush_dt, mush_X_train, mush_y_train, mush_X_test, mush_y_test