model: scratch: Create Linear Regression model about dffml HOT 5 CLOSED

@pdxjohnny I was implementing applicable features and found it leaded to the following functions. Do they need to be re-implemented or can I pick them up from dnn and change the conditions (for starters the feature length should be 2 and so on)

from dffml.

And can you suggest me how to debug my code like how can I go about testing (for now checking that whether I have received data successfully or not)?

from dffml.

Here's an outline for 2.a.iv and 2.a.v from https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.s3lkoesyhz9v

Is this what you're talking about?

class SimpleLinearRegression(Model):
    async def applicable_features(self, features):
        if len(features) != 1:
            raise ValueError("simple LR only supports a single feature")
        if features[0].dtype() != int and features[0].dtype() != float:
            raise ValueError("simple LR only supports int or float feature")
        if features[0].length() != 1:
            raise ValueError("simple LR only supports single values (non-matrix / array)")
        features_we_care_about = [features[0].NAME]
        return features_we_care_about

    async def train(self, sources, features):
        features_we_care_about = self.applicable_features(features)
        async for repo in sources.with_features(features_we_care_about):
            # Grab a subset of the feature data being stored within the repo
            # The subset is the feature_we_care_about and the feature we are want to predict
            feature_data = repo.features(features_we_care_about + [self.parent.config.predict])
            xData.append(feature_data[features_we_care_about[0]])
            yData.append(feature_data[self.parent.config.predict])

from dffml.

Here's an outline for 2.a.iv and 2.a.v from https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.s3lkoesyhz9v

Is this what you're talking about?

class SimpleLinearRegression(Model):
    async def applicable_features(self, features):
        if len(features) != 1:
            raise ValueError("simple LR only supports a single feature")
        if features[0].dtype() != int and features[0].dtype() != float:
            raise ValueError("simple LR only supports int or float feature")
        if features[0].length() != 1:
            raise ValueError("simple LR only supports single values (non-matrix / array)")
        features_we_care_about = [features[0].NAME]
        return features_we_care_about

    async def train(self, sources, features):
        features_we_care_about = self.applicable_features(features)
        async for repo in sources.with_features(features_we_care_about):
            # Grab a subset of the feature data being stored within the repo
            # The subset is the feature_we_care_about and the feature we are want to predict
            feature_data = repo.features(features_we_care_about + [self.parent.config.predict])
            xData.append(feature_data[features_we_care_about[0]])
            yData.append(feature_data[self.parent.config.predict])

well, this is pretty simple and understandable, I was using dnn as my reference wherein applicable features lead to self.features which in turn lead to feature_feature_column (which ultimately checked the types).

from dffml.

ya dnn is complicated by tensorflow APIs :)

from dffml.