DS & AI Cheat Sheet

Pandas

DataFrame manipulation, cleaning, aggregation, and I/O

I/O — Reading & Writing

Function	Description	Example
pd.read_csv()	Read CSV file into DataFrame	df = pd.read_csv('data.csv')
pd.read_excel()	Read Excel file into DataFrame	df = pd.read_excel('data.xlsx')
pd.read_json()	Read JSON into DataFrame	df = pd.read_json('data.json')
pd.read_sql()	Read SQL query into DataFrame	df = pd.read_sql(query, conn)
pd.read_parquet()	Read Parquet file	df = pd.read_parquet('data.parquet')
df.to_csv()	Write DataFrame to CSV	df.to_csv('out.csv', index=False)
df.to_excel()	Write DataFrame to Excel	df.to_excel('out.xlsx')
df.to_parquet()	Write DataFrame to Parquet	df.to_parquet('out.parquet')
df.to_dict()	Convert DataFrame to dict	d = df.to_dict(orient='records')

Inspection & Exploration

Function	Description	Example
df.head(n)	First n rows (default 5)	df.head(10)
df.tail(n)	Last n rows	df.tail(5)
df.info()	Column types, nulls, memory	df.info()
df.describe()	Summary stats (count, mean…)	df.describe(include='all')
df.shape	Tuple of (rows, columns)	rows, cols = df.shape
df.dtypes	Data type of each column	df.dtypes
df.columns	List of column names	list(df.columns)
df.nunique()	Count of unique values per col	df.nunique()
df.value_counts()	Frequency count	df['city'].value_counts(normalize=True)
df.memory_usage()	Memory used per column	df.memory_usage(deep=True)
df.sample(n)	Random sample of n rows	df.sample(5, random_state=42)

Selection & Filtering

Function	Description	Example
df['col']	Select single column (Series)	df['age']
df[['a','b']]	Select multiple columns	df[['name', 'age']]
df.loc[row, col]	Label-based indexing	df.loc[0:5, 'age']
df.iloc[row, col]	Integer-based indexing	df.iloc[0:5, 1:3]
df.at[row, col]	Single value by label	df.at[0, 'name']
df.iat[row, col]	Single value by position	df.iat[0, 1]
df.query(expr)	Filter using string expression	df.query('age > 30 & city=="NY"')
df[df.col > val]	Boolean mask filter	df[df['age'] > 30]
df.where(cond)	Keep values where True, else NaN	df.where(df > 0)
df.filter()	Filter by column/row labels	df.filter(like='price')

Cleaning & Transformation

Function	Description	Example
df.isnull()	Boolean mask of null values	df.isnull().sum()
df.dropna()	Drop rows with any null	df.dropna(subset=['age'])
df.fillna(val)	Fill nulls with value/method	df.fillna({'age': 0, 'name': 'N/A'})
df.interpolate()	Interpolate missing numeric vals	df['price'].interpolate()
df.duplicated()	Boolean mask of duplicate rows	df.duplicated(subset=['id'])
df.drop_duplicates()	Remove duplicate rows	df.drop_duplicates(keep='first')
df.rename()	Rename columns/index	df.rename(columns={'old':'new'})
df.astype()	Cast column to new type	df['age'].astype(int)
df.replace()	Replace values	df.replace({'Y':1, 'N':0})
df.clip()	Clip values to min/max	df['score'].clip(0, 100)
df.drop()	Drop rows or columns	df.drop(columns=['col1','col2'])
df.reset_index()	Reset row index	df.reset_index(drop=True)
df.set_index()	Set a column as index	df.set_index('id')

Aggregation, Merging & Reshaping

Function	Description	Example
df.sort_values()	Sort by column values	df.sort_values('age', ascending=False)
df.groupby()	Group rows & aggregate	df.groupby('city')['salary'].mean()
df.agg()	Apply multiple aggregations	df.groupby('dept').agg({'sal':['mean','max']})
df.pivot_table()	Create a pivot table	pd.pivot_table(df, values='sal', index='dept')
df.melt()	Wide to long format	df.melt(id_vars=['id'], value_vars=['q1','q2'])
df.pivot()	Long to wide format	df.pivot(index='date', columns='item', values='val')
df.apply()	Apply function along axis	df['bmi'].apply(lambda x: 'high' if x>25 else 'ok')
df.map()	Map values via dict/function	df['grade'].map({'A':4,'B':3,'C':2})
df.transform()	Transform keeping shape	df.groupby('dept')['sal'].transform('mean')
df.merge()	SQL-style join two DataFrames	pd.merge(df1, df2, on='id', how='left')
pd.concat()	Stack DataFrames row/colwise	pd.concat([df1, df2], ignore_index=True)
df.assign()	Add/overwrite columns fluently	df.assign(full_name=df.first+' '+df.last)
df.eval()	Evaluate expression string	df.eval('profit = revenue - cost')
df.explode()	Explode list column to rows	df.explode('tags')
df.value_counts()	Frequency count	df['city'].value_counts(normalize=True)
pd.cut()	Bin continuous data	pd.cut(df['age'], bins=[0,18,65,100])
pd.get_dummies()	One-hot encode categoricals	pd.get_dummies(df['color'])
df.crosstab()	Cross-tabulation of two cols	pd.crosstab(df['dept'], df['gender'])

String Methods (df.str.*)

Function	Description	Example
df.str.lower()	Lowercase all strings	df['name'].str.lower()
df.str.upper()	Uppercase all strings	df['name'].str.upper()
df.str.strip()	Strip whitespace	df['name'].str.strip()
df.str.contains()	Check if string contains pattern	df['email'].str.contains('@gmail')
df.str.replace()	Replace substring	df['text'].str.replace('old','new')
df.str.split()	Split string to list	df['full_name'].str.split(' ', expand=True)
df.str.len()	Length of each string	df['name'].str.len()
df.str.extract()	Extract regex groups	df['phone'].str.extract(r'(\d{3})')
df.str.startswith()	Check prefix	df['col'].str.startswith('pre')

DateTime Methods (df.dt.*)

Function	Description	Example
pd.to_datetime()	Parse string to datetime	pd.to_datetime(df['date'])
dt.year / .month / .day	Extract date components	df['date'].dt.year
dt.dayofweek	Day of week (0=Mon)	df['date'].dt.dayofweek
dt.hour / .minute	Extract time components	df['ts'].dt.hour
df.resample()	Resample time series	df.resample('M').mean()
df.shift()	Shift values by periods	df['price'].shift(1)
df.diff()	Difference between rows	df['price'].diff()
df.rolling(n)	Rolling window calculation	df['price'].rolling(7).mean()
df.expanding()	Expanding window	df['price'].expanding().mean()
df.ewm()	Exponentially weighted mean	df['price'].ewm(span=7).mean()

NumPy

Array creation, manipulation, math, and linear algebra

Array Creation

Function	Description	Example
np.array()	Create array from list/tuple	np.array([[1,2],[3,4]])
np.zeros(shape)	Array of zeros	np.zeros((3, 4))
np.ones(shape)	Array of ones	np.ones((3, 4), dtype=float)
np.full(shape, val)	Fill array with value	np.full((2,3), 7)
np.eye(n)	Identity matrix	np.eye(4)
np.arange()	Evenly spaced int range	np.arange(0, 10, 2)
np.linspace()	Evenly spaced floats	np.linspace(0, 1, 100)
np.random.rand()	Uniform random [0, 1)	np.random.rand(3, 4)
np.random.randn()	Standard normal random	np.random.randn(100)
np.random.randint()	Random integers	np.random.randint(0, 10, size=(3,3))
np.random.seed()	Set random seed	np.random.seed(42)

Shape & Manipulation

Function	Description	Example
arr.reshape()	Reshape without copying	arr.reshape(20, 20)
arr.ravel()	Flatten to 1D (view)	arr.ravel()
arr.flatten()	Flatten to 1D (copy)	arr.flatten()
arr.T / arr.transpose()	Transpose axes	arr.T
np.expand_dims()	Add new axis	np.expand_dims(arr, axis=0)
np.squeeze()	Remove size-1 axes	np.squeeze(arr)
np.concatenate()	Join arrays along axis	np.concatenate([a, b], axis=0)
np.stack()	Stack along new axis	np.stack([a, b], axis=1)
np.hstack()	Stack horizontally	np.hstack([a, b])
np.vstack()	Stack vertically	np.vstack([a, b])
np.split()	Split array	np.split(arr, 3)

Math & Statistics

Function	Description	Example
np.sum()	Sum all or along axis	np.sum(arr, axis=0)
np.mean()	Mean	np.mean(arr, axis=1)
np.median()	Median	np.median(arr)
np.std()	Standard deviation	np.std(arr, ddof=1)
np.min() / np.max()	Min / Max value	arr.min(axis=0)
np.argmin() / argmax()	Index of min / max	np.argmin(arr)
np.cumsum()	Cumulative sum	np.cumsum(arr)
np.abs()	Absolute value	np.abs(arr)
np.sqrt()	Square root	np.sqrt(arr)
np.exp()	Exponential (e^x)	np.exp(arr)
np.log()	Natural log	np.log(arr)
np.dot()	Dot product / matrix mul	np.dot(A, B)
np.matmul() / @	Matrix multiplication	A @ B
np.where()	Conditional select	np.where(arr > 0, arr, 0)
np.unique()	Unique values + counts	np.unique(arr, return_counts=True)
np.sort()	Sort array	np.sort(arr, axis=0)
np.percentile()	Percentile value	np.percentile(arr, 75)
np.corrcoef()	Correlation matrix	np.corrcoef(x, y)

Linear Algebra

Function	Description	Example
np.linalg.inv()	Matrix inverse	np.linalg.inv(A)
np.linalg.det()	Determinant	np.linalg.det(A)
np.linalg.eig()	Eigenvalues & eigenvectors	vals, vecs = np.linalg.eig(A)
np.linalg.svd()	Singular Value Decomposition	U, S, V = np.linalg.svd(A)
np.linalg.norm()	Vector / matrix norm	np.linalg.norm(arr)
np.linalg.solve()	Solve linear system Ax=b	x = np.linalg.solve(A, b)
np.linalg.pinv()	Pseudo-inverse	np.linalg.pinv(A)

SQL

Querying, aggregation, joins, window functions, and CTEs

Basic Querying

Syntax	Description	Example
SELECT col1, col2	Choose columns to return	SELECT name, age FROM users
SELECT DISTINCT	Remove duplicates	SELECT DISTINCT city FROM customers
WHERE condition	Filter rows	WHERE salary > 50000 AND dept = 'Eng'
IN (list)	Match any value in list	WHERE country IN ('US','UK','CA')
BETWEEN a AND b	Range filter (inclusive)	WHERE date BETWEEN '2024-01-01' AND '2024-12-31'
LIKE pattern	Pattern matching	WHERE email LIKE '%@gmail.com'
IS NULL / IS NOT NULL	Null checks	WHERE phone IS NOT NULL
ORDER BY col	Sort results	ORDER BY salary DESC, name ASC
LIMIT n / OFFSET n	Limit + paginate results	LIMIT 10 OFFSET 20
AS alias	Rename column/table	SELECT salary * 1.1 AS adjusted_salary

Aggregation & Grouping

Syntax	Description	Example
COUNT(*)	Count all rows	SELECT COUNT(*) FROM orders
SUM(col)	Sum of column	SELECT SUM(amount) FROM payments
AVG(col)	Average of column	SELECT AVG(score) FROM results
MIN(col) / MAX(col)	Min / Max value	SELECT MIN(price), MAX(price) FROM products
GROUP BY col	Group rows for aggregation	SELECT dept, AVG(salary) FROM emp GROUP BY dept
HAVING condition	Filter groups (after GROUP BY)	HAVING COUNT(*) > 5 AND AVG(sal) > 60000

Joins & Set Operations

Syntax	Description	Example
INNER JOIN	Rows matching in both tables	FROM a INNER JOIN b ON a.id = b.id
LEFT JOIN	All left + matched right	FROM a LEFT JOIN b ON a.id = b.id
RIGHT JOIN	All right + matched left	FROM a RIGHT JOIN b ON a.id = b.id
FULL OUTER JOIN	All rows from both tables	FROM a FULL OUTER JOIN b ON a.id = b.id
SELF JOIN	Join table with itself	FROM emp e1 JOIN emp e2 ON e1.mgr_id = e2.id
UNION	Combine + deduplicate results	SELECT city FROM a UNION SELECT city FROM b
UNION ALL	Combine keeping duplicates	SELECT id FROM a UNION ALL SELECT id FROM b
INTERSECT	Rows in both result sets	SELECT id FROM a INTERSECT SELECT id FROM b
EXCEPT / MINUS	Rows in first but not second	SELECT id FROM a EXCEPT SELECT id FROM b

Window Functions

Function	Description	Example
ROW_NUMBER()	Sequential row number	ROW_NUMBER() OVER (PARTITION BY dept ORDER BY sal DESC)
RANK()	Rank with gaps for ties	RANK() OVER (ORDER BY score DESC)
DENSE_RANK()	Rank without gaps	DENSE_RANK() OVER (PARTITION BY dept ORDER BY sal)
NTILE(n)	Divide into n equal buckets	NTILE(4) OVER (ORDER BY salary)
LAG(col, n)	Value from n rows before	LAG(price, 1) OVER (ORDER BY date)
LEAD(col, n)	Value from n rows ahead	LEAD(price, 1) OVER (ORDER BY date)
FIRST_VALUE()	First value in window	FIRST_VALUE(price) OVER (PARTITION BY id ORDER BY date)
SUM() OVER ()	Running total	SUM(amount) OVER (ORDER BY date)
AVG() OVER ()	Moving average	AVG(price) OVER (ORDER BY date ROWS 6 PRECEDING)
PARTITION BY	Define window partitions	OVER (PARTITION BY dept ORDER BY salary)

Advanced — CTEs, Subqueries & Functions

Syntax	Description	Example
WITH cte AS (...)	Common Table Expression	WITH top AS (SELECT * FROM emp ORDER BY sal DESC LIMIT 10)
CASE WHEN	Conditional logic	CASE WHEN sal>100k THEN 'high' ELSE 'low' END
COALESCE(a, b)	First non-null value	COALESCE(phone, email, 'N/A')
NULLIF(a, b)	Return null if a equals b	NULLIF(score, 0)
CAST(col AS type)	Type conversion	CAST(price AS DECIMAL(10,2))
SUBSTRING()	Extract part of string	SUBSTRING(name, 1, 3)
CONCAT()	Join strings	CONCAT(first_name, ' ', last_name)
DATE_TRUNC()	Truncate date to unit	DATE_TRUNC('month', created_at)
EXTRACT()	Extract date part	EXTRACT(YEAR FROM order_date)
EXISTS	Check if subquery has rows	WHERE EXISTS (SELECT 1 FROM orders WHERE user_id=u.id)
EXPLAIN	Show query execution plan	EXPLAIN SELECT * FROM large_table WHERE id > 100

Excel

Lookup, math, text, logic, date functions & dynamic arrays

Lookup & Reference

Function	Description	Example
VLOOKUP(val,range,col,0)	Vertical exact match lookup	=VLOOKUP(A2,Sheet2!A:C,2,0)
XLOOKUP(val,arr,ret)	Modern flexible lookup	=XLOOKUP(A2,B:B,C:C,"Not Found")
INDEX(range,row,col)	Return value at position	=INDEX(A1:D10,3,2)
MATCH(val,range,0)	Position of value in range	=MATCH("Bob",A:A,0)
INDEX+MATCH	Two-way flexible lookup	=INDEX(C:C,MATCH(A2,B:B,0))
HLOOKUP(val,range,row)	Horizontal lookup	=HLOOKUP("Q1",A1:D5,3,0)
OFFSET(ref,rows,cols)	Dynamic cell reference	=OFFSET(A1,3,2)
INDIRECT(ref_text)	Reference from text string	=INDIRECT("Sheet2!A"&B1)

Math & Statistical

Function	Description	Example
SUM(range)	Sum values	=SUM(A1:A100)
SUMIF(range,crit,sum)	Sum with condition	=SUMIF(B:B,"NY",C:C)
SUMIFS()	Sum with multiple conditions	=SUMIFS(D:D,B:B,"NY",C:C,">1000")
AVERAGEIF()	Average with condition	=AVERAGEIF(A:A,">18",B:B)
COUNTIF(range,crit)	Count with condition	=COUNTIF(B:B,"NY")
COUNTIFS()	Count with multiple conditions	=COUNTIFS(A:A,">18",B:B,"NY")
LARGE(range, k)	k-th largest value	=LARGE(A1:A100,3)
RANK(val,range)	Rank of value in range	=RANK(A2,A:A,0)
ROUND(val, digits)	Round to decimal places	=ROUND(3.14159, 2)
MOD(num, div)	Remainder after division	=MOD(10,3)

Text Manipulation

Function	Description	Example
TEXTJOIN(del,T,range)	Join with delimiter	=TEXTJOIN(", ",TRUE,A1:A5)
LEFT(text, n)	Extract left n chars	=LEFT(A2,3)
RIGHT(text, n)	Extract right n chars	=RIGHT(A2,4)
MID(text,start,n)	Extract from middle	=MID(A2,5,3)
LEN(text)	String length	=LEN(A2)
TRIM(text)	Remove extra spaces	=TRIM(A2)
PROPER(text)	Title Case	=PROPER("john doe")
SUBSTITUTE(t,old,new)	Replace substring	=SUBSTITUTE(A2,"-","/")
TEXT(val,format)	Format number as text	=TEXT(A2,"$#,##0.00")
FIND(sub,text)	Position of substring	=FIND("@",A2)

Logic & Conditionals

Function	Description	Example
IF(test,true,false)	Conditional value	=IF(A2>100,"High","Low")
IFS(t1,v1,t2,v2)	Multiple conditions	=IFS(A2>90,"A",A2>80,"B",TRUE,"C")
AND(cond1,cond2)	All conditions true	=IF(AND(A2>0,B2>0),"Both","No")
OR(cond1,cond2)	Any condition true	=IF(OR(A2="NY",A2="LA"),"Coast","Mid")
IFERROR(val,err)	Handle formula errors	=IFERROR(A2/B2,0)
IFNA(val,na_val)	Handle #N/A errors	=IFNA(VLOOKUP(A2,B:C,2,0),"Not found")
SWITCH(expr,v1,r1)	Switch statement	=SWITCH(A2,1,"Jan",2,"Feb","Other")

Dynamic Array Functions (Excel 365)

Function	Description	Example
FILTER(arr,cond)	Dynamic filtered range	=FILTER(A:B,C:C="NY")
SORT(range,col,ord)	Dynamic sorted range	=SORT(A1:C10,2,-1)
SORTBY(range,by)	Sort by another range	=SORTBY(A:B,C:C,-1)
UNIQUE(range)	Unique values list	=UNIQUE(A2:A100)
SEQUENCE(r,c,start)	Generate number sequence	=SEQUENCE(12,1,1,1)
VSTACK(r1,r2)	Stack ranges vertically	=VSTACK(A1:C5,Sheet2!A1:C3)
BYROW(range,fn)	Apply LAMBDA to each row	=BYROW(A1:C3,LAMBDA(r,SUM(r)))
LAMBDA(x,formula)	Create custom function	=LAMBDA(x,x*x)(5)

Matplotlib

Chart types, subplots, labels, annotations and styling

Chart Types

Function	Description	Example
plt.plot(x, y)	Line chart	plt.plot(x, y, color='blue', linewidth=2)
plt.scatter(x, y)	Scatter plot	plt.scatter(x, y, c=colors, s=sizes, alpha=0.7)
plt.bar(x, height)	Vertical bar chart	plt.bar(cats, vals, color='steelblue')
plt.barh(y, width)	Horizontal bar chart	plt.barh(labels, values)
plt.hist(x, bins)	Histogram	plt.hist(data, bins=30, density=True, alpha=0.7)
plt.pie(sizes)	Pie chart	plt.pie(sizes, labels=lbls, autopct='%1.1f%%')
plt.boxplot(data)	Box & whisker plot	plt.boxplot([grp1, grp2], labels=['A','B'])
plt.violinplot(data)	Violin plot	plt.violinplot(data, showmeans=True)
plt.fill_between(x,y1,y2)	Shade area between lines	plt.fill_between(x, y1, y2, alpha=0.3)
plt.errorbar(x,y,yerr)	Error bar plot	plt.errorbar(x, y, yerr=err, fmt='o')
plt.stackplot(x, y1,y2)	Stacked area chart	plt.stackplot(x, y1, y2, labels=['A','B'])
plt.imshow(data)	Display 2D array as image	plt.imshow(matrix, cmap='viridis')
plt.contourf()	Filled contour plot	plt.contourf(X, Y, Z, cmap='RdBu')
plt.hexbin(x,y)	Hexagonal bin plot	plt.hexbin(x, y, gridsize=30, cmap='Blues')

Figures, Subplots & Layout

Function	Description	Example
plt.figure(figsize)	Create figure with size	fig = plt.figure(figsize=(10, 6))
plt.subplots(r,c)	Grid of subplots	fig, axes = plt.subplots(2, 3, figsize=(12,8))
ax.twinx()	Second y-axis on same plot	ax2 = ax.twinx()
plt.tight_layout()	Auto-fix layout overlap	plt.tight_layout(pad=2.0)
plt.GridSpec()	Fine-grained grid control	gs = GridSpec(2, 2, width_ratios=[1,2])

Labels, Annotations & Styling

Function	Description	Example
plt.xlabel(text)	X-axis label	plt.xlabel('Time (s)', fontsize=12)
plt.ylabel(text)	Y-axis label	plt.ylabel('Price ($)')
plt.title(text)	Plot title	plt.title('Sales Over Time', fontsize=14, fontweight='bold')
plt.legend()	Show legend	plt.legend(loc='upper right', framealpha=0.9)
plt.xlim() / ylim()	Set axis limits	plt.xlim(0, 100); plt.ylim(-1, 1)
plt.xticks() / yticks()	Custom tick marks	plt.xticks(range(0,100,10), rotation=45)
plt.grid()	Add grid lines	plt.grid(True, linestyle='--', alpha=0.5)
plt.axhline(y)	Horizontal line	plt.axhline(y=0, color='red', linestyle='--')
plt.annotate(text)	Annotate with arrow	plt.annotate('Peak', xy=(5,10), xytext=(7,12), arrowprops={})
plt.savefig(fname)	Save to file	plt.savefig('chart.png', dpi=150, bbox_inches='tight')
plt.style.use()	Apply a style theme	plt.style.use('seaborn-v0_8-darkgrid')
plt.rcParams[]	Set global defaults	plt.rcParams['font.size'] = 12

Scikit-Learn

Preprocessing, models, evaluation, pipelines and tuning

Preprocessing & Encoding

Function	Description	Example
train_test_split()	Split into train/test sets	X_tr,X_te,y_tr,y_te = train_test_split(X,y,test_size=0.2)
StandardScaler()	Standardize (zero mean, unit var)	scaler = StandardScaler(); X_sc = scaler.fit_transform(X)
MinMaxScaler()	Scale to [0, 1]	MinMaxScaler().fit_transform(X)
RobustScaler()	Scale robust to outliers	RobustScaler().fit_transform(X)
LabelEncoder()	Encode target labels as int	le = LabelEncoder(); y = le.fit_transform(y)
OneHotEncoder()	One-hot encode categories	OneHotEncoder(sparse=False).fit_transform(X)
SimpleImputer()	Fill missing values	SimpleImputer(strategy='mean').fit_transform(X)
PolynomialFeatures()	Create polynomial features	PolynomialFeatures(degree=2).fit_transform(X)
ColumnTransformer()	Apply transforms per column	ColumnTransformer([('sc',StandardScaler(),[0,1])])

Models — Regression, Classification & Clustering

Model	Description	Example
LinearRegression()	Ordinary least squares	LinearRegression().fit(X_tr, y_tr)
Ridge(alpha)	L2 regularized regression	Ridge(alpha=1.0).fit(X_tr, y_tr)
Lasso(alpha)	L1 regularized regression	Lasso(alpha=0.1).fit(X_tr, y_tr)
LogisticRegression()	Binary/multi-class classifier	LogisticRegression(C=1.0, max_iter=1000)
SVC(kernel)	Support vector classifier	SVC(kernel='rbf', C=1.0, gamma='scale')
RandomForestClassifier()	Ensemble of decision trees	RandomForestClassifier(n_estimators=100)
GradientBoostingClassifier()	Gradient boosting trees	GradientBoostingClassifier(n_estimators=200)
XGBClassifier()	Extreme gradient boosting	XGBClassifier(n_estimators=200, learning_rate=0.1)
GaussianNB()	Naive Bayes classifier	GaussianNB().fit(X_tr, y_tr)
MLPClassifier()	Neural network classifier	MLPClassifier(hidden_layer_sizes=(100,50))
KMeans(k)	K-Means clustering	KMeans(n_clusters=5, random_state=42)
DBSCAN()	Density-based clustering	DBSCAN(eps=0.5, min_samples=5)
IsolationForest()	Anomaly detection	IsolationForest(contamination=0.05)

Dimensionality Reduction & Feature Selection

Function	Description	Example
PCA(n)	Principal Component Analysis	pca = PCA(n_components=2); X_red = pca.fit_transform(X)
TSNE(n)	t-SNE for visualization	TSNE(n_components=2, perplexity=30).fit_transform(X)
SelectKBest()	Select k best features	SelectKBest(f_classif, k=10).fit_transform(X, y)
RFE()	Recursive feature elimination	RFE(estimator=clf, n_features_to_select=5)
VarianceThreshold()	Remove low-variance features	VarianceThreshold(threshold=0.01)

Evaluation, Tuning & Pipelines

Function	Description	Example
accuracy_score()	Classification accuracy	accuracy_score(y_te, y_pred)
precision_score()	Precision	precision_score(y_te, y_pred, average='macro')
recall_score()	Recall / sensitivity	recall_score(y_te, y_pred)
f1_score()	F1 harmonic mean of P & R	f1_score(y_te, y_pred, average='weighted')
roc_auc_score()	AUC of ROC curve	roc_auc_score(y_te, y_proba)
confusion_matrix()	Confusion matrix	confusion_matrix(y_te, y_pred)
classification_report()	Full classification metrics	print(classification_report(y_te, y_pred))
mean_squared_error()	MSE / RMSE for regression	mean_squared_error(y_te, y_pred, squared=False)
r2_score()	R-squared coefficient	r2_score(y_te, y_pred)
cross_val_score()	K-fold cross-validation	cross_val_score(clf, X, y, cv=5, scoring='f1')
GridSearchCV()	Exhaustive hyperparameter search	GridSearchCV(clf, param_grid, cv=5, scoring='roc_auc')
RandomizedSearchCV()	Random hyperparameter search	RandomizedSearchCV(clf, params, n_iter=50, cv=5)
Pipeline(steps)	Chain transforms + estimator	Pipeline([('sc', StandardScaler()), ('clf', SVC())])

PyTorch

Tensors, neural network layers, optimizers, training loop, data loading

Tensor Creation & Manipulation

Function	Description	Example
torch.tensor(data)	Create tensor from data	x = torch.tensor([[1.,2.],[3.,4.]])
torch.zeros(shape)	Zeros tensor	torch.zeros(3, 4, dtype=torch.float32)
torch.ones(shape)	Ones tensor	torch.ones(3, 4)
torch.rand(shape)	Uniform random [0,1)	torch.rand(100, 10)
torch.randn(shape)	Standard normal random	torch.randn(batch_size, 128)
torch.arange()	Range tensor	torch.arange(0, 10, step=2)
torch.from_numpy(arr)	Tensor from NumPy array	x = torch.from_numpy(np_array)
tensor.numpy()	Convert tensor to NumPy	arr = x.cpu().numpy()
tensor.item()	Python scalar from 1-elem tensor	loss_val = loss.item()
tensor.to(device)	Move to device	x = x.to('cuda')
tensor.view(shape)	Reshape (contiguous)	x.view(batch_size, -1)
tensor.reshape(shape)	Reshape (safe)	x.reshape(-1, 28*28)
tensor.permute(dims)	Reorder dimensions	x.permute(0, 2, 3, 1)
tensor.squeeze(dim)	Remove size-1 dimension	x.squeeze(0)
tensor.unsqueeze(dim)	Add size-1 dimension	x.unsqueeze(1)
torch.cat(tensors, dim)	Concatenate along axis	torch.cat([a, b], dim=0)
torch.stack(tensors)	Stack along new axis	torch.stack([a, b], dim=1)
torch.matmul(a, b)	Matrix multiplication	out = torch.matmul(W, x) + b
tensor.detach()	Detach from computation graph	y = x.detach()
torch.no_grad()	Context: disable gradients	with torch.no_grad(): out = model(x)

Neural Network Layers & Losses

Module	Description	Example
nn.Module	Base class for all models	class Net(nn.Module): def __init__(self): super().__init__()
nn.Linear(in, out)	Fully-connected layer	nn.Linear(784, 256)
nn.Conv2d(in,out,k)	2D convolution	nn.Conv2d(3, 64, kernel_size=3, padding=1)
nn.MaxPool2d(k)	Max pooling	nn.MaxPool2d(kernel_size=2, stride=2)
nn.BatchNorm2d(c)	Batch normalization (2D)	nn.BatchNorm2d(64)
nn.Dropout(p)	Dropout regularization	nn.Dropout(p=0.5)
nn.Embedding(n,d)	Embedding lookup table	nn.Embedding(vocab_size, embed_dim)
nn.LSTM(in,h)	Long Short-Term Memory	nn.LSTM(input_size=128, hidden_size=256, num_layers=2)
nn.MultiheadAttention()	Multi-head self-attention	nn.MultiheadAttention(embed_dim=512, num_heads=8)
nn.TransformerEncoder()	Transformer encoder stack	nn.TransformerEncoder(encoder_layer, num_layers=6)
nn.Sequential(*layers)	Chain layers sequentially	nn.Sequential(nn.Linear(784,256), nn.ReLU(), nn.Linear(256,10))
nn.CrossEntropyLoss()	Classification loss	criterion = nn.CrossEntropyLoss()
nn.BCEWithLogitsLoss()	Binary classification loss	nn.BCEWithLogitsLoss()
nn.MSELoss()	Mean squared error loss	criterion = nn.MSELoss()

Optimizers & Training Loop

Function	Description	Example
optim.Adam()	Adam optimizer	optim.Adam(model.parameters(), lr=1e-3, weight_decay=1e-4)
optim.AdamW()	Adam with weight decay fix	optim.AdamW(model.parameters(), lr=2e-4)
optim.SGD()	Stochastic gradient descent	optim.SGD(model.parameters(), lr=0.01, momentum=0.9)
optimizer.zero_grad()	Clear accumulated gradients	optimizer.zero_grad()
loss.backward()	Compute gradients	loss.backward()
optimizer.step()	Update model parameters	optimizer.step()
nn.utils.clip_grad_norm_()	Clip gradient norms	nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)
CosineAnnealingLR()	Cosine LR annealing	CosineAnnealingLR(optimizer, T_max=100)
model.train()	Set to training mode	model.train() # enables dropout, batchnorm
model.eval()	Set to eval mode	model.eval() # disables dropout, batchnorm
torch.save(state_dict)	Save model weights	torch.save(model.state_dict(), 'model.pt')
torch.load()	Load model weights	model.load_state_dict(torch.load('model.pt'))

Data Loading & Transforms

Function	Description	Example
Dataset.__getitem__()	Custom dataset indexing	def __getitem__(self, idx): return X[idx], y[idx]
Dataset.__len__()	Dataset length	def __len__(self): return len(self.data)
DataLoader(ds, batch)	Batch + shuffle dataset	DataLoader(ds, batch_size=32, shuffle=True, num_workers=4)
transforms.Compose()	Chain image transforms	transforms.Compose([transforms.ToTensor(), transforms.Normalize()])
transforms.Normalize()	Normalize image tensor	transforms.Normalize(mean=[0.5], std=[0.5])
torchvision.datasets.*	Built-in datasets	datasets.MNIST('./data', train=True, download=True)
torchvision.models.*	Pretrained models	models.resnet50(pretrained=True)

LangGraph

State graphs, memory, HITL, tools, messages and subgraphs

Core State & Graph Primitives

Function	Description	Example
StateGraph(schema)	Create graph with typed state	graph = StateGraph(MyState)
MessageGraph()	Graph using message list state	graph = MessageGraph()
TypedDict	Define state schema	class MyState(TypedDict): messages: list; count: int
Annotated[T, reducer]	Field with custom reducer	messages: Annotated[list, operator.add]
END	Terminal node constant	graph.add_edge('node', END)
START	Entry node constant	graph.add_edge(START, 'first_node')

Graph Building & Execution

Function	Description	Example
graph.add_node(name, fn)	Register a node function	graph.add_node('agent', call_llm)
graph.add_edge(a, b)	Static directional edge	graph.add_edge('tool', 'agent')
graph.add_conditional_edges()	Route based on output	graph.add_conditional_edges('agent', route_fn, {'tool':'tool','end':END})
graph.compile()	Compile to runnable	app = graph.compile()
graph.compile(checkpointer)	Compile with memory	app = graph.compile(checkpointer=MemorySaver())
app.invoke(input)	Run graph synchronously	result = app.invoke({'messages': [HumanMessage('Hi')]})
app.stream(input)	Stream graph step by step	for chunk in app.stream(input, config): print(chunk)
app.ainvoke(input)	Async graph invocation	result = await app.ainvoke(input, config)

Memory & Persistence

Function	Description	Example
MemorySaver()	In-memory thread checkpointer	checkpointer = MemorySaver()
SqliteSaver.from_conn()	SQLite persistent memory	SqliteSaver.from_conn_string('memory.db')
thread_id in config	Isolate conversation sessions	config = {'configurable': {'thread_id': 'user-123'}}
app.get_state(config)	Read current graph state	state = app.get_state(config)
app.update_state(cfg, val)	Manually patch state	app.update_state(config, {'count': 5})
app.get_state_history(cfg)	Full state history	for s in app.get_state_history(config): print(s)

Human-in-the-Loop (HITL)

Function	Description	Example
interrupt_before=['node']	Pause before node runs	app = graph.compile(interrupt_before=['human_review'])
interrupt_after=['node']	Pause after node runs	app = graph.compile(interrupt_after=['agent'])
app.update_state(cfg, v)	Inject human feedback	app.update_state(config, {'approved': True})
NodeInterrupt(msg)	Raise interrupt from node	raise NodeInterrupt('Need human approval')
app.invoke(None, config)	Resume after interrupt	app.invoke(None, config)

Tools & Messages

Function	Description	Example
ToolNode(tools)	Built-in node to execute tools	tool_node = ToolNode([search, calculator])
tools_condition(state)	Route to tool or END	graph.add_conditional_edges('agent', tools_condition)
@tool decorator	Define a LangChain tool	@tool def search(q: str) -> str: return web_search(q)
bind_tools(tools)	Bind tools to LLM	llm_with_tools = llm.bind_tools([search, calculator])
HumanMessage(content)	User message	HumanMessage(content='What is the weather?')
AIMessage(content)	LLM response message	AIMessage(content='The weather is...')
SystemMessage(content)	System prompt message	SystemMessage(content='You are a helpful assistant')
ToolMessage	Result message from a tool	ToolMessage(content=result, tool_call_id=call.id)

Subgraphs, Fan-out & Advanced Routing

Function	Description	Example
Send(node, state)	Fan-out: send to multiple	return [Send('process', {'item': x}) for x in items]
Command(goto, update)	Dynamic routing + update	return Command(goto='agent', update={'count': n+1})
InjectedState	Pass full state to tool	def tool(query: str, state: Annotated[State, InjectedState])
RunnableConfig	Pass config to nodes	def node(state, config: RunnableConfig): ...