Thanks!

[coderschoolreview]

Awesome good work :D

[coderschoolreview]

could be written in shorter form : sum(ecom['Job']=='Lawyer')

[coderschoolreview]

overall, all answer are correct. great work :D

[coderschoolreview]

---

Assignment 4:

Visualization:

Should've plot in different plot. Sample code:

# Distribution of length between 3 species: 
## list of species
ls = list(data.Species.unique())

plt.rcParams['figure.figsize'] = (20,5)
f,axes = plt.subplots(1,4)

i=0;
for v in ('SepalLengthCm','SepalWidthCm','PetalLengthCm','PetalWidthCm'):
    plt.sca(axes[i])
    axes[i].set_title(v)

    for s in ls:
        sns.distplot(data[data.Species==s][v],
                     hist=False,
                     label=s
                    )
    i+=1

plt.legend()
plt.show()

[coderschoolreview]

[coderschoolreview]

Model:

Linear Regression only work for binary classification (you only have 2 classes) In this case you have 3.

Ref: https://www.quora.com/Can-you-do-multiclass-classification-with-logistic-regression

[coderschoolreview]

hyl0428: Assignment 5

The goal of this assignment was to introduce you following concepts in Machine Learning:

• Data quality check: missing value, anomaly, ...
• Exploratory Data Analysis: ratio of class, relationship of feature with another and relationship of feature with target, and data visualization.
• Preprocessing data: handling imbalance dataset, dealing with categorical variable.
• Modeling & evaluation: building classifier and evaluate, Hyper Parameter Tuning.

Things you did well

• Aweomse, you knew magic function !pip install missingno, more on magic command here
• Checking data quality: check for missing value and using missingno to visualize
• EDA: Ratio of classes, finding out that data is imbalance, visualizing relationship between features
• Preprocessing: under sampling / over sampling, train test split
• Modeling & evaluate: train and evaluate multiple model by comparing classification report of models against each other

Things to work on

• Preprocessing categorical features. Try pandas.get_dummies
• GridSearchCV and RandomSearchCV: common practice is using RandomSearchCV to narrow down search range then fine tuning with GridSearchCV, can read this article

Minor tips

• To know "which is the best combination of parameter":

  https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html

  # Using gridsearchcv, random forest model and this param grid to find the best combination of parameters 
  # Hint: example
  # https://stackoverflow.com/questions/30102973/how-to-get-best-estimator-on-gridsearchcv-random-forest-classifier-scikit
  
  param_grid = {
    'bootstrap': [True],
    'max_depth': [80, 90, 100, 110],
    'max_features': [2, 3],
    'min_samples_leaf': [3, 4, 5],
    'min_samples_split': [8, 10, 12],
    'n_estimators': [100, 200, 300, 1000]
  }
  
  from sklearn.model_selection import GridSearchCV
  gcv = GridSearchCV(RandomForestClassifier(),param_grid=param_grid)
  gcv.fit(X,y)
  gcv.best_params_

• Installing packages missingno (or any arbitrary package) on Win10:

  • Go to Anaconda Prompt
  • Type conda install -c conda-forge missingno

• Filtering series:

  • Your code:train_copy.isnull()

  • Could write like this to get column with null value only:

  ncols = train_copy.isnull().sum
  ncols[ncols!=0]

• Check if whole data frame have any null value:

  ​ train.isnull().any().any()

• For evaluation function, should print instead of return so when you loop through list of model and evaluate them, the result for each iteration printed to output.

[coderschoolreview]

---

Assignment 6

Goal of this Assignment

The goal of this assignment was to introduce you to following concepts:

• Unsupervised Learning (KMeans, Hierachical Clustering)
• PCA

You learn how to use PCA for dimension reduction, KMeans, and Hierarchical Clustering. Also you learn to visualize the result of both tenichque.

Things you did well:

• Ultilizing Pandas to read, summary and visualize data.
• Almost got everything right

To sum up:

• Great work! This showing you understood the concept and able to apply it. Keep it up!

Few minor tips

To create a range of integer, use range(low,high,step)

k_values = range(1,20)

To access first columns of pca_data, use pca_data.iloc[:,0] or pca_data['PC1']. pca_data[:,0] is wrong way of slicing a dataframe, hence the error.