singvals,
len(trainDF),
probObservation=1.0,
modelType="svd",
svdMethod="numpy",
otherSeriesKeysArray=otherStates,
)
# fit the model
rscModel.fit(trainDF)
# save the denoised training data
denoisedDF = rscModel.model.denoisedDF()
# predict - all at once
predictions = rscModel.predict(testDF)
# plot
yearsToPlot = range(yearStart, yearTestEnd, 1)
interventionYear = yearTrainEnd - 1
plt.plot(
yearsToPlot,
np.append(trainMasterDF[basqueKey], testDF[basqueKey], axis=0),
color="red",
label="observations",
)
plt.plot(
yearsToPlot,
np.append(denoisedDF[basqueKey], predictions, axis=0),
color="blue",
label="predictions",
def basque(filename):
# BASQUE COUNTRY STUDY
df = pd.read_csv(filename)
pivot = df.pivot_table(values='gdpcap', index='regionname', columns='year')
pivot = pivot.drop('Spain (Espana)')
dfBasque = pd.DataFrame(pivot.to_records())
allColumns = dfBasque.columns.values
states = list(np.unique(dfBasque['regionname']))
years = np.delete(allColumns, [0])
basqueKey = 'Basque Country (Pais Vasco)'
states.remove(basqueKey)
otherStates = states
yearStart = 1955
yearTrainEnd = 1971
yearTestEnd = 1998
singvals = 1
p = 0.8
trainingYears = []
for i in range(yearStart, yearTrainEnd, 1):
trainingYears.append(str(i))
testYears = []
for i in range(yearTrainEnd, yearTestEnd, 1):
testYears.append(str(i))
trainDataMasterDict = {}
trainDataDict = {}
testDataDict = {}
for key in otherStates:
series = dfBasque[dfBasque['regionname'] == key]
trainDataMasterDict.update({key: series[trainingYears].values[0]})
# randomly hide training data
(trainData, pObservation) = tsUtils.randomlyHideValues(copy.deepcopy(trainDataMasterDict[key]), p)
trainDataDict.update({key: trainData})
testDataDict.update({key: series[testYears].values[0]})
series = dfBasque[dfBasque['regionname'] == basqueKey]
trainDataMasterDict.update({basqueKey: series[trainingYears].values[0]})
trainDataDict.update({basqueKey: series[trainingYears].values[0]})
testDataDict.update({basqueKey: series[testYears].values[0]})
trainMasterDF = pd.DataFrame(data=trainDataMasterDict)
trainDF = pd.DataFrame(data=trainDataDict)
testDF = pd.DataFrame(data=testDataDict)
# model
rscModel = RobustSyntheticControl(basqueKey, singvals, len(trainDF), probObservation=1.0, modelType='als',
otherSeriesKeysArray=otherStates)
# fit the model
rscModel.fit(trainDF)
# save the denoised training data
denoisedDF = rscModel.model.denoisedDF()
# predict - all at once
predictions = rscModel.predict(testDF)
# plot
yearsToPlot = range(yearStart, yearTestEnd, 1)
interventionYear = yearTrainEnd - 1
plt.plot(yearsToPlot, np.append(trainMasterDF[basqueKey], testDF[basqueKey], axis=0), color='red',
label='observations')
plt.plot(yearsToPlot, np.append(denoisedDF[basqueKey], predictions, axis=0), color='blue', label='predictions')
plt.axvline(x=interventionYear, linewidth=1, color='black', label='Intervention')
# plt.ylim((-1, 0))
legend = plt.legend(loc='upper right', shadow=True)
plt.title('Abadie et al. Basque Country Case Study - $p = %.2f$' % p)
plt.show()
def runAnalysis(N, T, TrainingEnd, rowRank, colRank):
# generate metric matrices
genFunctionOne = simpleFunctionOne
genFunctionTwo = simpleFunctionTwo
trueWeights = np.random.uniform(0.0, 1.0, N)
trueWeights = trueWeights / np.sum(trueWeights)
thetaArrayParams = np.random.uniform(0.0, 1.0, rowRank)
rhoArrayParams = np.random.uniform(0.0, 1.0, colRank)
rowParams = np.random.choice(thetaArrayParams, N)
colParams = np.random.choice(rhoArrayParams, T)
# metric 1
(observationMatrix1, meanMatrix1, trainDF1, testDF1, meanTrainingDict1,
meanTestDict1) = generateOneMetricMatrix(N, T, TrainingEnd, rowRank,
colRank, genFunctionOne,
trueWeights, rowParams,
colParams)
# metric 2
(observationMatrix2, meanMatrix2, trainDF2, testDF2, meanTrainingDict2,
meanTestDict2) = generateOneMetricMatrix(N, T, TrainingEnd, rowRank,
colRank, genFunctionTwo,
trueWeights, rowParams,
colParams)
keySeriesLabel = '0'
otherSeriesLabels = []
for ind in range(1, N + 1):
otherSeriesLabels.append(str(ind))
# RSC analysis
singvals = 8
############################
#### RSC for metric 1
rscmodel1 = RobustSyntheticControl(keySeriesLabel,
singvals,
len(trainDF1),
probObservation=1.0,
svdMethod='numpy',
otherSeriesKeysArray=otherSeriesLabels)
# fit the model
rscmodel1.fit(trainDF1)
predictionsRSC1 = rscmodel1.predict(testDF1)
rscRMSE1 = np.sqrt(
np.mean((predictionsRSC1 - meanTestDict1[keySeriesLabel])**2))
#print("\n\n *** RSC rmse1:")
#print(rscRMSE1)
############################
##### RSC for metric 2
rscmodel2 = RobustSyntheticControl(keySeriesLabel,
singvals,
len(trainDF2),
probObservation=1.0,
svdMethod='numpy',
otherSeriesKeysArray=otherSeriesLabels)
# fit the model
rscmodel2.fit(trainDF2)
predictionsRSC2 = rscmodel2.predict(testDF2)
rscRMSE2 = np.sqrt(
np.mean((predictionsRSC2 - meanTestDict2[keySeriesLabel])**2))
#print("\n\n *** RSC rmse2:")
#print(rscRMSE2)
############################
#### multi RSC model (combined) --
relative_weights = [1.0, 1.0]
# instantiate the model
mrscmodel = MultiRobustSyntheticControl(
2,
relative_weights,
keySeriesLabel,
singvals,
len(trainDF1),
probObservation=1.0,
svdMethod='numpy',
otherSeriesKeysArray=otherSeriesLabels)
# fit
mrscmodel.fit([trainDF1, trainDF2])
# predict
combinedPredictionsArray = mrscmodel.predict(
[testDF1[otherSeriesLabels], testDF2[otherSeriesLabels]])
# split the predictions for the metrics
predictionsmRSC_1 = combinedPredictionsArray[0]
predictionsmRSC_2 = combinedPredictionsArray[1]
# compute RMSE
mrscRMSE1 = np.sqrt(
np.mean((predictionsmRSC_1 - meanTestDict1[keySeriesLabel])**2))
mrscRMSE2 = np.sqrt(
np.mean((predictionsmRSC_2 - meanTestDict2[keySeriesLabel])**2))
#print("\n\n *** mRSC rmse1:")
#print(mrscRMSE1)
#print("\n\n *** mRSC rmse2:")
#print(mrscRMSE1)
return ({
"rsc1": rscRMSE1,
"rsc2": rscRMSE2,
"mrsc1": mrscRMSE1,
"mrsc2": mrscRMSE2
})
def prop99(filename):
# CALIFORNIA PROP 99 STUDY
df = pd.read_csv(filename)
df = df[df['SubMeasureDesc'] == 'Cigarette Consumption (Pack Sales Per Capita)']
pivot = df.pivot_table(values='Data_Value', index='LocationDesc', columns=['Year'])
dfProp99 = pd.DataFrame(pivot.to_records())
allColumns = dfProp99.columns.values
states = list(np.unique(dfProp99['LocationDesc']))
years = np.delete(allColumns, [0])
caStateKey = 'California'
states.remove(caStateKey)
otherStates = states
yearStart = 1970
yearTrainEnd = 1989
yearTestEnd = 2015
singvals = 2
p = 1.0
trainingYears = []
for i in range(yearStart, yearTrainEnd, 1):
trainingYears.append(str(i))
testYears = []
for i in range(yearTrainEnd, yearTestEnd, 1):
testYears.append(str(i))
trainDataMasterDict = {}
trainDataDict = {}
testDataDict = {}
for key in otherStates:
series = dfProp99[dfProp99['LocationDesc'] == key]
trainDataMasterDict.update({key: series[trainingYears].values[0]})
# randomly hide training data
(trainData, pObservation) = tsUtils.randomlyHideValues(copy.deepcopy(trainDataMasterDict[key]), p)
trainDataDict.update({key: trainData})
testDataDict.update({key: series[testYears].values[0]})
series = dfProp99[dfProp99['LocationDesc'] == caStateKey]
trainDataMasterDict.update({caStateKey: series[trainingYears].values[0]})
trainDataDict.update({caStateKey: series[trainingYears].values[0]})
testDataDict.update({caStateKey: series[testYears].values[0]})
trainMasterDF = pd.DataFrame(data=trainDataMasterDict)
trainDF = pd.DataFrame(data=trainDataDict)
testDF = pd.DataFrame(data=testDataDict)
# model
rscModel = RobustSyntheticControl(caStateKey, singvals, len(trainDF), probObservation=1.0, modelType='als',
otherSeriesKeysArray=otherStates)
# fit the model
rscModel.fit(trainDF)
# save the denoised training data
denoisedDF = rscModel.model.denoisedDF()
# predict - all at once
predictions = rscModel.predict(testDF)
# plot
yearsToPlot = range(yearStart, yearTestEnd, 1)
interventionYear = yearTrainEnd - 1
plt.plot(yearsToPlot, np.append(trainMasterDF[caStateKey], testDF[caStateKey], axis=0), color='red',
label='observations')
plt.plot(yearsToPlot, np.append(denoisedDF[caStateKey], predictions, axis=0), color='blue', label='predictions')
plt.axvline(x=interventionYear, linewidth=1, color='black', label='Intervention')
legend = plt.legend(loc='lower left', shadow=True)
plt.title('Abadie et al. Prop 99 Case Study (CA) - $p = %.2f$' % p)
plt.show()