def process(self, df):
df = self._processor(df)
if self.verbose:
printv("Values after passing through %s like: " %
self.__class__.__name__)
printv(df[:5])
return df
def _format_data(self, df):
""" Needs already cleaned up data """
printv("Splitting data into train and test")
# Split into training and test data.
tr, ts = Splitter(TS_SIZE).split(df)
# Split columns into input and output
tr, ts = self.model.io_split(tr), self.model.io_split(ts)
printv("Data split with element count: (train, test) = (%d, %d)"
% (len(tr.input), len(ts.input)))
return tr, ts
def preprocess_data(df):
print = printv
df = df.iloc[1:323]
for processor in FILTERS:
printv("Applying filter: %s" % processor.__class__.__name__)
df = processor.process(df)
# print(np.isnan(df.any()), np.isfinite(df.all()))
df.to_csv("temp.csv")
printv("Filters applied. Final columns:")
printv(df.columns)
printv("\nFinal cleaned up data looks like:\n", df[:3])
printv()
return df
def run(self, df):
# Make the neural net model
if not self.model_name:
self.model_name = DEFAULT_MODEL
self.model = make_model(self.model_name)
# Split data
tr, ts = self._format_data(df)
with SequentialNeuralNet(self.model, self.args.modelfile) as net:
net.train(tr)
predicted = pd.DataFrame(net.predict(ts.input),
columns=self.model.output_cols)
predicted = self.model.post_process(ts, predicted)
printv("Predicted data like:\n", predicted[:5])
if self.show_graph:
self.graph(ts, predicted)
def graph(self, ts_data, predicted):
""" Plot the predicted data vs. original data """
printv("Graphing predicted values")
fig, ax = plt.subplots(nrows=int(math.ceil(len(predicted.columns) / 2)),
ncols=2)
fig.suptitle("Actual vs. predicted values for neural net")
for i, ax in enumerate(ax.reshape(-1)):
if i == len(predicted.columns): break
col_name = predicted.columns[i]
self._graph_column(col_name, ax, ts_data.output[:, i],
predicted.loc[:, col_name])
plt.show()
# Prompt to save
save = self.args.graphfile if self.args.graphfile\
else input("Save Graph (Enter filename to save, blank to skip)?: ")
if save:
self._save_graph(fig, save)
def main(args):
df = pd.read_csv(DATA1)
# Basic cleanup
utils.printv("\n### Preprocessing data ###\n")
df = preprocess_data(df)
if args.basic_graph or TO_GRAPH:
utils.printv("\n### Graphing Data ###\n")
basic_graph(df)
if args.nnmodel != NOMODEL:
# Make an graph the neural net
utils.printv("\n### Running the network ###\n")
runner = NeuralNetRunner(args)
runner.run(df)
def plot(self, block=True):
utils.printv("Plotting graphs...")
plt.show(block)
