Machine Learning Machine Learning Tools and Libraries Syllabus

Machine Learning Tools and Libraries syllabus — full chapter and topic list

Expand any chapter to see its topics and sub-topics. This is the whole examinable outline for Machine Learning Tools and Libraries in Machine Learning, not a summary of it.

1. 1

   Visualization Tools

   6 topics

   • Matplotlib
   • Seaborn
   • Plotly
   • Bokeh
   • Altair
   • Dash
2. 2

   Deep Learning Frameworks

   6 topics

   • TensorFlow
   • Keras
   • PyTorch
   • MXNet
   • Caffe
   • Theano
3. 3

   Natural Language Processing (NLP) Libraries

   6 topics

   • NLTK
   • SpaCy
   • Gensim
   • Transformers (Hugging Face)
   • TextBlob
   • Flair
4. 4

   Reinforcement Learning Libraries

   5 topics

   • OpenAI Gym
   • Stable Baselines
   • Ray RLlib
   • Keras-RL
   • TensorForce
5. 5

   Data Storage and Management

   6 topics

   • SQL Databases
   • NoSQL Databases
   • HDFS
   • Apache Spark
   • Dask
   • BigQuery
6. 6

   Deployment and Production

   9 topics

   • Flask
   • Django
   • FastAPI
   • TensorFlow Serving
   • ONNX
   • Docker
   • Kubernetes
   • MLflow
   • Kubeflow
7. 7

   Automated Machine Learning (AutoML)

   5 topics

   • Auto-sklearn
   • TPOT
   • H2O.ai
   • Google Cloud AutoML
   • Microsoft Azure AutoML

Machine Learning Tools and Libraries flashcards for Machine Learning

21 of 50 cards from the Machine Learning Tools and Libraries deck — real questions with worked answers.

1. What is Matplotlib and what is its core object hierarchy?

   Matplotlib is a low-level Python 2D plotting library. Its hierarchy is: a Figure (the whole canvas) contains one or more Axes (an individual plot), and each Axes contains Artists such as lines, text, and ticks. The state-based pyplot interface (plt) wraps the object-oriented Figure/Axes API.

2. In Matplotlib, what is the difference between the pyplot (state-machine) interface and the object-oriented interface?

   The pyplot interface (e.g. plt.plot, plt.title) implicitly tracks the 'current' figure and axes, convenient for quick scripts. The object-oriented interface explicitly creates and manipulates objects, e.g. fig, ax = plt.subplots(); ax.plot(...); ax.set_title(...). The OO style is preferred for complex, multi-axes figures.

3. In Matplotlib, what call creates a figure with a grid of subplots and returns both the figure and an array of axes?

   fig, axs = plt.subplots(nrows, ncols). It returns the Figure and a NumPy array of Axes objects, which you index (e.g. axs[0, 1]) to draw on each subplot.

4. What is Seaborn and how does it relate to Matplotlib?

   Seaborn is a high-level statistical data visualization library built on top of Matplotlib. It provides attractive default themes and concise functions for statistical plots (distributions, regressions, categorical comparisons) and integrates tightly with pandas DataFrames. Because it sits on Matplotlib, you can further customize Seaborn plots with Matplotlib calls.

5. In Seaborn, what is the distinction between figure-level and axes-level functions? Give an example of each.

   Axes-level functions draw onto a single Matplotlib Axes and accept an ax= argument (e.g. scatterplot, histplot, boxplot). Figure-level functions manage their own figure and can create multi-panel grids via row/col faceting (e.g. relplot, displot, catplot). Figure-level functions return a FacetGrid-type object rather than an Axes.

6. Which Seaborn function draws a scatterplot matrix of all pairwise relationships in a dataset?

   sns.pairplot(data), which plots pairwise scatterplots for every pair of numeric variables off the diagonal and univariate distributions (histograms or KDEs) on the diagonal. The hue= argument colors points by a categorical variable.

7. What is Plotly and what key feature distinguishes it from Matplotlib?

   Plotly is a graphing library that produces interactive, web-based (HTML/JavaScript) visualizations rendered with D3.js/WebGL. Unlike Matplotlib's static images, Plotly charts support hover tooltips, zooming, panning, and selection out of the box, and can be embedded in web pages or Dash apps.

8. In Plotly, what is the difference between Plotly Express and Plotly Graph Objects?

   Plotly Express (plotly.express, px) is a high-level concise API where one function call (e.g. px.scatter) builds an entire figure from a DataFrame. Plotly Graph Objects (plotly.graph_objects, go) is the lower-level API where you assemble figures from explicit trace and layout objects for fine-grained control.

9. What is Bokeh and what is its primary target output?

   Bokeh is a Python interactive visualization library that targets modern web browsers, generating standalone HTML/JavaScript (BokehJS) output. It excels at building interactive dashboards and streaming/large-dataset plots, and can run a Bokeh server to link Python callbacks to browser widgets.

10. In Bokeh, what is the role of ColumnDataSource?

    ColumnDataSource is Bokeh's core data structure: a mapping of column names to data sequences shared across glyphs. Using a shared ColumnDataSource lets multiple plots/glyphs reference the same data and enables linked interactions such as linked brushing, selection, and hover tooltips.

11. What is Altair and what visualization paradigm is it based on?

    Altair is a declarative Python statistical visualization library based on the Vega and Vega-Lite grammar of graphics. You describe what to plot by mapping data columns to visual encoding channels (x, y, color, size) rather than imperative drawing commands, and Altair emits a Vega-Lite JSON spec rendered in the browser.

12. In Altair, what is the typical chaining pattern to build a chart, and what does each part specify?

    alt.Chart(data).mark_point().encode(x='a', y='b', color='c'). Chart(data) binds the dataset, mark_*() chooses the geometric mark (point, bar, line, area), and encode() maps data fields to visual channels. Transforms and interactions can be chained on as well.

13. What is Dash and what three technology layers does it combine?

    Dash is a Python framework by Plotly for building analytical web applications with no JavaScript required. It combines Flask (web server/back end), Plotly.js (interactive charts), and React.js (front-end components). Apps are defined by a layout of components plus reactive callbacks.

14. In Dash, how is interactivity defined and what decorator is used?

    Interactivity is defined by callback functions decorated with @app.callback (or @callback). The decorator declares Output(s) (component properties to update) and Input(s)/State(s) (component properties that trigger or supply values). When an Input changes, Dash calls the function and updates the Outputs reactively.

15. What is TensorFlow and who develops it?

    TensorFlow is an open-source end-to-end machine learning and deep learning framework developed by Google. It represents computations as dataflow graphs of tensors, supports automatic differentiation, runs on CPUs/GPUs/TPUs, and scales from research prototyping to large-scale production deployment (e.g. via TensorFlow Serving and TensorFlow Lite).

16. What is a 'tensor' in the context of TensorFlow/PyTorch, and what do rank and shape mean?

    A tensor is a multi-dimensional array of numerical values. Its rank (or ndim) is the number of axes/dimensions (rank-0 scalar, rank-1 vector, rank-2 matrix), and its shape is the tuple giving the size along each axis, e.g. a tensor of shape $(3, 4)$ has rank 2 with 3 rows and 4 columns.

17. What major architectural change did TensorFlow 2.x introduce regarding graph execution?

    TensorFlow 2.x made eager execution the default, so operations run immediately like normal Python (rather than building a static graph then running a session as in TF1). Graph optimization is opt-in via the @tf.function decorator, which traces a Python function into a callable TensorFlow graph for performance.

18. What is Keras and what is its relationship to TensorFlow?

    Keras is a high-level neural-network API focused on fast, user-friendly model building. It is the official high-level interface of TensorFlow (tf.keras) and provides layers, optimizers, and losses with a simple declarative style. Keras 3 is multi-backend, able to run on TensorFlow, JAX, or PyTorch.

19. What are the three model-building APIs offered by Keras?

    1) The Sequential API: a linear stack of layers, simplest for single-input/single-output models. 2) The Functional API: builds models as a directed graph of layers, supporting multiple inputs/outputs and shared layers. 3) Model subclassing: define a custom class extending keras.Model with your own call() method for full flexibility.

20. In Keras, what two steps must you perform before and during training a model, and which methods implement them?

    First compile the model with model.compile(optimizer=..., loss=..., metrics=...) to configure the training process, then train with model.fit(x, y, epochs=..., batch_size=...). Evaluation uses model.evaluate() and inference uses model.predict().

21. What is PyTorch and which feature is central to its 'define-by-run' design?

    PyTorch is an open-source deep learning framework (originating at Facebook/Meta AI) known for a Pythonic interface. Its define-by-run design uses dynamic computation graphs built on the fly during the forward pass, so the graph can change each iteration. Autograd provides automatic differentiation by recording operations on tensors.

See more Machine Learning Tools and Libraries flashcards →

Planning Machine Learning Tools and Libraries for Machine Learning

Machine Learning Tools and Libraries (Machine Learning) FAQ