[SOLVED] T81-559- Assignment 2: Applications of Deep Neural Networks

Description

Listing 1 shows a sample submission skeleton that you can use as a starting point for this assignment.

Listing 1: Sample Submission Skeleton

1

2

3 path = “./data/”

4

5 # Helpful functions

6

• # Encode text values to dummy variables(i.e. [1,0,0],[0,1,0],[0,0,1] for red,green,blue)
• def encode_text_dummy(df, name): 9 dummies = pd.get_dummies(df[name]) 10              for x in columns:
• dummy_name = “{}-{}”.format(name, x)
• df[dummy_name] = dummies[x]
• drop(name, axis=1, inplace=True)

14

15

• # Encode text values to a single dummy variable. The new columns (which do not replace the old) will have a 1
• # at every location where the original column (name) matches each of the One column is added for 18 # each target value.
• def encode_text_single_dummy(df, name, target_values):
• for tv in target_values:
• l = list(df[name].astype(str))
• l = [1 if str(x) == str(tv) else 0 for x in l]
• name2 = “{}-{}”.format(name, tv)
• df[name2] = l

25

26

• # Encode text values to indexes(i.e. [1],[2],[3] for red,green,blue).
• def encode_text_index(df, name):
• le = preprocessing.LabelEncoder()
• df[name] = le.fit_transform(df[name])
• return classes_

32

33

• # Encode a numeric column as zscores
• def encode_numeric_zscore(df, name, mean=None, sd=None):
• if mean is None: 37 mean = df[name].mean()

38

• if sd is None:
• sd = df[name].std()

41

42                  df[name] = (df[name] – mean) / sd

43

44

45 # Convert all missing values in the specified column to the median 46 def missing_median(df, name): 47    med = df[name].median()

48                  df[name] = df[name].fillna(med)

49

50

51 # Convert all missing values in the specified column to the default 52 def missing_default(df, name, default_value): 53         df[name] = df[name].fillna(default_value)

54

55

56 # Convert a Pandas dataframe to the x,y inputs that TensorFlow needs 57 def to_xy(df, target):

58         result = [] 59           for x in df.columns:

• if x != target:
• append(x)
• # find out the type of the target column. Is it really this hard? 🙁
• target_type = df[target].dtypes
• target_type = target_type[0] if hasattr(target_type, ’__iter__’) else target_type
• # Encode to int for classification, float otherwise. TensorFlow likes 32 bits.
• if target_type in (np.int64, np.int32):
• # Classification
• dummies = pd.get_dummies(df[target])
• return as_matrix(result).astype(np.float32), dummies.as_matrix ().astype(np.float32)
• else:
• # Regression
• return as_matrix(result).astype(np.float32), df.as_matrix([ target]).astype(np.float32) 73

74 # Nicely formatted time string 75 def hms_string(sec_elapsed):

• h = int(sec_elapsed / (60 * 60))
• m = int((sec_elapsed % (60 * 60)) / 60)
• s = sec_elapsed % 60
• return “{}:{:>02}:{:>05.2f}”.format(h, m, s)

80

81

• # Regression chart.
• def chart_regression(pred,y,sort=True):
• t = pd.DataFrame({’pred’ : pred, ’y’ : y.flatten()})
• if sort:
• sort_values(by=[’y’],inplace=True)
• a = plt.plot(t[’y’].tolist(),label=’expected’)
• b = plt.plot(t[’pred’].tolist(),label=’prediction’)
• ylabel(’output’)
• legend()
• show()

92

93 # Remove all rows where the specified column is +/- sd standard deviations 94 def remove_outliers(df, name, sd):

• drop_rows = df.index[(np.abs(df[name] – df[name].mean()) >= (sd * df[ name].std()))]
• drop(drop_rows, axis=0, inplace=True)

97

98

99 # Encode a column to a range between normalized_low and normalized_high. 100 def encode_numeric_range(df, name, normalized_low=-1, normalized_high=1,

• data_low=None, data_high=None):
• if data_low is None:
• data_low = min(df[name])
• data_high = max(df[name])

105

106         df[name] = ((df[name] – data_low) / (data_high – data_low)) \ 107    * (normalized_high – normalized_low) + normalized_low

108

109 # Solution

110

• def encode_toy_dataset(filename):
• df = pd.read_csv(filename, na_values=[’NA’, ’?’])
• encode_numeric_zscore(df, ’length’)
• encode_numeric_zscore(df, ’width’)
• encode_numeric_zscore(df, ’height’)
• encode_text_dummy(df, ’metal’)
• encode_text_dummy(df, ’shape’)
• return df

119

120 # Encode the toy dataset 121 def question1():

• print()
• print(“***Question 1***”)

124

125                path = “./data/”

126

• filename_read = os.path.join(path,”toy1.csv”)
• filename_write = os.path.join(path,”submit-jheaton-prog2q1.csv”)
• df = encode_toy_dataset(filename_read) # You just have to implement encode_toy_dataset above
• to_csv(filename_write,index=False)
• print(“Wrote {} lines.”.format(len(df)))

132

133

134 # Model the toy dataset, no cross validation 135 def question2():

• print()
• print(“***Question 2***”)

138

• def question3():
• print()
• print(“***Question 3***”)

142

• # Z-Score encode these using the mean/sd from the dataset (you got this in question 2)
• testDF = pd.DataFrame([
• {’length’:1, ’width’:2, ’height’: 3},
• {’length’:3, ’width’:2, ’height’: 5},
• {’length’:4, ’width’:1, ’height’: 3}
• ])

149

150

• def question4():
• print()
• print(“***Question 4***”)

154

155

• def question5():
• print()
• print(“***Question 5***”)

159

160

• question1()
• question2()
• question3()
• question4()
• question5()

Listing 2 shows what the output from this assignment would look like. Your numbers might di er from mine slightly. Every question, except 2, also generates an output CSV file. For your submission please include your Jupyter notebook and any generated CSV files that the questions specified. Name your output CSV files something such as submit-jheaton-prog2q1.csv. Submit a ZIP file that contains your Jupyter notebook and 4 CSV files to Blackboard. This will be 5 files total.

Listing 2: Expected Output

• ***Question 1***
• Wrote 10001 lines.

3

• ***Question 2***
• Epoch 00144: early stopping
• Final score (RMSE): 75.46247100830078

7

• ***Question 3***
• length: (5.5258474152584744, 2.8609014041584113)
• width: (5.5340465953404658, 2.8598366585224158)
• height: (5.5337466253374661, 2.8719829476156122)
• height length      width
• 0 -0.882205 -1.581907 -1.235659
• 1 -0.185856 -0.882861 -1.235659
• 2 -0.882205 -0.533338 -1.585338

16

• ***Question 4***
• Fold #1
• Epoch 00060: early stopping
• Fold score (RMSE): 0.21216803789138794
• Fold #2
• Epoch 00061: early stopping
• Fold score (RMSE): 0.14340682327747345
• Fold #3
• Epoch 00028: early stopping
• Fold score (RMSE): 0.3336745500564575
• Fold #4
• Epoch 00058: early stopping
• Fold score (RMSE): 0.2133668214082718
• Fold #5
• Epoch 00077: early stopping
• Fold score (RMSE): 0.1796143352985382
• Final, out of sample score (RMSE): 0.22570167481899261 34
• ***Question 5***
• Fold #1
• Epoch 00182: early stopping
• Fold score: 0.3625
• Fold #2
• Epoch 00425: early stopping
• Fold score: 0.9875
• Fold #3
• Epoch 00169: early stopping
• Fold score: 0.975
• Fold #4
• Epoch 00111: early stopping
• Fold score: 0.8987341772151899
• Fold #5
• Epoch 00203: early stopping
• Fold score: 0.8227848101265823
• Final, out of sample score: 0.8090452261306532

Question 1

Use the dataset found here for this question: [click for toy dataset].

Encode the toy1.csv dataset. Generate dummy variables for the shape and metal. Encode height, width and length as z-scores. Include, but do not encode the weight. If this encoding is performed in a function, named encode_toy_dataset, you will have an easier time reusing the code from question 1 in question 2.

Write the output to a CSV file that you will submit with this assignment. The CSV file will look similar to Listing 3.

Listing 3: Question 2 Output Sample

Question 2

Use the dataset found here for this question: [click for toy dataset].

Use the encoded dataset from question 1 and train a neural network to predict weight. Use 25% of the data as validation and 75% as training, make sure you shu            e the data. Report the RMSE error for the validation set. No CSV file need be generated for this question.

Question 3

Use the dataset found here for this question: [click for toy dataset].

Using the toy1.csv dataset calculate and report the mean and standard deviation for height, width and length. Calculate the z-scores for the dataframe given by Listing 4. Make sure that you use the mean and standard deviations you reported for this question. Write the results to a CSV file.

Listing 4: Question 3 Input Data

• testDF = pd.DataFrame([
• {’length’:1, ’width’:2, ’height’: 3},
• {’length’:3, ’width’:2, ’height’: 5},
• {’length’:4, ’width’:1, ’height’: 3}
• ])
• …

Your resulting CSV file should look almost exactly like Listing 5.

Listing 5: Question 3 Output Sample

• height,length,width
• -0.8822049883269626,-1.5819074849494659,-1.2356589865858818
• -0.18585564084337075,-0.8828608931337095,-1.2356589865858818
• -0.8822049883269626,-0.5333375972258314,-1.5853375067165896

Question 4

Use the dataset found here for this question: [click for iris dataset].

Usually the iris.csv dataset is used to classify the species. Not this time! Use the fields species, sepal-l, sepal-w, and petal-l to predict petal-w. Use a 5-fold cross validation and report ONLY out-of-sample predictions to a CSV file. Make sure to shu e the data. Your generated CSV file should look similar to Listing 6. Encode each of the inputs in a way that makes sense (e.g. dummies, z-scores).

Listing 6: Question 4 Output Sample

• sepal_l,sepal_w,petal_l,petal_w,species-Iris-setosa,species-Iris- versicolor,species-Iris-virginica,0,0
• 30995914214417364, -0.5903951331558184, 0.5336208818725668, 1.2,0.0,1.0,0.0,1.2,1.444551944732666
• -0.1730940663922016, 1.7038864723719687, -1.1658086782311483, –

0.3,1.0,0.0,0.0,0.3,0.\

• …

Question 5

Use the dataset found here for this question: [click for auto mpg dataset].

Usually the auto-mpg.csv dataset is used to regress the mpg. Not this time! Use the fields to predict how many cylinders the car has. Treat this as a classification problem, where there is a class for each number of cylinders. Use a 5-fold cross validation and report ONLY out-ofsample predictions to a CSV file. Make sure to shu e the data. Your generated CSV file should look similar to Listing 7. Encode each of the inputs in a way that makes sense (e.g. dummies, z-scores). Report the final out of sample accuracy score.

Listing 7: Question 4 Output Sample

• mpg,cylinders,displacement,horsepower,weight,acceleration,year,origin,name ,ideal,predict
• -0.7055506566787514, 8, 1.0892327311042995, 0.6722714619460141, 6300768256149949, -1.2938698102195594, 70, -0.7142457922976494, chevrolet chevelle malibu,8,8
• -1.0893794720944747, 8, 1.5016242793620063, 1.5879594901955474, –

0.8532590135498572, -1.4751810504376373, 70, -0.7142457922976494,buick skylark 320,8,8

• -0.7055506566787514, 8, 1.1947282434492943, 1.19552176380289, –

0.5497784722839334, -1.6564922906557151, 70, -0.7142457922976494, plymouth satellite,8,8

• …